A building micro-grid coordinated global AI adaptive lighting, air conditioning and fresh air integrated intelligent flexible energy load regulation system and AI intelligent regulation method

Abstract

The application discloses a building micro-grid collaborative global AI adaptive lighting, air conditioning and fresh air integrated intelligent flexible energy load regulation system and an AI intelligent regulation method. The system constructs a cloud digital twin platform, an edge wireless centralized control, a flexible load device, a global AI sensing, a micro-grid source storage five-layer architecture, integrates eight decoupling modules of a distributed micro-grid source load storage collaborative unit, a cloud-edge-end three-level global AI computing power, a 0-100% stepless integrated flexible load unit, a three-layer heterogeneous redundant communication, a digital twin simulation, a virtual power plant dispatching and a blockchain carbon storage, and contains seven closed-loop steps of data acquisition, hierarchical AI optimization, micro-grid collaborative scheduling, offline redundant execution, virtual power plant interaction, digital twin iteration and blockchain carbon accounting. The method realizes global AI predictive stepless flexible regulation, can bidirectionally participate in virtual power plant demand response, and has the advantages of digital twin self-evolution, blockchain credible carbon accounting and multiple communication / power grid fault offline operation.

Description

Technical Field

[0001] This invention relates to the cross-technical fields of smart buildings, low-carbon energy saving, flexible load regulation, and Internet of Things adaptive sensing control. It is particularly applicable to the coordinated and flexible energy management of multi-type buildings such as public buildings, commercial buildings, office complexes, industrial parks, hospitals, schools, and hotels, and belongs to the category of building energy digitalization and green low-carbon energy saving technology. Background Technology

[0002] Against the backdrop of global energy transition and the deep implementation of the dual-carbon strategy, the building sector is a core carrier of energy consumption and carbon emissions for the entire society. In my country, the proportion of building operation energy consumption in total social energy consumption remains consistently high. Traditional building electromechanical equipment generally suffers from pain points such as extensive operation, simplistic control logic, supply-demand imbalance, and serious energy waste. Therefore, there is an urgent need to provide a smart, flexible energy system with comprehensive perception, multi-system linkage, and millisecond-level adaptive flexible control to address the industry pain points of existing building electromechanical equipment, including independent management, extensive control, limited energy-saving effects, and poor living experience. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a building microgrid-based, AI-adaptive, integrated intelligent flexible energy load control system for lighting, air conditioning, and fresh air. The system constructs a standardized, layered architecture from top to bottom: a cloud-based digital twin scheduling layer, an edge-controlled computing layer, a flexible load equipment layer, a global sensing layer, and a microgrid source / storage equipment layer. These five layers are vertically independent and decoupled, yet horizontally interconnected, enabling full-domain linkage of computing power, load, sensing, source / storage, and scheduling. The system integrates eight mutually decoupled, independently selectable, and modularly pluggable core functional modules. It also adds an auxiliary flexible linkage interface for electric louvers and reserves flexible expansion interfaces for electric sunshades, smart curtains, and building water supply and drainage, forming a multi-source coupled intelligent energy control system encompassing photovoltaics, energy storage, sunshades, lighting, air conditioning, and fresh air. The eight modules include: The distributed microgrid source-load-storage collaborative unit is equipped with a photovoltaic grid-connected inverter, a multi-type energy storage BMS management host, a park AC / DC charging pile controller, a millisecond-level microgrid on-grid / off-grid switching switch, and a source-storage-load synchronous sampling module. The unit incorporates source-load-storage dynamic matching subroutines, power smoothing subroutines, and islanding fault-tolerant control subroutines, and is compatible with three mainstream energy storage devices: lithium iron phosphate, lead-carbon, and vanadium redox flow batteries, as well as energy storage systems of different voltage levels. It collects real-time data on photovoltaic instantaneous power generation, real-time energy storage SOC, individual cell voltage, charging / discharging current, battery temperature, and charging pile dynamics. The system autonomously manages the entire process, including load, equipment operating status, grid time-of-use peak and off-peak electricity prices, grid voltage, frequency, and quality signals, to maximize local photovoltaic power consumption, utilize energy storage for peak-valley charging and discharging arbitrage, manage and control isolated loads with graded fault tolerance, restore grid fault recovery without impact and smooth grid connection, and gradually backtrack power after grid connection. The entire microgrid scheduling logic, security constraint logic, and power optimization logic are all embedded in the edge local flash memory, requiring no continuous cloud-based instructions or cloud computing power support. It can achieve localized, fully functional, autonomous, and stable operation even under grid outages, cloud disconnection, and abnormal grid conditions. The cloud-edge-device full-domain AI adaptive computing architecture constructs a three-tiered distributed heterogeneous computing power system: lightweight inference at the terminal, real-time optimization at the edge, and iterative training in the cloud. The bottom layer consists of a lightweight AI inference unit built into the sensing terminal, responsible for pre-sensing and predicting the environment, personnel, and operating conditions. The middle layer is the NPU local inference engine mounted on the intelligent flexible controller, handling millisecond-level real-time optimal control solutions. The top layer is a cloud-based reinforcement learning iterative training platform, responsible for accumulating large-scale data samples, iterating models, optimizing parameters, and upgrading strategies. This three-tiered computing power system is uniformly connected to a standardized multi-source time-series database, integrating real-time microgrid output, energy storage operation status, grid time-of-use pricing, millimeter-wave human multimodal point cloud data, indoor temperature and humidity, carbon dioxide, PM2.5, TVOC, natural light intensity, outdoor weather time series, and building load. The system performs global collaborative optimization calculations using time-series multi-dimensional parameters, completely eliminating the problems of isolated computing power, data fragmentation, and isolated strategies in traditional building systems. The intelligent flexible central controller serves as the core computing power and control carrier at the system edge, locally embedding a complete set of microgrid collaborative algorithms, multi-load linkage adjustment algorithms, DRL deep reinforcement learning multi-objective optimization algorithms, and fault-tolerant redundancy algorithms. It is equipped with a large-capacity power outage protection flash memory and a power failure data retention mechanism, and has millisecond-level local offline fully autonomous control capabilities. When single or multiple overlapping link failures occur, such as external fiber optic cable disconnection, cloud server disconnection, public power grid power failure, or single communication link interruption, the system can independently execute all flexible control logic in a closed loop, continuously ensuring uninterrupted and uncontrolled adaptive and refined adjustment of lighting, air conditioning, and fresh air, as well as microgrid energy storage collaborative scheduling and load optimization allocation. The integrated flexible load control unit for lighting, air conditioning, and fresh air is equipped with a stepless dimming LED driver module, a variable frequency air conditioning continuous cooling and heating adjustment module, and a fresh air stepless speed regulation airflow adjustment module. The three types of equipment output a unified and standardized 0~100% continuous analog control signal with a minimum adjustment step size accurate to 1%, without fixed levels, graded jumps, or start / stop dead zones. The system unifies the modeling, aggregation, and management of all flexible adjustable loads of lighting, air conditioning, and fresh air throughout the building, integrating them into a standardized, quantifiable, dispatchable, and tradable flexible load resource pool. It outputs continuously smooth and adjustable power, and can accurately match the millisecond-level fine-grained dispatching commands of the power grid and the demand response commands of the virtual power plant, realizing the upgrade of building load from rigid fixed load to flexible adjustable resource. The all-domain multi-dimensional sensing terminal is uniformly equipped with a 60GHz frequency-modulated continuous wave high-precision millimeter-wave radar, a wide-range illumination acquisition probe, a high-precision temperature and humidity sensor, and a four-in-one air quality detection module for carbon dioxide / PM2.5 / TVOC / formaldehyde. All components utilize a low-power lithium battery power supply and a LoRa long-range wireless communication solution. Retrofitting existing buildings requires no trenching or wiring, no laying of high-voltage control lines, and no modification to existing equipment wiring, making it suitable for rapid deployment in older buildings without modification. The terminal incorporates a lightweight CNN convolutional neural network and a 72-hour sliding window LSTM time-series prediction submodule: the CNN network can accurately identify four core postures: standing while working, sitting while working, lying down while resting, and in unoccupied spaces. Output parameters include real-time number of people, personnel density, and duration of stay in each zone; the LSTM model predicts fluctuations in natural light intensity, trends in indoor pollutant concentration, and time-series fluctuations in photovoltaic power generation based on long-term historical data, enabling proactive prediction and control of environmental and operational changes, eliminating control lag issues; all AI perception, recognition, and prediction calculations at the terminal are completed locally, without needing to upload to the cloud computing power, and the function is completely unaffected by network conditions such as isolated offline or network outages; the cloud backend supports custom configuration and batch distribution of parameters such as unattended latency, sensitivity threshold, and comfort zone, which can accurately adapt to different personnel stay patterns and different energy consumption scenarios such as office buildings, hotels, campuses, shopping malls, hospitals, and industrial parks; A three-layer heterogeneous redundant communication network of 5G+WiFi+LoRa is deployed in parallel with industrial 4G or 5G modules, dual-band 2.4G / 5G WiFi modules, and LoR star-shaped networking modules. The three types of communication links have built-in real-time fault monitoring, automatic identification, and seamless switching mechanisms, with no delay and no packet loss during link switching. Among them, LoRa is responsible for low-power, long-distance, and interference-resistant data collection from massive sensor terminals at the bottom layer; WiFi is responsible for local high-speed data interaction and policy synchronization between floor central control and sub-controllers and terminal devices; 5G industrial modules carry long-distance high-speed and reliable transmission of real-time synchronous data of digital twin 3D, millisecond-level scheduling commands of virtual power plants, and power grid demand response interaction commands. The three-layer network has a clear division of labor, serves as a backup for each other, and is collaboratively redundant, comprehensively ensuring the real-time performance and reliability of data transmission at different levels. The digital twin virtual-real mapping unit integrates four sub-modules: offline full-condition simulation before commissioning, millisecond-level online virtual-real synchronization after commissioning, hierarchical fault early warning for all equipment and the entire link, and visualization analysis of operational data. Offline simulation can input multi-dimensional parameters such as building structure, window-to-wall ratio, heat transfer coefficient of the building envelope, photovoltaic and energy storage installed capacity, rated power of electromechanical equipment, annual personnel work and rest schedules, hourly weather, and hourly electricity prices. It automatically simulates the microgrid output, load conditions, and energy consumption characteristics for 8760 hours throughout the year, outputting optimal collaborative control thresholds and strategy parameters. Online synchronization... The step module synchronizes microgrid power generation, energy storage SOC, real-time power of zoned equipment, environmental parameters, and carbon emission reduction data to a 3D twin model every 100ms, visually and dynamically presenting the building's energy flow path, load peak shaving and valley filling amplitude, photovoltaic green electricity consumption ratio, and equipment operating status; the fault early warning module collects all types of fault signals in real time, including sensor offline, communication interruption, equipment abnormality, parameter exceeding limits, energy storage fault, and grid abnormality, and pushes alarms in three levels: general, important, and urgent, with precise location, fault cause analysis, standard handling plan, and operation and maintenance suggestions. The virtual power plant aggregation and dispatch unit reserves standardized API interfaces for electricity market access, grid demand response access, and regional energy internet access, enabling seamless integration with regional virtual power plant platforms, grid market-based dispatch platforms, and integrated energy service platforms. It integrates flexible loads such as building-wide lighting, air conditioning, and fresh air into a standardized distributed adjustable resource pool, uniformly quantifying adjustable capacity, adjustable rate, and continuously adjustable duration. It supports participation in various market-based electricity transactions, including grid peak-valley shaving, auxiliary frequency regulation, electricity spot trading, orderly electricity consumption, and demand response compensation, thereby realizing the resource-based, asset-based, and market-oriented management of building loads. This blockchain-based, full-lifecycle carbon data trusted storage unit features a built-in dedicated encrypted node on the consortium blockchain and an AI-powered intelligent accounting subroutine for the carbon footprint of buildings throughout their entire lifecycle. The system integrates regional power grid time-of-use carbon emission factors, photovoltaic zero-carbon offset coefficients, energy storage charging and discharging carbon conversion coefficients, and equipment loss carbon correction coefficients. It accurately calculates the real-time carbon emissions of lighting, air conditioning, and fresh air systems, overlaying the implicit carbon from the entire process of production, installation, transportation, and scrapping of various electromechanical equipment. This constructs a complete full-lifecycle carbon asset ledger covering equipment production, engineering construction, building operation, and equipment decommissioning. All original energy consumption, power generation, energy storage, load regulation, environmental monitoring, and carbon accounting data are asymmetrically encrypted to generate unique hash values ​​for on-chain hash storage. The data is tamper-proof, undeletable, fully traceable, and cross-verifiable. It can automatically generate standardized carbon verification certificates compatible with Green Building Three-Star, LEED, BREEAM, corporate ESG annual reports, CCER carbon sink development, green credit, and green bond applications. Furthermore, it automatically calculates the project's annual carbon emission reduction and potential carbon trading revenue through a built-in real-time carbon trading price dynamic model. The system's eight modules employ a fully decoupled, pluggable modular design for both hardware and software. The distributed microgrid source-load-storage coordination unit, digital twin virtual-real mapping unit, virtual power plant aggregation and scheduling unit, and blockchain-based full-lifecycle carbon data trusted storage unit can all be selected, removed, or expanded as needed to adapt to different renovation and new construction scenarios. For energy-saving renovations of existing buildings without photovoltaics or energy storage, energy saving can be achieved simply by deploying a full-domain multi-dimensional sensing terminal, a cloud-edge-device full-domain AI adaptive computing architecture, a flexible load control unit integrating lighting / air conditioning / fresh air, and a 5G+WiFi+LoRa three-layer heterogeneous redundant communication network. Effectiveness: For newly constructed zero-carbon and near-zero-energy buildings, a complete set of full-function modules can be installed to achieve full-dimensional capabilities such as source-load-storage synergy, AI self-adaptation, digital twin visualization, reliable carbon data management and control, and grid market participation; The system also reserves a dedicated control interface and data interaction interface for the linkage of electric shading louvers, allowing the electric louvers to be connected to the edge control as a passive energy-saving flexible resource, and constructing a six-element synergistic energy-saving system of photovoltaic-energy storage-electric shading-lighting-air conditioning-fresh air. By dynamically adjusting the louver angle in conjunction with light, temperature, and operating conditions, the solar radiation heat gain of the building is reduced from the source, significantly reducing the basic load of air conditioning in summer.

[0004] Furthermore, the distributed microgrid source-load-storage collaborative unit is equipped with a dedicated microgrid on-grid / off-grid switching controller, which has a built-in hardware-level millisecond-level islanding detection circuit and voltage and frequency distortion identification algorithm. The islanding voltage and frequency anomaly detection delay is ≤10ms, which can accurately identify real islands, instantaneous power grid fluctuations, and load change interference, and prevent misjudgment and incorrect switching. When the public power grid fails and the voltage and frequency exceed the limit to enter islanding mode, the cloud-edge-device full-domain AI computing power architecture automatically starts the preset hierarchical flexible load reduction strategy, which divides the building load into a three-level load system: Level 1 basic guarantee load, Level 2 office regular load, and Level 3 public area non-essential load. The Level 1 basic guarantee load continues to operate at full power without interruption, ensuring basic lighting, computer room constant temperature, basic fresh air, and security monitoring. The core load operates uninterruptedly; secondary and tertiary loads are not directly cut off or shut down instantaneously, but are smoothly and steplessly reduced in power and load by integrated flexible units according to priority, achieving shock-free load reduction; surplus photovoltaic power generation is preferentially stored in energy storage batteries for local consumption; when the energy storage SOC is lower than the system's preset safety threshold, the AI ​​engine automatically and adaptively tightens the indoor temperature, humidity, illuminance, and air quality comfort range, further reducing equipment operating power and strictly controlling energy consumption while ensuring basic living experience; after the grid voltage, frequency, and phase parameters are restored to grid connection standards, the system gradually and gradually increases the load power in multiple stages, simultaneously completing precise phase matching of the microgrid, with no voltage surge or current surge throughout the process, and no need for manual on-site operation, achieving fully automatic and imperceptible grid connection restoration; The integrated flexible load unit for lighting / air conditioning / fresh air adopts a smooth stepless continuous power regulation mode from 0 to 100% throughout the entire process. This completely avoids the instantaneous high current impact caused by the frequent switching of relays, repeated start-stop of compressors, and fan speed jumps in traditional equipment, as well as the harmonic disturbances of the power grid and wear and tear on the mechanical structure. It effectively extends the service life of core equipment such as LED lighting fixtures, air conditioning compressors, and fresh air fans, and significantly reduces the energy consumption and failure probability caused by the frequent start-stop of central air conditioning units. The system is equipped with an intelligent operation and maintenance early warning module in the cloud, supporting real-time alarm pushes on both mobile app and PC web management terminal. All fault information is accompanied by a precise building, floor, and room location QR code and a unique equipment code. Operation and maintenance personnel can scan the code to accurately locate the fault point, retrieve equipment files and historical operation and maintenance records, and completely reduce ineffective inspections and blind troubleshooting. The system supports customizable fault push blocking time periods and fault level push rules. Low-priority faults at night and on holidays can be pushed and handled centrally on weekdays, balancing the rest rights of operation and maintenance personnel with the timeliness and standardization of fault handling. The cloud-based backend of the full-domain multi-dimensional millimeter-wave AI sensing terminal supports batch configuration and one-click distribution of unattended latency parameters, sensing sensitivity, and human recognition thresholds. It can set differentiated departure latency strategies for different functional scenarios such as hotel rooms, independent office areas, open office areas, conference rooms, and corridors, accurately adapting to diverse human activity patterns. All LoRa sensing terminals have built-in intelligent low-power sleep and wake-up mechanisms, with high-frequency data collection when people are present and low-frequency sleep when people are absent. This effectively extends the device's battery life and reduces the frequency of terminal maintenance while ensuring sensing accuracy.

[0005] Furthermore, the intelligent flexible controller's control logic engine incorporates a DRL deep reinforcement learning octet optimization algorithm specifically tailored for complex scenarios involving strong coupling of source, load, and storage in building microgrids, multiple interferences, and nonlinearity. Unlike general standard reinforcement learning algorithms, it specifically adds source-storage constraints, equipment lifespan constraints, human comfort constraints, and grid quality constraints. The algorithm constructs a four-fold weighted constraint reward function, namely, a positive economic reward for minimizing the building's monthly electricity purchase cost, a negative penalty for exceeding the threshold of indoor environmental parameters, a positive reward for extending equipment lifespan without frequent start-stop cycles, and a low-carbon positive reward for improving photovoltaic green electricity consumption. The weights of the four reward categories support cloud-based online visualization and adjustment, as well as local offline fixed effects. It can switch between three core operating modes—priority energy saving, priority human comfort, and priority green electricity consumption—with one click, and also supports custom weight allocation to form personalized operating strategies. The DRL algorithm has the ability to iterate online and train offline around the clock: the cloud-based digital twin platform automatically summarizes the big data samples of the previous day's full microgrid power generation, energy storage operation, personnel flow, environmental time series, electricity price fluctuations, and load changes, and completes offline model iterative training and neural network weight optimization. The optimized neural network weight parameters are automatically encrypted and sent to the edge NPU inference engine of each floor for permanent fixation, continuously reducing the building's overall energy consumption, increasing the photovoltaic self-consumption rate and energy storage arbitrage income. The longer the algorithm iteration cycle and the more data accumulated, the more significant the system's global optimization effect, adaptability, and energy-saving benefits. The algorithm incorporates six standardized preset scenario templates: weekday office work, lunch break, nighttime security, holiday hibernation, meeting reception, and hotel room sleep, covering all normal operating conditions of the building at all times. It also supports drag-and-drop visual customization, editing, adding, modifying, and deleting exclusive scenario strategies. Custom scenario logic is permanently stored in edge flash memory, and can be called and executed normally and stably even when the cloud is disconnected, the network is interrupted, or the microgrid is isolated mode, without being affected by the status of the upper-level system. The algorithm also incorporates a scene intelligent recognition subroutine, which can automatically match the optimal scenario strategy based on time, personnel, lighting, and load, and achieve unattended adaptive switching.

[0006] Furthermore, the distributed microgrid source-load-storage coordination unit and the lighting / air conditioning / fresh air flexible control unit achieve complete interconnection of underlying data, deep linkage of top-level algorithms, and coordinated matching of operating conditions and strategies, and execute refined and differentiated coordinated scheduling strategies for four typical extreme operating conditions of building operation. Operating Condition 1: Solar power generation surplus at midday; The AI ​​engine actively and synchronously reduces the power of zoned lighting to reduce the indoor heat increase caused by the heat generated by the lamps, thereby reducing the air conditioning cooling load from the source; The reasonable range of air conditioning cooling temperature is appropriately relaxed to avoid excessive cooling and waste, maximizing the local consumption of real-time solar power generation. The surplus electricity that cannot be consumed locally is automatically stored in the energy storage battery to complete green electricity storage, achieving nearly 100% local consumption of solar power generation. Operating Condition 2: Peak electricity prices and high demand conditions in the power grid; the microgrid unit controls the energy storage to discharge and replenish energy at high power levels, replacing the high-priced electricity in the power grid; synchronously linking stepless dimming and power reduction of lighting, precise cooling of air conditioning, and speed reduction of fresh air ventilation, steplessly and smoothly reducing the instantaneous peak load of buildings, effectively reducing electricity costs during peak periods, lowering the maximum demand declaration value of the power grid, avoiding demand penalties, and significantly improving the energy economy of buildings; Operating Condition 3: Insufficient photovoltaic output, cloudy and rainy weather with weak light, and low SOC and low power consumption of energy storage; The AI ​​engine automatically and adaptively tightens the indoor comfort constraint threshold, compressing the air conditioning temperature and humidity adjustment range, limiting the maximum brightness threshold of lighting, and reducing the basic fresh air supply volume while ensuring the basic living experience, accurately and strictly controlling the total building power load, minimizing the amount of electricity purchased from the grid, and reducing the grid electricity cost outside of off-peak hours; Operating Condition 4: Virtual power plant grid load reduction invitation and demand response triggering condition; The system constructs a three-dimensional dynamic adjustable load capacity matrix of lighting, air conditioning and fresh air in real time, accurately calculates the adjustable upper limit, adjustable rate and sustainable adjustment duration of the three types of equipment, and intelligently and optimally allocates the adjustment ratio of the three types of equipment according to the grid target power reduction, and outputs continuous, stable and controllable standardized adjustable load resources to meet the grid demand response accuracy requirements; The system provides real-time statistics on the adjustable load capacity, adjusted capacity, and remaining adjustable capacity across the entire building area within seconds, and periodically reports these statistics to the virtual power plant dispatch platform with encryption. This forms a standardized, flexible load resource ledger that is dynamically updated, traceable, and tradable, supporting regional unified dispatch of the power grid, market-based trading, and demand response subsidy settlement.

[0007] Furthermore, the digital twin virtual-real mapping unit is configured as follows: The offline simulation submodule supports custom input of multi-dimensional refined parameters such as building envelope heat transfer coefficient, window shading coefficient, wall insulation parameters, photovoltaic energy storage installation parameters, electromechanical equipment rated parameters, personnel work and rest patterns, hourly electricity price, and annual meteorological and solar radiation time series. The simulation engine is based on a fusion algorithm of building energy consumption mechanism and AI big data model to automatically complete 8760 hours of hourly dynamic simulation of all working conditions throughout the year. It accurately outputs a complete set of quantitative data, including the building's annual power consumption, photovoltaic green electricity absorption rate, grid peak load, monthly electricity cost, hourly carbon emission reduction, and optimal equipment control threshold. Operation and maintenance personnel can send the optimal simulation parameters to all edge controllers and terminal controllers in batches with one click, eliminating the tedious process of traditional manual debugging, repeated trial and error, and experience parameter settings, and significantly shortening the system debugging cycle. The online virtual-real synchronization submodule completes a full synchronization update of equipment, energy consumption, energy storage, environment, and carbon data every 100ms. The 3D twin interface displays real-time operating power of equipment, real-time output of photovoltaic power generation, remaining energy storage capacity and charging / discharging status, indoor environmental parameters, and hourly carbon emission reduction data in four hierarchical levels: building, floor, zone, and individual room. It supports year-on-year, month-on-month, and time-period comparative visualization analysis of energy consumption, power generation, and carbon data. Managers can remotely, intuitively, and accurately grasp the energy flow path, load control effect, and energy-saving and low-carbon benefits of the entire building. The digital twin platform automatically aggregates big data from the entire operation every day and sends it to the cloud-based DRL training platform to complete offline iterative optimization and parameter upgrades of the model. The updated optimal algorithm parameters are automatically encrypted and sent to the NPU computing chips of each central controller, enabling the system to learn autonomously throughout its entire life cycle, continuously evolve, and become better with use. The longer the building is in operation, the more prominent the system's energy-saving effect, green electricity absorption efficiency, and comprehensive carbon emission reduction benefits become, completely breaking through the technical drawbacks of traditional building automation systems, such as fixed logic, fixed parameters, and no evolutionary ability.

[0008] Furthermore, the cloud-based AI-powered carbon footprint calculation unit can dynamically calculate real-time carbon emissions during the building's operation phase at multiple time scales, including minute, hour, day, and month. It automatically and accurately distinguishes between carbon emissions from grid-connected thermal power purchases, zero-carbon electricity offsets from photovoltaic power, carbon conversions from energy storage charging and discharging, and additional carbon emissions from equipment wear and tear. Simultaneously, it supports batch input of implicit carbon data from the entire process of manufacturing, logistics, engineering installation, and waste disposal of various electromechanical equipment, overlaying annual carbon reduction data during the building's long-term operation period to construct a carbon asset ledger covering the entire lifecycle, all dimensions, and all time series from equipment production to building decommissioning. The system is equipped with a dedicated consortium blockchain encrypted storage node. All minute-level raw energy consumption, photovoltaic power generation, energy storage charging and discharging, environmental monitoring, load regulation, and carbon accounting details are asymmetrically encrypted to generate unique, non-repeatable hash values ​​and stored on the blockchain in real time. The data is tamper-proof, non-deletable, fully traceable, and cross-verifiable, possessing judicial, financial, and certification-level credibility. The cloud-based system supports the automatic generation and export of standardized PDF carbon verification reports and carbon asset statements by year, quarter, month, and custom time period, accurately adapting to five core application scenarios: green building star rating certification, LEED international green building certification, corporate ESG annual report disclosure, CCER carbon sink project development, green finance credit, and green bond application. It also features a built-in real-time carbon trading average price dynamic calculation model, which can automatically and accurately calculate a project's annual carbon emission reduction, potential carbon trading revenue, and energy-saving and carbon-reduction economic value based on real-time prices in the national carbon market, providing comprehensive quantitative data support for the refined, market-oriented, and value-driven operation of building carbon assets.

[0009] Furthermore, the intelligent flexible controller serves as the core hardware carrier at the system edge. The hardware highly integrates a LoRa long-distance communication module, a dual-band WiFi high-speed module, a 5G full-network industrial module, a multi-core NPU high-speed computing chip, a large-capacity power-loss protection flash memory, multiple high-precision analog / digital input / output interfaces, and an Ethernet wired backup communication interface, forming a 5G+WiFi+LoRa+wired Ethernet quadruple communication redundancy guarantee system to prevent system control failure and data interruption caused by a single communication failure. The central controller's large-capacity power-loss protection flash memory can permanently store custom scenario strategies, algorithm weight parameters, historical energy consumption data, sensor monitoring data, microgrid operation raw data, and fault log data. It will not lose or fail when power is off. In the event of any communication link failure, cloud disconnection, or network fluctuation, it can independently and completely run the entire set of AI algorithms, microgrid scheduling strategies, and load flexible control logic locally, and continuously complete millisecond-level high-precision flexible adjustment. After the communication link is restored, the system automatically and intelligently identifies the offline period and completely retransmits all offline raw data and operation logs to the cloud platform and blockchain node, without data loss, computational gaps, or policy interruptions. A single intelligent flexible controller can concurrently and stably connect up to 200 LoRa sensor terminals and end devices, with a full-domain data polling cycle of ≤100ms and a local NPU multi-target AI inference and optimization operation latency of ≤50ms. It fully meets the ultra-high timeliness requirements of microgrid on-grid and off-grid switching, virtual power plant millisecond-level scheduling, and real-time building load optimization. It can efficiently adapt to networking and control scenarios with massive points and wide coverage in large industrial parks, commercial complexes, and multi-building groups.

[0010] Furthermore, the system adopts a six-level hierarchical, finely distributed management and control architecture: park-multiple buildings-single building-floor-independent zone-independent room. Each level of the architecture operates independently, with real-time data exchange and global collaborative linkage. The park cloud platform enables centralized monitoring of multiple buildings, batch policy distribution, group-level carbon data aggregation, unified energy consumption analysis, and unified statistics of transaction revenue. The edge controller of a single building is responsible for the power balance of the microgrid's source, load, and storage, global load optimization, and security management of the entire building. Each floor is equipped with a domain-specific sub-control unit, dividing open office areas, independent offices, meeting rooms, guest rooms, and public corridors into independent management and control zones. Each sub-zone is equipped with a dedicated sensing network and flexible control terminal to achieve fine-grained management and control of the zones. Each independent zone can autonomously execute AI adaptive closed-loop control, supporting customized exclusive living comfort thresholds, unattended delay strategies, scene operation modes, and equipment adjustment precision; the floor and building central controllers coordinate the overall load balance, intelligently coordinate the staggered operation and staggered start-up and shutdown of each zone, effectively eliminating the problem of simultaneous load surges and excessive instantaneous peaks in multiple zones, and achieving optimal energy scheduling, lowest energy consumption, and maximum benefits for the entire building while ensuring personalized living comfort in each zone.

[0011] Furthermore, an AI intelligent control method applied to the system described in any one of claims 1 to 8 is a continuous, uninterrupted, self-iterative closed-loop execution system that operates automatically throughout the entire process without human intervention, comprising the following seven standardized continuous closed-loop steps: S1. Millisecond-level multi-source multi-dimensional data synchronous acquisition: The system collects real-time photovoltaic power generation, a complete set of energy storage BMS operating parameters (SOC, cell voltage, charging and discharging current, battery temperature, equalization status), instantaneous load of charging piles, grid time-of-use electricity price, grid voltage frequency quality, millimeter-wave human posture point cloud data, indoor illuminance, temperature and humidity, carbon dioxide, PM2.5, TVOC, formaldehyde multi-dimensional time-series data in parallel at high frequency, and uploads them to the edge NPU inference engine in real time through LoRa / WiFi / 5G redundant links; it also collects a complete set of safety parameters for energy storage overcharging, over-discharging, over-temperature, and voltage difference anomalies around the clock, providing multiple safety constraints for AI algorithms and ensuring the inherent safety of system operation; S2, Edge-Cloud Layered AI Multi-Objective Fusion Optimization Computation: Divided into three progressive processes: terminal pre-processing, edge local real-time optimization, and cloud daily iterative upgrades; The terminal CNN network accurately identifies human posture and real-time personnel count, and the LSTM 72-hour sliding window time series model predicts environmental parameter fluctuations and photovoltaic output changes, achieving pre-condition prediction and proactive adjustment; The edge NPU loads a dedicated eight-tuple DRL deep reinforcement learning algorithm, integrating multi-dimensional parameters such as microgrid status, time-of-use electricity price, energy storage safety constraints, and electric shading status, to solve for the optimal stepless power combination of lighting, air conditioning, and fresh air from 0 to 100% in real time, supporting one-click switching of three operating modes and custom weight adaptation; The cloud automatically summarizes the full amount of time series big data daily to complete offline model training and weight optimization, and the upgraded algorithm parameters are automatically sent to the edge flash memory for permanent fixation, ensuring stable and consistent execution of strategies under offline network outage and isolated operating conditions. S3, Microgrid Source-Load-Storage and Six-Element Flexible Resource Co-scheduling: The algorithm automatically identifies five core operating conditions in real time: high photovoltaic power generation, high grid peak electricity price, low energy storage SOC, microgrid islanding faults, and virtual power plant power invitation, and adaptively executes differentiated optimal control logic; it links electric shading louvers to construct a six-element active and passive integrated collaborative energy-saving system of photovoltaic-energy storage-shading-lighting-air conditioning-fresh air, and synchronously adjusts the opening of shading louvers, the output brightness of lamps, and the cooling load of air conditioning under strong light and high temperature conditions, reducing building heat gain and cooling energy consumption from the source; in islanding mode, it smoothly and steplessly reduces load according to load level, avoiding grid voltage and frequency fluctuations and equipment tripping; after grid parameters recover, the load is gradually restored in stages, achieving smooth grid connection without impact, surge, or risk of reconnection; S4. Local three-layer redundant offline flexible closed-loop execution: The edge controller sends the optimal stepless adjustment command obtained from the solution to the end device in real time, realizing smooth and shock-free power adjustment of the device; The system has a built-in three-layer fault redundancy protection mechanism: When the cloud connection is lost, the complete set of algorithms and strategies will run independently locally; Wireless link failure will automatically and seamlessly switch to the backup LoRa link; Microgrid will autonomously coordinate and schedule power supply in the power outage island mode of the public power grid; The system monitors the operating status of all devices, sensors and communications in real time at the second level, and accurately locates abnormal faults, issues graded alarms and pushes handling solutions in an instant; After the fault is recovered, all original data and operation logs of the offline period will be automatically and completely retransmitted, with no data loss, no computational gaps and no strategy interruptions throughout the process; S5. Virtual Power Plant Two-Way Flexible Load Market Interaction: The cloud monitors the two-way power invitation commands between the power grid and the virtual power plant in real time, supporting two-way flexible adjustment of load increase and decrease; during peak grid invitations, it intelligently disassembles the adjustable capacity of three types of equipment to stably and accurately reduce the peak load of buildings; during off-peak grid periods, it appropriately relaxes the human comfort threshold and reasonably increases the operating power of equipment to maximize the absorption of surplus power from the grid during off-peak periods, and simultaneously controls the full charging of energy storage during off-peak periods to complete arbitrage energy storage; the system dynamically matches the grid's time-of-use pricing strategy, peak load reduction, off-peak energy storage, and flat-rate optimization to continuously reduce the building's electricity purchase cost, and automatically calculates the demand response adjustment amount and adjustment duration for accurate market-based subsidy calculation. S6. Digital Twin Simulation and AI Model Full Lifecycle Iterative Evolution: Through offline simulation, all parameters of buildings, equipment, weather, personnel, and electricity prices are entered to complete 8760 hours of full-time-series operating condition simulation throughout the year. The optimal control parameters are pre-generated and distributed to edge devices in batches to achieve optimal performance upon commissioning. In online mode, all data is synchronized between virtual and real systems every 100ms. The daily summary of operating data is sent to the cloud training platform to complete model iteration and optimization, continuously improving the algorithm's prediction accuracy and optimization capabilities, and realizing the system's autonomous learning and continuous evolution throughout its entire lifecycle. S7. Blockchain-based Full Lifecycle Carbon Accounting and Trusted Evidence Storage: The AI ​​carbon accounting unit accurately calculates energy consumption data for each device and zone in a time-sharing and itemized manner, integrating official carbon conversion coefficients and implicit carbon data throughout the device's lifecycle to generate a refined carbon asset ledger; all original operating data, control data, and carbon data are encrypted and stored on the blockchain, achieving tamper-proof, full traceability, and cross-verification; the system supports the automatic export of standardized carbon verification reports in multiple dimensions and formats, fully meeting various compliance requirements for green building certification, ESG disclosure, carbon sink development, and green credit; the system executes the above seven steps in a 24 / 7 loop, achieving a fully closed-loop intelligent operation of microgrid collaboration, AI-based pre-judgment, integrated flexible control of three devices, market-based interaction, and trusted carbon data management.

[0012] Furthermore, the AI ​​intelligent control method incorporates multi-level, full-process, and multi-dimensional security protection, energy-saving optimization, scenario adaptation, and market-adaptive segmented control logic in the corresponding steps, as detailed below: S1. The data acquisition process enables the simultaneous collection of seven dimensions of data, including microgrid, equipment, environment, personnel, electricity price, grid quality, and energy storage safety. This constructs a complete, detailed, and timely underlying data foundation, providing comprehensive data support for upper-level AI optimization and safety constraint control, and preventing control distortion caused by data gaps. The S2 AI computing stage incorporates multiple hard constraint mechanisms for energy storage safety, dynamically identifying risks such as overcharging, over-discharging, overheating, and excessive individual cell pressure differences in real time. Once a safety threshold is triggered, it prioritizes automatically reducing flexible loads and suspending energy storage charging and discharging, prioritizing equipment and system safety. The LSTM time-series model continuously iterates and predicts operating conditions, performing pre-adjustment actions in advance, completely resolving energy waste and fluctuations in human comfort caused by lagging control and parameter mutations in traditional systems. All custom scenario strategies are permanently stored locally, ensuring consistent performance under offline, offline, isolated, and other extreme conditions compared to networked conditions. S3, the collaborative scheduling process incorporates a dedicated energy-saving subprocess linked to electric shading louvers, achieving a deep integration of passive shading energy saving and active electromechanical equipment control, reducing building heating and cooling loads from the source; through islanded load tiered smooth load reduction and grid-connected phased logic, it effectively avoids grid voltage fluctuations, equipment tripping, closing impacts, and harmonic disturbances, ensuring the safety of equipment and grid operation in all aspects. S4. The execution phase has triple fault redundancy and seamless switching capabilities, covering all fault scenarios such as network outage, disconnection, power grid failure, and equipment malfunction, ensuring uninterrupted control, unbroken strategies, and no data loss; combined with accurate fault location and standardized handling plan push, it significantly reduces the cost of manual operation and maintenance inspection and the time for fault handling. S5, the market-oriented interaction link supports two-way high-precision power regulation of virtual power plants, deeply adapts to the grid time-of-use pricing, demand response, and electricity spot trading mechanism, dynamically optimizes load operation strategy, maximizes the off-peak power consumption rate, peak-valley arbitrage income and demand response subsidy income, and continuously reduces the electricity cost of the entire life cycle of buildings. S7, the carbon accounting process supports the export and statistics of multi-level hierarchical carbon ledgers for single rooms, single floors, single buildings, and multiple buildings within a group, adapting to the refined carbon asset management needs of group-based building clusters; all on-chain data fully complies with domestic green building standards, international green certification, ESG disclosure, carbon sink development, and green finance credit compliance requirements, realizing one-stop credible traceability, quantitative statistics, compliant declaration, and value transformation of building carbon data. Compared with the prior art, the present invention has the following advantages: 1. Break down the technological silos between microgrid EMS and building electromechanical automation, build a unified edge AI computing power hub, realize the global integrated and coordinated scheduling of photovoltaic, energy storage, charging piles, lighting, air conditioning and fresh air, significantly improve the local photovoltaic self-consumption rate, reduce the peak load of building power grid, and reduce the monthly electricity purchase cost at the same time; 2. Build a three-level full-domain AI computing power architecture consisting of terminal lightweight AI, edge NPU local inference, and cloud reinforcement learning iteration. It is equipped with millimeter wave human body recognition, LSTM time series prediction, and DRL multi-objective optimization algorithms to achieve forward-predictive flexible control, taking into account both extreme energy saving and indoor living comfort. 3. Integrating digital twin offline simulation and online virtual-real synchronization modules, the optimal control strategy is pre-generated in the early stage of the project, which greatly shortens the on-site commissioning cycle; after commissioning, the DRL model is iterated daily to achieve autonomous learning and continuous optimization of energy-saving benefits throughout the system's entire life cycle; 4. Establish a blockchain-based full-lifecycle carbon footprint AI accounting unit. Energy consumption and carbon emission reduction data are encrypted and uploaded to the blockchain, making them tamper-proof. Output green building, ESG, CCER, and green loan compliance carbon verification certificates in one stop, thereby enhancing the added value of zero-carbon building assets. Attached Figure Description

[0013] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the system architecture in this invention. Detailed Implementation

[0014] Specific embodiments of the present invention will be described in detail with reference to the accompanying drawings, but are not intended to limit the invention. It should be understood that various modifications can be made to the embodiments disclosed herein. Therefore, the following description should not be considered limiting, but merely as examples of embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention. Other modifications within the scope of this disclosure will be apparent to those skilled in the art: the accompanying drawings, which are included in and constitute a part of this specification, illustrate embodiments of the present disclosure and, together with the general description of the present disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the present disclosure. The above and other aspects, features, and advantages of the present disclosure will become more apparent when taken in conjunction with the accompanying drawings, in view of the following detailed description. Specific embodiments of the present disclosure are described thereafter with reference to the accompanying drawings; however, it should be understood that the disclosed embodiments are merely examples of the present disclosure, which can be implemented in various ways. Well-known or repeated functions and structures are not described in detail to avoid obscuring the present disclosure with unnecessary or redundant details. Therefore, the specific structural and functional details disclosed herein are not intended to be limiting, but rather serve as the basis and representative examples of the claims, to guide those skilled in the art to use this disclosure in a variety of substantially any suitable detailed structures.

[0015] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings, such as... Figure 1As shown in some embodiments of this application, a building microgrid-coordinated, AI-adaptive, integrated intelligent flexible energy load control system for lighting, air conditioning, and fresh air is proposed. The system constructs a standardized five-layer architecture from top to bottom: a cloud-based digital twin scheduling layer, an edge-controlled computing power layer, a flexible load equipment layer, a global sensing layer, and a microgrid source / storage equipment layer. These five layers are vertically independent and decoupled, while horizontally interconnected, enabling full-domain linkage of computing power, load, sensing, source / storage, and scheduling. The system integrates eight mutually decoupled, independently selectable, and modularly pluggable core functional modules, and also adds an electric louver auxiliary... Flexible linkage interfaces and reserved flexible expansion interfaces for electric sunshades, smart curtains, and building water supply and drainage form a smart energy control system with multi-source coupling of photovoltaic, energy storage, sunshade, lighting, air conditioning, and fresh air. The eight modules include: a distributed microgrid source-load-storage collaborative unit, a cloud-edge-device full-domain AI adaptive computing power architecture, an integrated flexible load control unit for lighting / air conditioning / fresh air, a full-domain multi-dimensional sensing terminal, a 5G+WiFi+LoRa three-layer heterogeneous redundant communication network, a digital twin virtual-real mapping unit, a virtual power plant aggregation and scheduling unit, and a blockchain full life cycle carbon data trusted storage unit.

[0016] In one embodiment, the distributed microgrid source-load-storage collaborative unit is an edge-side core intelligent control terminal developed for new power system park-level microgrid scenarios. It deeply integrates multiple technologies such as power electronic control, energy storage management, high-precision data acquisition, smart grid switching, and AI flexible regulation to construct an integrated collaborative control hardware foundation for "source-load-storage-grid". The unit adopts a highly integrated and modular hardware architecture, featuring convenient installation, strong adaptability, stable operation, and excellent scalability. It can fully adapt to microgrid systems in various scenarios such as newly built industrial parks, commercial complexes, industrial parks, office buildings, and smart communities, while also supporting the intelligent upgrade and transformation of existing microgrids and traditional distribution systems. The entire hardware system consists of five core hardware components: a photovoltaic grid-connected inverter, a multi-type energy storage BMS management host, a park AC / DC charging pile controller, a millisecond-level microgrid on-grid / off-grid switching switch, and a source-storage-load synchronous sampling module. Each device operates independently yet collaboratively, forming a complete closed-loop hardware system for state perception, logical operation, instruction execution, and operating condition switching. This provides solid hardware support for the autonomous operation of the microgrid under all operating conditions, as detailed below: I. The overall hardware configuration system of the unit includes: 1. Grid-connected Photovoltaic Inverter: As a core execution device on the power generation side, it adopts a high-frequency isolated power conversion topology, possessing efficient AC / DC power conversion, grid-connected voltage stabilization, and dual-dimensional active and reactive power regulation capabilities. The device can precisely adapt to the dynamic output characteristics of distributed photovoltaic modules as illumination and temperature change, converting the unstable DC power output from the photovoltaic array into stable AC power conforming to the industrial frequency grid standard. It also supports millisecond-level power response, adjusting output power in real time according to system scheduling commands, effectively adapting to the fluctuating characteristics of photovoltaic power generation. From a hardware perspective, it ensures high-quality grid connection and local consumption of photovoltaic power, preventing problems such as photovoltaic curtailment and power backflow.

[0017] 2. Multi-type Energy Storage BMS Management Host: This is a universal, all-domain energy storage management and control device that overcomes the technical bottleneck of traditional energy storage management systems that only adapt to a single energy storage medium and a single voltage level. It is perfectly compatible with three mainstream energy storage devices on the market: lithium iron phosphate, lead-carbon, and vanadium redox flow storage. Lithium iron phosphate energy storage is suitable for high-frequency charging and discharging and short-term power regulation scenarios; lead-carbon energy storage is suitable for long-term stable power generation and low-cost energy storage scenarios; and vanadium redox flow storage is suitable for large-capacity, long-endurance, and high-safety energy storage scenarios. The host can perform differentiated and precise control based on the electrochemical characteristics, charging and discharging thresholds, and degradation patterns of different energy storage media. Simultaneously, the device supports high- and low-voltage multi-level energy storage cluster adaptation and can simultaneously monitor the operating parameters of energy storage system clusters and individual batteries. This enables unified networking, coordinated scheduling, and unified fault diagnosis for multiple types and voltage levels of energy storage devices, completely solving the pain points of fragmented management and poor coordination in multi-energy storage mixed-use scenarios.

[0018] 3. Park AC / DC Charging Pile Controller: This is the core device for flexible load control, supporting full access to all AC slow-charging and DC fast-charging piles within the park. It can collect real-time network data on single-pile charging power, charging current, charging voltage, charging time, and equipment fault status, among other dimensions. Based on big data analysis, it accurately identifies the tidal characteristics and instantaneous load mutation patterns of the park's charging load. It can dynamically control the start / stop status and charging power rate of charging piles according to the real-time power margin of the microgrid, achieving flexible and controllable electric vehicle charging load and orderly off-peak operation, effectively avoiding problems such as grid load spikes, voltage drops, and power imbalances caused by concentrated charging.

[0019] 4. Millisecond-level microgrid grid-connected / off-grid transfer switch: This is the core power hardware for microgrid switching. Employing vacuum arc extinguishing and rapid electromagnetic switching technologies, it boasts advantages such as fast breaking speed, high closing accuracy, low electrical loss, and strong impact resistance. It can precisely support millisecond-level switching between two core operating conditions: microgrid grid-connected operation and islanded independent operation. Simultaneously, it features multiple hardware protection functions including overvoltage, overcurrent, overload, and short circuit protection, effectively resisting electrical shocks during the switching process and ensuring the safety of system equipment and lines.

[0020] 5. Source-Storage-Load Synchronous Sampling Module: Serving as the system's sensory nerves, this module employs a high-precision synchronous sampling chip to achieve synchronous high-frequency data acquisition from four core components: photovoltaic power generation, energy storage operation, park load, and the public power grid. The module boasts a sampling accuracy of up to 0.2%, with a data sampling period down to the millisecond level. It can accurately capture subtle changes in system operating conditions such as instantaneous power fluctuations, parameter distortions, and sudden load changes, eliminating control decision lags and inaccuracies caused by data delays and sampling deviations. This provides a precise, real-time, and synchronous data foundation for intelligent system scheduling and fault management.

[0021] II. This unit abandons the traditional microgrid's "cloud-based decision-making, edge-based execution" dependency architecture, and instead locally embeds the entire set of core control algorithm subroutines into edge flash memory. This includes three core programs: a source-load-storage dynamic matching subroutine, a power smoothing subroutine, and an islanded fault-tolerant control subroutine. All subroutines can operate independently locally and execute autonomously in a closed loop, requiring no cloud computing power or network command transmission. They can function normally and fully under extreme conditions such as network outages, cloud disconnection, and grid anomalies, achieving full-scenario adaptive intelligent control of the microgrid. This includes: 1. Source-Load-Storage Dynamic Matching Subroutine: This is the core algorithm for microgrid energy optimization and economic operation. Based on real-time collected multi-dimensional data such as photovoltaic power generation, energy storage SOC status, battery available capacity, dynamic load of the park, and time-of-use electricity price, it constructs a high-precision three-dimensional dynamic coupling matching model of source-load-storage. The algorithm can perform second-level power reallocation based on the instantaneous fluctuations in photovoltaic power generation caused by changes in sunlight, the tidal changes in park load during commuting hours, and the load mutation characteristics of concentrated charging of charging piles, dynamically optimizing the matching relationship between photovoltaic output, energy storage charging and discharging power, and load power consumption. In terms of clean energy consumption, the subroutine always adheres to the principle of "prioritizing local photovoltaic consumption," prioritizing the direct supply of photovoltaic power to the local load of the park, and storing all surplus electricity in the energy storage system to minimize the redundant photovoltaic power on the grid and reduce transmission line losses and curtailment rates. In terms of economic dispatch, it deeply integrates with the peak-valley time-of-use pricing mechanism of the power grid to intelligently formulate energy storage charging and discharging arbitrage strategies: during the off-peak electricity price period, it controls the energy storage to charge at full power and store electricity at low prices; during the peak electricity price period, it dispatches the energy storage to discharge and replenish energy, replacing the high-priced grid-purchased electricity, and continuously reduces the electricity cost of the park through the peak-valley price difference, thereby maximizing the economic benefits of the microgrid system operation.

[0022] 2. Power Smoothing Subroutine: Photovoltaic power generation is characterized by intermittency, randomness, and volatility. Cloudy, rainy, and shading weather conditions can cause drastic fluctuations in photovoltaic output power. These fluctuations, coupled with load disturbances such as equipment start-ups and shutdowns in the industrial park and concentrated charging by charging piles, can easily lead to microgrid bus voltage deviations, frequency fluctuations, and power oscillations, affecting power quality and equipment safety. To address this issue, the power smoothing subroutine leverages the millisecond-level rapid charge-discharge response characteristics of the energy storage system to achieve dynamic compensation and steady-state control of microgrid power. The subroutine can monitor the amplitude of photovoltaic power fluctuations, grid voltage and frequency deviations, and load change rates in real time. It accurately calculates the system's active and reactive power deficits and redundancies, and rapidly absorbs redundant power and replenishes power deficits by controlling energy storage, quickly smoothing short-term power oscillations in the system. This effectively solves the problem of power quality exceeding standards caused by photovoltaic fluctuations and load disturbances, stabilizes the microgrid bus voltage and power frequency, and avoids faults such as frequent equipment start-ups and shutdowns, relay protection malfunctions, grid tripping, and equipment damage caused by drastic power fluctuations, comprehensively ensuring the stability of the microgrid system's power supply and compliance with power quality standards.

[0023] 3. Islanding Fault-Tolerant Control Subroutine: This is the core fallback algorithm for microgrid fault conditions, specifically adapted to islanding operation scenarios under various abnormal conditions such as public grid power outages, line faults, voltage and frequency exceeding limits, and grid distortion. Traditional microgrid islanding operations are prone to problems such as disordered load outages, source-load power imbalances, energy storage overloads, and critical load power failures. This fault-tolerant control subroutine can work with hardware detection equipment to accurately pinpoint the type and severity of islanding faults, and execute a series of closed-loop control logics, including tiered load management, power fault-tolerant balancing, equipment overload protection, and fault state fallback correction. Throughout the islanding operation, the subroutine continuously and dynamically balances the internal generation power and load power of the microgrid, corrects operating parameters in real time, avoids risks such as system power imbalances, equipment overloads, and voltage and frequency runaway, while ensuring uninterrupted power supply to core loads. This minimizes the impact of grid faults on the park's production, office, and maintenance, significantly improving the reliability, security, and fault tolerance of microgrid islanding operations.

[0024] III. The distributed microgrid source-load-storage collaborative unit relies on a high-precision source-storage-load synchronous sampling module to build a full-volume, high-frequency, synchronous data acquisition system covering four dimensions: "source, storage, load, and grid." This enables comprehensive and seamless perception of the microgrid's operational status, providing core data support for localized intelligent decision-making, precise control, and fault diagnosis. The core collected data is categorized as follows: Generation-side data includes photovoltaic instantaneous active / reactive power generation, photovoltaic module operating temperature, module open-circuit voltage, array operating status, inverter AC / DC voltage and current, inverter operating temperature, equipment fault codes, real-time power generation efficiency, and other full-dimensional photovoltaic power generation operating parameters, providing complete feedback on the real-time output status and equipment health of the photovoltaic system.

[0025] Energy storage side data: covering refined parameters such as overall SOC, SOH, real-time charging and discharging power, total voltage and current of the cluster, individual cell voltage, individual cell temperature, cell temperature difference, charge and discharge rate, insulation status of energy storage equipment, fault alarm information, start-up and shutdown status, etc., to accurately control the operational safety and available capacity of the energy storage system.

[0026] Load-side data includes the overall total power load of the park, the load power of each zone branch, the dynamic charging load of AC / DC charging piles (single pile and cluster), the operating power, start / stop status, and load change rate of terminal electrical equipment such as office equipment, lighting systems, air conditioning and fresh air systems, and security equipment, providing a comprehensive understanding of the dynamic changes in the park's load.

[0027] Grid-side data includes core parameters such as time-of-use peak and off-peak electricity prices, real-time power frequency, three-phase voltage amplitude, phase angle, voltage frequency distortion rate, grid harmonic content, and grid connection outlet power. It monitors the grid operation status and power quality in real time, providing grid benchmark data for grid-connected and off-grid switching and power optimization scheduling.

[0028] All collected data adopts a millisecond-level synchronous sampling mechanism, which has extremely low data transmission latency, high acquisition accuracy, and strong synchronization. It can accurately capture the instantaneous and subtle changes in the system's operating conditions, ensuring that the system control strategy can adapt to operating condition fluctuations in real time and guarantee the accuracy and timeliness of regulation.

[0029] IV. The core technological innovation of this unit lies in its fully functional, localized, autonomous closed-loop operation, completely overturning the traditional microgrid's reliance on cloud platforms. The system embeds all microgrid scheduling and optimization logic, equipment safety constraint logic, power dynamic optimization logic, fault protection logic, and operating condition switching logic into a local edge flash memory chip. Algorithm execution, logic judgment, and instruction issuance are all completed locally at the edge, eliminating the need for continuous control commands from the cloud and avoiding the consumption of cloud computing power and network bandwidth. This unit completely eliminates dependence on the cloud and network, and can independently maintain stable full-function operation under various extreme abnormal conditions such as network outages, cloud disconnection, abnormal grid voltage and frequency, line disturbances, and communication failures. The unit can autonomously complete six core closed-loop control operations, covering all microgrid operation scenarios: First, local photovoltaic (PV) absorption maximization control, tapping the potential of clean energy utilization and reducing reliance on external grid power purchases; second, energy storage peak-valley charging and discharging arbitrage control, refining energy storage operation strategies and improving system economic benefits; third, islanded load hierarchical fault-tolerant management, providing differentiated protection for core loads and flexibly regulating ordinary loads; fourth, grid fault-tolerant operation control, proactively preventing fault escalation and ensuring system stability and self-sufficiency; fifth, grid recovery-free smooth grid connection control, achieving seamless switching of operating conditions; and sixth, post-grid connection power gradual retrospective optimization control, automatically restoring the system to its optimal operating conditions. The entire process requires no manual intervention or cloud support, resulting in a significant improvement in system autonomy, operational stability, and anti-interference capabilities under extreme conditions.

[0030] V. Dedicated On-Grid Switching Controller and Precise Islanding Detection Technology: Addressing the industry pain points of traditional microgrid islanding detection, such as slow response, weak interference identification, susceptibility to misjudgment and incorrect switching, and significant electrical impact during operational condition switching, this unit features a dedicated microgrid on-grid switching controller. Equipped with a hardware-level high-speed islanding detection circuit and a self-developed high-precision voltage and frequency distortion identification algorithm, it achieves rapid perception, accurate identification, and smooth switching of islanding conditions. In terms of response performance, the device's hardware detection circuit possesses ultra-high sensitivity, with a detection delay of ≤10ms for abnormal signals such as voltage drops, frequency shifts, and phase distortions caused by islanding. This is far superior to the second-level detection performance of traditional microgrids, enabling rapid perception of fault conditions and reserving sufficient control time for subsequent load regulation and operational condition switching. Regarding identification accuracy, the self-developed algorithm, through cross-comparison and logical judgment of multi-dimensional parameters (voltage, frequency, phase, and power), can accurately distinguish three easily confused operational scenarios: real islanding faults in the public power grid, instantaneous voltage and frequency fluctuations in the power grid, and sudden load changes within the industrial park. It eliminates the misjudgment and incorrect switching problems caused by single parameter detection in traditional equipment, and avoids problems such as power outages, equipment start-up and shutdown shocks, and system malfunctions caused by unnecessary operating condition switching, greatly improving the accuracy and stability of microgrid operating condition switching.

[0031] VI. Hierarchical Flexible Load Reduction Control Strategy in Island Mode: When the public power grid experiences a power outage, voltage or frequency exceeds limits, or power quality fails to meet grid connection standards, the microgrid system immediately triggers an island independent operation mode. The cloud-edge-device full-domain AI computing architecture simultaneously activates the preset hierarchical flexible load reduction intelligent strategy. The system abandons the traditional microgrid's crude management model of "one-size-fits-all fault handling and instantaneous power outage." Based on the importance, safety level, and interruptibility attributes of the park's power load, it constructs a standardized load level system of Level 1, Level 2, and Level 3, achieving differentiated, flexible, and shock-free load control, balancing power supply security, equipment stability, and user experience. Details are as follows: 1. Primary Basic Load Guarantee: This category comprises the core, essential loads for the safe operation, equipment maintenance, and basic office needs of the park. It mainly includes critical loads such as basic emergency lighting, constant temperature and humidity equipment in the computer room, basic fresh air ventilation system, comprehensive security monitoring, fire control system, and core server equipment. Throughout the entire operation cycle of the microgrid island, the system will prioritize the allocation of power from photovoltaic power generation and energy storage, ensuring 100% full-power, uninterrupted, and non-derating operation of the primary loads. This guarantees uninterrupted basic security and control, core equipment maintenance, and basic office conditions, eliminating the risk of power outages for core business and security systems.

[0032] 2. Level 2 Regular Office Loads: These mainly consist of daily office loads such as general office lighting, office computers, regular office air conditioning, and basic power consumption in meeting rooms. These are auxiliary loads for the normal operation of the park and pose no risk of instantaneous power outages. Level 3 Non-Essential Public Area Loads: These are redundant loads that can be completely de-loaded or suspended. They include non-essential electrical equipment such as landscape lighting in public areas, air conditioning in unused areas, corridor auxiliary lighting, and non-functional display equipment. In the event of grid failure and islanded operation or tight system power supply, Level 2 and 3 loads will not be directly de-energized or instantly shut down, effectively avoiding current surges and equipment damage caused by instantaneous power outages to office equipment and electrical wiring. The system, controlled by an integrated flexible control unit, smoothly and steplessly reduces the load power level step by step according to the priority of "Level 3 first, then Level 2," gradually reducing the overall electrical load and achieving shock-free flexible load reduction. While ensuring a basic office experience, it quickly balances the microgrid's source-load power gap, extending the self-sustaining operation time of the islanded system.

[0033] VII. AI Adaptive Energy Consumption Optimization Control Logic: In the isolated operation state of the microgrid, the system prioritizes the local consumption strategy of clean energy, storing all surplus power generated by the photovoltaic system into energy storage batteries to continuously replenish the available energy storage capacity, maximizing the utilization rate of clean energy and reducing energy waste. When the system monitors in real time that the energy storage SOC drops to the preset safe operating threshold, the available energy storage capacity is insufficient, or the overall power supply capacity of the microgrid cannot continuously support stable operation at full load, the AI ​​intelligent energy consumption optimization engine automatically triggers an adaptive and refined energy consumption control strategy. This strategy takes "ensuring basic living experience, meeting basic office environment standards, and not affecting core operations" as its bottom line. It will not rudely shut down equipment, but will instead use intelligent algorithms to adaptively tighten the comfortable range of the indoor environment, and reasonably optimize and fine-tune environmental parameters such as indoor temperature, humidity, illuminance, and air freshness. The system dynamically adjusts the cooling and heating power of air conditioning, the ventilation frequency of the fresh air system, and the brightness and power of lighting equipment. Under the premise that there is no significant difference in human perception and the basic office experience is not affected, it accurately compresses the ineffective operating power of various environmental devices and strictly controls the redundant energy consumption of the system. This AI adaptive control logic operates in a fully automatic closed loop without human intervention. It can accurately balance the energy supply and demand contradiction under islanded conditions, effectively alleviate the pressure on energy storage power supply, significantly extend the self-sustaining operation time of microgrid islands, and significantly improve the power supply guarantee capability and system fault tolerance level of the park under grid failure scenarios.

[0034] 8. Fully Automatic Seamless Grid Connection Restoration Mechanism: Once the public power grid fault is completely eliminated and core parameters such as grid voltage, frequency, phase, and harmonics are restored to national grid connection standards, the microgrid system automatically initiates a fully automatic grid connection restoration process. The entire process requires no manual on-site verification, no manual operation of closing the circuit breaker, and no manual parameter adjustment, achieving unattended, seamless, impact-free, and surge-free safe grid connection. To completely solve the problems of voltage and phase mismatch, large closing inrush current, sudden power fluctuations, and sudden load shutdowns and restarts in traditional grid connection operations, the system adopts a phased, gradient, and smooth backtracking grid connection restoration mechanism. The first step involves the system accurately collecting voltage and phase parameters between the microgrid and the public power grid. An algorithm is used to precisely match the phases at both ends, completely eliminating grid connection phase differences and preventing closing surges at their source. The second step involves gradually and smoothly restoring the power of third- and second-level conventional loads according to load priority, from low to high, thus gradually restoring auxiliary power to the park. The third step, once the system's grid-connected operation is stable, fully restores the normal operating power of all loads in the park, and the system's power operation gradually returns to its optimal scheduling condition. The entire grid connection restoration process is smooth and controllable, without closing voltage surges, instantaneous current surges, or violent power fluctuations. It will not cause any electrical damage to the park's electrical equipment, energy storage systems, photovoltaic inverters, or power grid distribution lines. It perfectly achieves a seamless and smooth transition from islanded operation mode to normal grid-connected operation mode, significantly improving the automation level and safety of the microgrid system during operation transitions.

[0035] In one embodiment, the cloud-edge-device global AI adaptive computing architecture and intelligent flexible central controller technology are described in detail below: I. Cloud-Edge-End Full-Domain AI Adaptive Computing Power Architecture: This system innovatively constructs a three-level distributed heterogeneous computing power system: lightweight terminal inference, real-time edge optimization, and iterative training in the cloud. This completely breaks through the technical bottlenecks of traditional intelligent building and microgrid control systems, which suffer from "dispersed computing power, fragmented data, rigid strategies, and lagging linkage." The architecture is based on the core design concepts of layered computing power adaptation, hierarchical task scheduling, and full-domain data interoperability. Addressing the diverse characteristics of building microgrids, such as strong source-load-storage coupling, dynamic fluctuations in living environment parameters, time-of-use changes in grid electricity prices, and complex external meteorological interference, it achieves intelligent collaborative control across all scenarios, all time periods, and all dimensions, comprehensively improving building energy efficiency, living comfort, and system operational stability. Details are as follows: 1. Terminal Layer: Lightweight AI Pre-sensing Inference Unit. The underlying architecture consists of a lightweight AI inference unit built into the all-domain sensing terminal, serving as the fundamental carrier for system data acquisition and pre-judgment. The terminal layer abandons the traditional single-function mode of sensors that only collect data without analysis. It is equipped with a tailored and optimized lightweight AI inference model, possessing localized data preprocessing, feature extraction, and status prediction capabilities. It is primarily responsible for the pre-sensing and preliminary analysis of all-domain environmental parameters, personnel status, and equipment operating conditions. Specific data collection and prediction dimensions include dozens of core parameters such as real-time microgrid output data, energy storage battery charging and discharging operation status, grid time-of-use electricity price dynamic data, millimeter-wave human multimodal point cloud data, indoor temperature and humidity, carbon dioxide concentration, PM2.5, TVOC, natural light intensity, outdoor full-dimensional meteorological time-series data, and building zone load time-series data. The raw data collected by all terminal devices and the preliminary prediction results are uniformly and standardizedly accessed into the multi-source time-series database of the system, realizing unified storage, unified parsing, and unified scheduling of data across the entire domain. This provides complete, accurate, and real-time data support for real-time optimization at the mid-layer edge and iterative training at the top-layer cloud, solving the industry pain points of inconsistent data formats, high data latency, invalid data redundancy, and severe data silos in traditional building systems from the source.

[0036] 1.2 Edge Layer: The real-time computing power hub of the intelligent flexible controller. The middle layer of the architecture uses the intelligent flexible controller as the core computing and control carrier, equipped with a high-performance NPU local inference engine. It is the core unit for real-time control, dynamic optimization, and emergency fault tolerance of the system. Unlike traditional PLC controllers and ordinary gateway devices that can only execute fixed logic instructions, this controller has a complete set of independently developed core algorithms locally embedded, including four core algorithm modules: microgrid collaborative scheduling algorithm, multi-load linkage precise adjustment algorithm, DRL deep reinforcement learning multi-objective optimization algorithm, and multi-level fault-tolerant redundancy algorithm. Relying on the high-speed computing power of the local NPU, the controller can achieve millisecond-level real-time optimal control solutions, dynamically adapting and adjusting for building lighting, air conditioning, fresh air, energy storage, photovoltaics, and various office and production loads, balancing the microgrid's source, load, and storage operation in real time. Meanwhile, the device is equipped with a large-capacity power-loss protection flash memory and a dedicated power-loss data retention mechanism. All local algorithm parameters, operating strategies, and historical data can be permanently stored, completely avoiding parameter loss and strategy reset issues caused by power outages or restarts. This provides core assurance for the system's offline autonomous operation and uninterrupted control. Based on its robust local computing power and algorithm reserves, the system possesses extremely strong fault resistance capabilities. When on-site fault scenarios such as external fiber optic cable disconnection, cloud server outage, public power grid failure, or single or multiple communication link interruptions occur, the intelligent flexible controller can autonomously switch to local closed-loop operation mode, independently executing all flexible control logic. This ensures that core functions such as adaptive dimming of lighting, precise control of air conditioning temperature and humidity, dynamic adjustment of fresh air exchange volume, coordinated scheduling of microgrid energy storage charging and discharging, and optimal allocation of building load remain uninterrupted and uncontrolled, greatly improving the operational reliability and scenario adaptability of the building microgrid system.

[0037] 3. Cloud Layer: The top layer of the architecture is a cloud-based reinforcement learning iterative training platform, coupled with a digital twin visualization platform. It undertakes the core functions of big data accumulation, model iterative optimization, parameter updates, strategy upgrades, and global scheduling. The cloud platform continuously aggregates full-volume time-series data uploaded from all terminals and edge devices, including microgrid photovoltaic power generation output, energy storage operation status, personnel flow trajectories and density, indoor and outdoor environmental parameters, grid price fluctuations, and building zoning load changes, constructing a massive, realistic, and dynamic scenario sample library. Based on this massive amount of real-world operational samples, the cloud automatically completes offline model iterative training and neural network weight optimization daily, continuously correcting algorithm deviations, optimizing constraint weights, and improving the accuracy of multi-objective collaborative optimization. The iteratively optimized neural network weight parameters and strategy model parameters are permanently stored in the edge NPU inference engines of each floor via encrypted channels, realizing a closed-loop iterative mechanism of cloud training, edge deployment, and global updates. As system runtime increases and data samples continue to accumulate, the algorithm's scenario adaptability, global optimization effect, energy-saving benefits, and green electricity consumption capacity will continue to improve, achieving long-term evolution of system performance.

[0038] II. Customized DRL Deep Reinforcement Learning Octave Optimization Algorithm: Addressing the complex operating scenarios of building microgrids characterized by strong source-load-storage coupling, multiple interferences, nonlinearity, and strong dynamics, the system's intelligent flexible controller control logic engine incorporates a custom-designed DRL deep reinforcement learning octave optimization algorithm. This algorithm differs from general open-source and standardized reinforcement learning algorithms, overcoming the shortcomings of general algorithms such as limited adaptability to specific scenarios, simple constraints, and inability to simultaneously address multi-dimensional requirements related to economy, comfort, equipment lifespan, and grid quality. It specifically adds four exclusive constraint terms to construct a customized algorithm system adapted to the complex operating conditions of building microgrids. Details are as follows: 1. Core Constraint System of the Algorithm: The algorithm is based on an octet core architecture, superimposed with four dimensions of rigid constraints: source-storage constraints, equipment lifespan constraints, human comfort constraints, and grid quality constraints. These constraints comprehensively cover the core requirements of system operation in terms of safety, economy, comfort, low carbon emissions, and stability. Among them, the source-storage constraints are used to regulate the output range and operating logic of photovoltaics and energy storage, avoiding operation beyond rated conditions; the equipment lifespan constraints reduce equipment losses and extend the service life of electromechanical equipment by avoiding frequent start-ups and shutdowns and overload operation; the human comfort constraints strictly limit the threshold values ​​of parameters such as indoor temperature, humidity, air quality, and illuminance to ensure the quality of indoor living and office environments; and the grid quality constraints effectively suppress problems such as harmonics, voltage fluctuations, and abnormal power factors, ensuring the power quality of microgrid grid-connected operation.

[0039] 2. Quadruple Weighted Constraint Reward Function: To achieve multi-objective collaborative optimal control, the algorithm constructs a quadruple weighted constraint reward function. Through a mechanism combining positive rewards and negative penalties, it dynamically optimizes the system's operating strategy. The four reward dimensions mutually constrain and coordinate with each other: First, a positive economic reward for minimizing the building's monthly electricity purchase cost. The algorithm dynamically matches the grid's time-of-use pricing, optimizes the charging and discharging periods of energy storage, and adjusts the building load during peak hours to reduce peak electricity purchases and off-peak electricity waste losses, maximizing the reduction of building electricity costs and achieving the economic operation goal. Second, a negative penalty for indoor environmental parameters exceeding thresholds. When indoor environmental parameters such as temperature, humidity, carbon dioxide concentration, PM2.5, and TVOC exceed human comfort thresholds, the algorithm automatically triggers a penalty mechanism, forcing the system to quickly adjust the operating status of air conditioning, fresh air, and lighting equipment to prioritize the quality of the living environment. Third, a positive reward for extending equipment lifespan by avoiding frequent start-ups and shutdowns. The algorithm intelligently smooths equipment operating conditions, avoiding frequent start-ups and shutdowns and significant fluctuations in operating conditions for equipment such as air conditioners, fans, and energy storage converters. This reduces mechanical and electrical losses, earns equipment lifespan extension rewards, and lowers equipment maintenance costs and replacement frequency. Fourth, there are low-carbon positive rewards for improved photovoltaic green electricity consumption. The algorithm prioritizes local consumption of photovoltaic green electricity, optimizes photovoltaic output matching load curves, reduces photovoltaic curtailment rates, increases the proportion of green electricity self-consumption, reduces building carbon emissions, and earns low-carbon operation rewards.

[0040] 3. Flexible Multi-Mode Switching Mechanism: Four reward weights support cloud-based online visualization and adjustment, as well as local offline fixed effects, offering convenient operation and flexible adaptation. Based on differences in weight ratios, the system presets three core operating modes, which can be quickly switched with one click: priority energy-saving mode, priority human comfort mode, and priority green energy consumption mode, adapting to different scenarios such as weekday office work, holiday hibernation, peak electricity consumption, and green energy surplus. Simultaneously, it supports user-defined weight ratios, allowing users to freely combine priorities of economy, comfort, low carbon, and lifespan extension to form exclusive personalized operating strategies, adapting to the differentiated operating needs of different buildings and business types.

[0041] III. DRL algorithm iterative upgrades and intelligent scene adaptation capabilities, as detailed below: 1. 24 / 7 Online Iteration and Offline Upgrade Mechanism: The DRL algorithm of this system possesses the dual capabilities of continuous online iteration and offline training and upgrade, forming a closed-loop intelligent evolution system. The cloud-based digital twin platform automatically aggregates the previous day's full-domain operational big data daily, covering microgrid power generation data, energy storage charging and discharging operation data, personnel flow time-series data, indoor and outdoor environmental monitoring data, grid electricity price fluctuation data, and building zoning load change data, constructing a complete scenario sample dataset. Based on massive real-world operational samples, the cloud completes offline model iterative training and precise optimization of neural network weights, correcting strategy deviations under different scenarios and improving the algorithm's adaptability to complex operating conditions. The optimized neural network weight parameters and strategy model parameters are permanently fixed to the edge NPU inference engine of each floor through an encrypted transmission channel, ensuring the latest and optimal strategy is implemented in real time at the edge. The longer the system iteration cycle and the richer the data accumulation, the more significant the global optimization accuracy, scenario adaptability, energy-saving benefits, and green electricity utilization efficiency become, realizing the system's evolutionary characteristics of "becoming smarter with use and more efficient with operation."

[0042] 2. Standardized Scene Templates and Custom Editing Functions: To cover the operational needs of buildings across all time periods and operating conditions, the algorithm incorporates six standardized preset scene templates, encompassing six core scenarios: weekday office work, low-load lunch break, nighttime security duty, holiday hibernation, conference reception, and hotel room sleep. These templates are fully adaptable to the routine operating conditions of various business types, including office buildings, industrial parks, commercial complexes, and hotels, enabling basic intelligent control without complex configuration. The system also boasts strong customization and expansion capabilities, supporting drag-and-drop visual scene editing. Users can independently add, modify, and delete exclusive scene strategies, freely configuring equipment operating logic, load adjustment thresholds, energy storage scheduling rules, and environmental control parameters for different scenarios. All custom scene logic is permanently stored in high-capacity edge flash memory. Even in extreme scenarios such as cloud connectivity failure, complete network outage, or isolated microgrid operation, custom strategies can still be invoked and executed stably, completely unaffected by the upper-level system's operating status, ensuring the continuity and stability of the control logic.

[0043] 3.3 Intelligent Scene Recognition and Unmanned Adaptive Switching: The algorithm is equipped with an intelligent scene recognition subroutine, possessing multi-dimensional parameter fusion and judgment capabilities. It can collect and analyze diverse parameters in real time, such as time information, on-site personnel density, natural light intensity, real-time building load, and grid electricity price status, automatically matching the optimal preset or custom scene strategy to achieve unmanned adaptive switching at all times. This completely eliminates the extensive management mode of traditional systems with fixed timed control and manual switching, truly realizing the autonomous perception, autonomous judgment, and autonomous control of building energy and environmental systems.

[0044] IV. The hardware architecture and redundancy protection system of the intelligent flexible central controller are as follows: 1. Highly Integrated Hardware Configuration: The intelligent flexible controller, serving as the core hardware carrier at the system edge, adopts an industrial-grade highly integrated design, integrating multiple communication modules, high-speed computing chips, storage units, and interface modules. The hardware boasts stable performance and strong scalability. Core hardware configurations include: a LoRa long-range communication module, a dual-band WiFi high-speed communication module, a 5G full-network industrial communication module, a multi-core NPU high-speed computing chip, large-capacity power-loss protected flash memory, multiple high-precision analog / digital input / output interfaces, and an Ethernet wired backup communication interface. The overall hardware meets industrial-grade standards for wide-temperature operation, interference resistance, and high reliability, making it suitable for complex industrial environments such as computer rooms, power distribution rooms, and equipment rooms.

[0045] 2. Quadruple Communication Redundancy Mechanism: Leveraging the advantages of multi-module hardware integration, the system constructs a quadruple communication redundancy system of 5G + WiFi + LoRa + wired Ethernet, enabling multiple communication links to serve as backups and automatically switch over each other. When a single communication link fails, experiences lag, or suffers from excessive latency, the system can automatically switch to a backup communication link within milliseconds, preventing system control failures, data interruptions, and control lags caused by single communication failures, significantly improving system communication stability and operational reliability. Specifically, the LoRa module handles long-distance, low-power terminal device networking communication, the WiFi module ensures high-speed local data transmission, the 5G module enables high-speed remote interaction with the cloud, and wired Ethernet serves as a backup communication method, comprehensively covering the needs of different networking scenarios.

[0046] 3. Permanent Data Storage and Resume Capability During Power Loss: The central controller is equipped with high-capacity power-loss protected flash memory, which can permanently store core system data and strategy parameters, including custom scenario strategies, algorithm optimization weight parameters, historical energy consumption statistics, raw data from full-domain sensor monitoring, microgrid runtime sequence data, and device fault log data. All data is guaranteed to be unlost, unaffected, and recoverable upon restart. In any abnormal scenario such as communication link failure, cloud disconnection, network fluctuations, or external network interruption, the device can independently and completely run the entire set of AI algorithms, microgrid scheduling strategies, and load flexible control logic locally, continuously outputting millisecond-level high-precision flexible adjustment commands to ensure uninterrupted system operation. When the communication link is restored, the system can automatically and intelligently identify the offline operation period and completely retransmit all offline raw data, operation logs, and control records for that period to the cloud platform and blockchain nodes, achieving a fully closed-loop operation without data loss, computational gaps, or policy interruptions.

[0047] V. System Computing Power and Large-Scale Network Adaptability: The intelligent flexible controller boasts ultra-high computing efficiency and large-scale networking capabilities, adapting to the massive point control needs of large building clusters. A single device can concurrently and stably connect up to 200 LoRa sensor terminals and end devices, covering the terminal networking needs of a single large building or multiple small buildings. The data polling cycle for all devices is ≤100ms, enabling rapid collection of real-time terminal status across the entire domain, ensuring data timeliness; the local NPU's multi-objective AI inference and optimization computation latency is ≤50ms, allowing for extremely rapid completion of multi-objective optimization solutions and control command output. The ultra-high timeliness computing power fully meets the ultra-high timeliness requirements of rapid switching between microgrid and off-grid operations, millisecond-level precise scheduling of virtual power plants, and real-time dynamic optimization of building loads. It can efficiently adapt to network control scenarios with massive points, wide coverage, and complex operating conditions, such as large industrial parks, commercial complexes, multi-building clusters, and smart parks, possessing strong scenario versatility and scalability.

[0048] In one embodiment, the implementation details of the 5G+WiFi+LoRa three-layer heterogeneous redundancy communication network technology are as follows: I. Overall Network Architecture Deployment Scheme: The 5G+WiFi+LoRa three-layer heterogeneous redundant communication network built in this embodiment adopts an overall architecture of parallel deployment of multiple communication modules, hierarchical networking, and collaborative redundancy. At the hardware level, it synchronously integrates three core modules: an industrial 5G full-network compatible module, a dual-band 2.4G / 5G WiFi module, and a LoRa low-power star topology, abandoning the limitations of traditional single communication modes. The three types of communication modules are deployed independently and perform their respective functions, while achieving deep integration through a system-level linkage mechanism. This constructs a full-scenario industrial communication system covering low-power acquisition, local high-speed interaction, and remote reliable transmission, fully adapting to the communication needs of multiple business scenarios such as industrial power grids, virtual power plants, and digital twin systems. At the hardware deployment level, all communication modules adopt industrial-grade hardware specifications, possessing industrial characteristics such as wide-temperature operation, resistance to electromagnetic interference, dust and water resistance, and long-term stable operation, making them suitable for complex industrial site environments such as factories, substations, and building clusters. The three types of modules are integrated into the core communication terminal equipment of the system, eliminating the need for additional independent networking hardware. They feature high hardware integration, low deployment costs, and strong adaptability, enabling rapid deployment in various industrial intelligent transformation scenarios.

[0049] II. Core Communication Fault Tolerance and Switching Mechanisms: The core advantage of this network lies in its built-in intelligent mechanisms: real-time end-link fault monitoring, automatic fault identification, and seamless link switching. These mechanisms thoroughly resolve industry pain points such as link interruptions, data packet loss, transmission delays, and system downtime in traditional communication networks, ensuring stable 24 / 7 operation of the communication system, including: 1. Real-time fault monitoring mechanism: The system is equipped with a high-precision link status monitoring algorithm, which can continuously collect and monitor core parameters such as signal strength, transmission rate, latency fluctuation, packet loss rate, link connectivity, and device operating status of 5G, WiFi, and LoRa communication links at the millisecond level 24 / 7. It can capture various abnormal problems such as link interference, signal attenuation, equipment failure, and network congestion in real time, and realize the early detection of potential faults.

[0050] 2. Automatic Fault Identification Mechanism: Based on the built-in industrial communication fault identification model, the system can intelligently analyze the abnormal data detected, accurately distinguish different fault types such as instantaneous signal fluctuations, local link interference, hardware module failures, external network interruptions, and network congestion, and determine the scope and duration of the fault impact. This prevents invalid link switching caused by misjudgment or incorrect judgment, and ensures the stability of system operation.

[0051] 3. Seamless Automatic Switching Mechanism: When a communication link experiences an anomaly or failure, the system can automatically complete link priority switching, data link migration, and communication resource reconstruction without manual intervention, achieving zero-latency link switching and zero packet loss in data transmission. The three types of links serve as mutual backups and dynamic safety nets, ensuring that a single link failure will not affect the overall system communication function. This significantly enhances the redundancy and fault tolerance capabilities of the multi-link system, ultimately achieving an overall system communication reliability of 99.99%, meeting the high reliability and stability communication standards for core industrial applications.

[0052] III. The Layer 3 network layering functions and service adaptation instructions are as follows: 1. LoRa Low-Power Star Topology Module: As the underlying sensing and communication carrier of the network, the LoRa module adopts a star topology architecture. It possesses core characteristics such as large node capacity, long transmission distance, low power consumption, strong anti-interference capability, and flexible deployment, specifically designed for data collection from massive numbers of sensor terminals in industrial settings. In scenarios such as industrial power grids, smart factories, and virtual power plants, various sensor monitoring terminals are deployed on-site, including those for temperature, humidity, voltage, current, power, equipment operating status, and environmental security. These terminals are numerous, widely distributed, and scattered, with most relying on battery power, placing extremely high demands on communication power consumption and transmission stability. LoRa star topology allows a single gateway to connect to hundreds of sensor terminals, effectively covering a large area and long distance within the factory. Compared to traditional WiFi networking, it has stronger anti-electromagnetic interference capabilities, effectively avoiding electromagnetic signal interference from inverters, high-voltage equipment, motors, and other equipment in industrial settings, ensuring the stability of data collection in complex environments. Meanwhile, LoRa's ultra-low power consumption can significantly reduce the energy consumption of sensing terminal devices, extend the battery life of devices, reduce the frequency and cost of later maintenance, and continuously and stably complete the routine collection and uploading of status data, monitoring data, and alarm data of massive terminals at the underlying level, providing complete and accurate underlying data support for data analysis, strategy control, and status judgment of the upper-layer system.

[0053] 2. Dual-band 2.4G / 5G WiFi Module: The dual-band 2.4G / 5G WiFi module undertakes the core functions of mid-level localized communication, mainly responsible for high-speed data interaction and control strategy synchronization between floor-level centralized control equipment, sub-controllers, and various end-user devices within local areas such as buildings, single factory areas, and single zones. The 2.4G band has the advantages of strong wall penetration and wide coverage, adapting to complex indoor environments with multiple walls obstructing the view, ensuring communication connectivity for remote end-user devices. The 5G band features high transmission speed, low network latency, and sufficient bandwidth, supporting the rapid transmission of large amounts of local data. In actual operation, each sub-controller and end-user device within the zone needs to synchronize the control strategies, operating parameters, and control commands issued by the system in real time, while simultaneously uploading local device operating status and execution feedback data. This involves high data interaction frequency, high real-time requirements, and large data volume. Dual-band WiFi modules achieve millisecond-level data interaction and policy synchronization between devices within a local network through adaptive frequency band switching, balancing communication coverage and high-speed transmission. This ensures coordinated operation and joint control of devices within the area, avoiding operational anomalies caused by lagging local device commands and asynchronous policies, and building a high-speed transmission barrier for mid-level localized communication.

[0054] 3. Industrial 5G Full-Network Compatibility Module: As a top-level remote communication carrier, the industrial 5G full-network compatibility module leverages the technological advantages of 5G networks—high bandwidth, low latency, high reliability, and wide coverage—to specifically carry the core, high-real-time, and high-priority remote business data transmission of the system. It serves as the core communication channel connecting local devices with the cloud platform and remote dispatch center. Its main business transmission includes three types of core data: First, real-time synchronization data of digital twin 3D data. Industrial digital twin systems require real-time acquisition of 3D modeling data, dynamic image data, and equipment location data of field equipment, plant conditions, and power grid operation status. This type of data has a large capacity and extremely high real-time requirements. The 5G module, relying on its high bandwidth, can achieve uninterrupted, high-definition, and real-time remote transmission of 3D data, ensuring millisecond-level synchronization between the cloud-based digital twin model and the actual field conditions, enabling visualized and precise control. Second, millisecond-level dispatch instructions for virtual power plants. The core of virtual power plants lies in the precise coordinated dispatch of distributed energy resources, energy storage devices, and load equipment. Dispatch instructions have extremely high timeliness; even millisecond-level delays can lead to power grid load imbalance and dispatch failure. Industrial 5G modules enable ultra-low-speed remote transmission of dispatch commands, ensuring rapid issuance of commands from the cloud-based dispatch center and rapid feedback of equipment execution status, supporting efficient, accurate, and dynamic load regulation and resource coordination in virtual power plants. Third, power grid demand response interaction commands. During power grid peak-valley regulation, load optimization, and supply-demand balance control, continuous interaction of demand response commands, load adjustment data, and supply-demand status data is required between the power grid dispatch center and field terminals, energy storage, and load equipment. 5G modules can ensure long-distance, highly reliable, and latency-free transmission of such critical interactive data, accurately matching the dynamic control needs of the power grid and improving power grid operational stability and resource utilization.

[0055] IV. Three-Layer Network Cooperative Redundancy and Overall Advantages: The 5G, WiFi, and LoRa three-layer heterogeneous network built in this solution achieves a hierarchical division of labor, independent performance, mutual backup, and cooperative redundancy operation mode, completely breaking the capability limitations of a single communication network. From a hierarchical division of labor perspective, the bottom layer LoRa solves the problem of massive low-power terminal data collection, the middle layer WiFi solves the problem of high-speed local interaction, and the top layer 5G solves the problem of high-speed and reliable remote core service transmission. The three-layer network accurately matches the transmission needs of different levels and types of data, exhibiting extremely strong service adaptability. From a redundancy and coordination perspective, the three types of communication links can serve as backups for each other. When a single network experiences congestion, failure, or signal interruption, the system can automatically schedule other links to take over the data transmission task, achieving uninterrupted service operation. For low-priority massive data collection, high-priority local interaction data, and core-level remote scheduling data, the system can intelligently allocate the optimal transmission link, achieving optimized configuration of communication resources. Overall, this three-layer heterogeneous redundant communication network comprehensively covers the communication needs of scenarios such as industrial intelligence, power grid digitalization, virtual power plant control, and digital twin visualization. From bottom-level data acquisition and mid-level device linkage to top-level remote scheduling, the entire link ensures the real-time, stability, and reliability of data transmission. The 99.99% comprehensive communication reliability can effectively avoid risks such as equipment downtime, scheduling failure, data loss, and system crashes caused by communication failures, providing core communication support for the safe and stable operation of industrial intelligent systems, smart grids, and virtual power plants.

[0056] In one embodiment, the integrated flexible load control technology for lighting / air conditioning / fresh air is described in detail below: I. Modular Stepless Control Technology for Core Equipment: The integrated flexible load control unit for lighting, air conditioning, and fresh air systems used in this solution is equipped with dedicated high-precision control modules for each of the three main types of energy-consuming building equipment. This completely breaks through the limitations of traditional graded control and start-stop control of building equipment, achieving continuous, refined, and standardized control of the power of all types of equipment. The specific technical characteristics of each supporting module are as follows: For architectural lighting systems, a stepless dimming LED driver module is provided. This module abandons the traditional segmented dimming and on / off control mode of LED lamps, supporting continuous adjustment of the full range of illuminance power from 0% to 100%, without fixed brightness levels. Through high-precision analog signal output, the module accurately controls the operating current and power output of the LED lamps, with a minimum adjustment step of 1%, enabling smooth and gradual brightness control without brightness jumps or dimming dead zones. It can adapt to the fine-tuning needs of lighting loads in different scenarios and at different times.

[0057] For building central air conditioning systems, a variable frequency air conditioning continuous cooling and heating adjustment module is configured. Traditional air conditioners only support fixed cooling and heating levels and start-stop operation, resulting in strong intermittent and large fluctuations in load regulation. This module, relying on the core technology of variable frequency control, achieves continuous adjustment of the air conditioning unit's cooling and heating capacity from 0% to 100%, eliminating the need for frequent compressor start-stop to adapt to load demands. A minimum fine adjustment step of 1% allows for bidirectional micro-adjustment of the air conditioning operating power based on indoor ambient temperature and power grid dispatch requirements, completely eliminating power gaps and adjustment blind spots in the air conditioning system control process.

[0058] For building ventilation systems, a stepless speed-regulating airflow control module is incorporated. Breaking through the limitations of traditional multi-speed control of ventilation fans, it achieves smooth, stepless adjustment of fan speed and ventilation volume from 0% to 100%, eliminating the impact of speed switching and the problem of start-up / shutdown issues. The module can precisely fine-tune the ventilation supply based on indoor air quality, occupancy density, and load scheduling commands. Each 1% adjustment step ensures a stable increase or decrease in airflow, guaranteeing the continuity and stability of the ventilation system's operation.

[0059] The three core control modules uniformly adopt a 0-100% continuous analog control signal output standard, with fully standardized hardware interfaces, control protocols, and signal formats. This avoids industry pain points such as inconsistent control logic, signal incompatibility, and difficulty in scheduling adaptation among different types of equipment, laying a hardware foundation for unified management and control of the entire building load. All modules support millisecond-level response, with stable and deviation-free adjustment accuracy, and no graded jumps, start / stop dead zones, or operational lag issues throughout the process.

[0060] II. Construction and Scheduling Application of a Full-Domain Flexible Load Resource Pool: Based on the standardized and continuous control capabilities of three types of equipment, the system digitally integrates all flexibly adjustable lighting, air conditioning, and fresh air loads across the entire building area. Through a core technical architecture of unified modeling, unified aggregation, and unified management, it builds a standardized, quantifiable, schedulable, and tradable building flexible load resource pool, completely reconstructing the management and control of building energy consumption resources. At the unified modeling level, the system combines building functional zoning, equipment parameters, load characteristics, and operational constraints to build a dynamic mathematical model of the entire load area. This model accurately depicts the adjustable power range, response speed, and adjustment cost of each lighting, air conditioning, and fresh air device, enabling digital and visual management and control of scattered equipment loads. This solves the problems of traditional building load dispersion, data fragmentation, and disordered management. At the unified aggregation level, the system aggregates and integrates massive, dispersed, single-device flexible adjustable loads, consolidating fragmented small adjustable loads into a large-scale, continuous overall adjustable load resource. This eliminates the shortcomings of limited adjustment capabilities of single devices and their inability to participate in high-level grid scheduling, forming a load resource cluster with large-scale adjustment capabilities. At the unified management and control level, the system relies on standardized control protocols to achieve centralized scheduling, dynamic regulation, and real-time monitoring of flexible loads across the entire domain. It can accurately control the upward and downward adjustment of the overall load and output continuous and smooth adjustable power, completely changing the shortcomings of traditional rigid and fixed building loads and discrete regulation. The completed flexible load resource pool has extremely strong scenario adaptability and scheduling response capabilities, and can accurately match the millisecond-level fine-grained scheduling commands of the power grid and the demand response commands of virtual power plants. Whether it is the high-precision scheduling requirements of real-time frequency regulation, peak shaving, and voltage regulation of the power grid, or the market-oriented requirements of virtual power plant aggregation response, power trading, and load peak shaving and valley filling, all can be quickly adapted through the continuous power adjustment capability of the resource pool. This upgrades building loads from traditional passive energy-consuming rigid fixed loads to flexible and adjustable energy resources that actively participate in power grid scheduling and power market trading, significantly improving the energy utilization value of buildings and their adaptability to the power grid.

[0061] III. Operational Benefits and Equipment Protection of Stepless Continuous Regulation Mode: This technology employs a smooth, stepless continuous power regulation mode from 0% to 100% throughout the entire process. It fundamentally avoids many drawbacks of traditional building electromechanical equipment control modes, achieving significant improvements in three dimensions: grid stability, equipment operation and maintenance, and energy conservation. At the grid operation level, traditional lighting, air conditioning, and fresh air systems often use relay switching, gear shifting, and start-stop control. Sudden changes in equipment operating states can generate instantaneous large current surges, causing grid voltage fluctuations and frequency disturbances. Simultaneously, it easily triggers excessive grid harmonics, affecting the power supply quality and operational stability of the regional grid. The stepless smooth continuous regulation mode, however, achieves gradual adjustment of equipment power, eliminating sudden changes in operating states and sudden increases or decreases in instantaneous current. This completely eliminates grid disturbances during equipment regulation, effectively purifying the grid operating environment and improving the stability and reliability of the power distribution system. At the equipment operation and maintenance level, frequent start-stop operations, gear shifting, and repeated compressor start-stop operations of traditional equipment cause severe wear and tear on the core mechanical structure and electrical components, significantly shortening equipment lifespan and increasing equipment failure rates. This solution's continuous stepless control mode ensures that core equipment such as LED lighting fixtures, air conditioning compressors, and fresh air fans operate in a stable and flexibly adjustable state, free from mechanical and electrical shocks. This significantly reduces the wear rate of core components, effectively extending the service life of various core equipment and substantially lowering daily maintenance and replacement costs. In terms of energy consumption management, the frequent start-stop operation of traditional central air conditioning units and fresh air systems generates substantial start-stop losses. The energy consumption at startup is far higher than the steady-state operation energy consumption, leading to serious energy waste from repeated start-stop cycles. Furthermore, the coarse control mode of start-stop and speed-up switching cannot accurately match actual load demand, easily resulting in over-supply and ineffective energy consumption. This technology, through 1% high-precision continuous fine-tuning, achieves precise matching between equipment operating power and actual energy demand, eliminating ineffective energy consumption and start-stop energy losses, significantly reducing the overall building energy consumption level, and effectively reducing the failure probability of major energy-consuming equipment such as central air conditioning and fresh air systems, thus improving the stability and economy of the building's electromechanical system.

[0062] IV. Core Technological Innovations: This integrated flexible load control technology is based on modular, high-precision, stepless control, with full-domain load aggregation modeling as its core, and aims to adapt the power grid and virtual power plant to meet their demands. It breaks through the bottlenecks of traditional building energy consumption equipment, characterized by "segmented control, rigid operation, and isolated management." Through standardized, continuous, and refined load control, it achieves energy-saving, stable, and long-term operation of building energy consumption, while also completing the resource-based transformation of building loads. This provides core technological support for the collaborative operation of intelligent buildings, virtual power plants, and new power systems.

[0063] In one embodiment, the full-domain multi-dimensional millimeter-wave AI sensing terminal technology is described as follows: I. Terminal Hardware Configuration and Deployment: The all-domain multi-dimensional sensing terminal adopts an integrated hardware design, with the entire unit uniformly equipped with multiple types of high-precision sensing devices to comprehensively cover the full-dimensional needs of human body status perception and environmental parameter monitoring. The core hardware modules include a 60GHz frequency-modulated continuous wave high-precision millimeter-wave radar, a wide-range illumination acquisition probe, a high-precision temperature and humidity sensor, and a four-in-one air quality detection module for carbon dioxide / PM2.5 / TVOC / formaldehyde. The multi-module integrated design avoids the industry pain points of traditional monitoring equipment, such as single monitoring, cumbersome multi-device splicing deployment, and large data synchronization errors. It enables a single device to complete the synchronous acquisition of data on all elements such as human body status, indoor environmental quality, illumination, temperature, and humidity. In terms of power supply and communication solutions, the terminal adopts an integrated solution of low-power lithium battery power supply and LoRa long-range wireless communication. Compared with the high-voltage wiring and wired communication mode of traditional building intelligent equipment, this solution has strong scene adaptability, especially suitable for the renovation of existing old buildings. In actual deployment, there is no need to cut grooves in the building walls for wiring, no need to lay high-voltage control lines, and no need to modify the building's original equipment lines and electrical systems. This completely avoids the problems of large construction volume, long construction period, damage to decoration, high renovation cost, and impact on normal building use in traditional intelligent renovation. It can achieve renovation-free, zero-destruction, and rapid deployment in old buildings and existing buildings, significantly reducing the construction cost and time cost of building intelligent and energy-saving renovation.

[0064] II. Terminal-Localized AI Algorithm Architecture and Core Functions: The core advantage of this terminal is its fully localized AI computing power deployment. The device incorporates a lightweight CNN convolutional neural network and a 72-hour sliding window LSTM time-series prediction dual AI sub-modules. Relying on the dual-algorithm collaborative mechanism, it achieves accurate human pose recognition, personnel status statistics, and environmental and operational condition time-series prediction, completely solving problems such as control lag, network dependence, and data latency inherent in traditional cloud computing devices. This includes: 1. Lightweight CNN Convolutional Neural Network (Human Perception and Recognition Module): The lightweight CNN convolutional neural network on the terminal has undergone scenario-specific optimization. It has been trained and iterated for high-frequency human postures within buildings, accurately identifying four core postures: standing while working, sitting while working, lying down while resting, and when the space is empty. The recognition accuracy is high with a low false positive rate. Based on real-time posture recognition results, the algorithm can simultaneously statistically output three core operational parameters: the real-time number of people in each spatial area, personnel distribution density, and the duration of each person's single / cumulative stay. This provides accurate personnel status data support for building air conditioning, lighting, ventilation, air quality control, and energy consumption management. This lightweight model is optimized for the terminal's local computing power, consuming minimal device resources and operating at high speed. It can achieve millisecond-level data acquisition and recognition output, ensuring the real-time and stable perception of human status within the space and meeting the refined management needs of various indoor scenarios.

[0065] 2.72-Hour Sliding Window LSTM Time Series Prediction Submodule: To address the industry shortcomings of traditional building equipment's "real-time monitoring and lagging control," the terminal incorporates a 72-hour sliding window LSTM long-term time series prediction model. Based on continuously collected long-term historical operating data, the model can predict multi-dimensional operating conditions and trends. The core prediction scenarios include three dimensions: first, prediction of natural light intensity fluctuations, which can anticipate dynamic changes in indoor light intensity caused by day-night cycles, weather changes, and cloud cover; second, prediction of indoor pollutant concentration changes, which, based on historical data from the air quality four-in-one module, accurately predicts the accumulation and dissipation patterns of pollutants such as carbon dioxide, PM2.5, TVOC, and formaldehyde; and third, prediction of photovoltaic power generation time series fluctuations, adapting to photovoltaic-linked energy-saving scenarios and predicting the dynamic changes in new energy power supply. By leveraging the LSTM timing prediction capability, the system can achieve advanced pre-control of indoor environment, equipment operating conditions, and new energy supply, fundamentally eliminating the lag problem in the entire process of data collection, cloud analysis, and equipment execution in traditional intelligent control systems, and significantly improving the comfort of building environment control and the accuracy of energy consumption management.

[0066] III. Offline Independent Operation Capability (No Network Adaptation): All AI sensing, human body recognition, time-series prediction, data processing, and logical judgment functions of this terminal are completed locally on the device, without relying on cloud server computing power or uploading data to the cloud in real time. This localized computing architecture gives the device extremely strong environmental adaptability. In extreme conditions such as network isolation, offline operation, and network outages, all sensing, recognition, prediction, and control linkage functions of the terminal are completely unaffected by network status and can operate continuously and stably. This completely avoids the problems of paralysis and functional failure of traditional cloud-based intelligent devices when the network is down, greatly improving the stability, reliability, and fault tolerance of the building intelligent control system. After the network is restored, the device can automatically synchronize local cached data to the cloud, ensuring that offline data is not lost and operation is uninterrupted.

[0067] IV. Refined Cloud-Based Backend Configuration and Multi-Scenario Adaptability: The terminal is equipped with a dedicated cloud-based management backend, featuring customizable parameter configuration, batch deployment, and differentiated strategy settings, adaptable to the personalized management needs of various building types. The backend supports custom editing and batch one-click deployment of core indicators such as unattended latency, sensing sensitivity threshold, human body recognition threshold, indoor environmental comfort range, and control logic parameters, eliminating the need for on-site equipment debugging and significantly improving operational efficiency. Addressing the differences in personnel activity patterns across different functional areas within a building, the cloud-based backend enables scenario-based differentiated departure delay strategy configuration, adapting to different functional scenarios such as hotel rooms, independent office areas, open office areas, meeting rooms, corridors, and public service areas. For example, short-delay control strategies can be set for meeting rooms with short-term concentrated personnel and rapid departures; flexible delay strategies can be set for office areas with long-term stays and intermittent departures; and personalized control strategies can be set for the private lifestyles of hotel rooms, accurately matching the activity characteristics of different spaces while balancing environmental comfort and energy efficiency. Leveraging this flexible configuration capability, the equipment can be fully adapted to various commercial and public building energy consumption formats such as office buildings, hotels, campuses, shopping malls, hospitals, and industrial parks.

[0068] V. Low-Power Intelligent O&M Mechanism: All LoRa wireless sensor terminals have a built-in intelligent low-power sleep and adaptive wake-up mechanism, achieving an optimal balance between sensing accuracy and device battery life. The system dynamically adjusts the device's data acquisition and operation frequency based on CNN human perception results: When personnel are present, the device automatically switches to a high-frequency acquisition mode, collecting personnel status and environmental data at high density to ensure control accuracy and real-time performance; when no one is present, the device automatically enters a low-frequency sleep operation state, reducing data acquisition frequency, computational power consumption, and communication power consumption. This intelligent power control mechanism requires no manual intervention, automatically and dynamically adapting to different scenarios. While ensuring the effectiveness of spatial perception, environmental monitoring, and intelligent control throughout the process, it significantly reduces the overall power consumption of the device, significantly extends the lithium battery life, effectively reduces the frequency of device charging, replacement, and fault repair, and significantly reduces the later-stage maintenance costs and manpower investment of building intelligence systems, adapting to the maintenance needs of large-scale, large-area building equipment deployment.

[0069] In one embodiment, the digital twin virtual-real mapping unit is the core functional unit of the smart building microgrid energy-saving and low-carbon management system. It deeply integrates building energy consumption mechanism models, artificial intelligence big data algorithms, real-time data transmission technology, and 3D visualization simulation technology, comprehensively covering all business scenarios of the building microgrid system, including pre-commissioning debugging and optimization, real-time monitoring during operation, fault early warning during operation, and full-cycle iterative evolution. This unit integrates four core sub-modules: a pre-commissioning offline full-condition simulation sub-module, a post-commissioning millisecond-level online virtual-real synchronization sub-module, a full-equipment, full-link hierarchical fault early warning sub-module, and an operational data visualization analysis sub-module. These modules work collaboratively and complement each other, thoroughly solving industry pain points of traditional building microgrid management systems such as cumbersome parameter debugging, opaque operating status, delayed fault diagnosis, rigid control strategies, and lack of autonomous optimization capabilities. This comprehensively improves the energy efficiency, stability, low-carbon operation, and intelligence level of the building microgrid, as detailed below: I. Offline Full-Condition Simulation Submodule (Core Function Before Commissioning): The offline full-condition simulation submodule is mainly used for scheme optimization, parameter debugging, and strategy finalization before the formal commissioning of the building microgrid system. It is the core support for achieving optimal system control, avoiding manual debugging errors, and shortening the project implementation cycle. This module abandons the outdated mode of traditional building systems that relies on manual experience to set parameters and repeatedly try and debug. It builds a dedicated simulation engine based on building energy consumption mechanisms and AI big data fusion algorithms, supports multi-dimensional and refined custom parameter input, and can realize hourly, full-condition, fully automatic dynamic simulation for 8760 hours a year. It accurately outputs a complete set of quantitative operating data and optimal control parameters, providing standardized, digital, and optimized parameter support for the formal commissioning of the system, including: 1. Multi-dimensional and refined simulation parameter input system: The module supports the custom input and configuration of parameters for the entire building body, equipment, personnel, environment, and economy, covering six core categories of parameters: building envelope, new energy equipment, electromechanical equipment, personnel work and rest, meteorological environment, and electricity pricing mechanism. This achieves a 1:1 replication of the simulation scene with the actual operating scene, ensuring the authenticity and accuracy of the simulation results. Specific parameter categories are as follows: Building envelope parameters include building unit structure, window-to-wall ratio, overall heat transfer coefficient of the envelope, shading coefficient of external windows, parameters of wall insulation material and insulation thickness, and parameters of roof and ground insulation performance. These parameters accurately match the heat dissipation, heat storage and heat insulation characteristics of the building itself and are the basic core parameters for building energy consumption simulation.

[0070] New energy and energy storage equipment parameters: covering core installation and operation parameters of photovoltaic energy storage, such as photovoltaic module installed capacity, photovoltaic array layout, module conversion efficiency, rated capacity of energy storage system, maximum charging and discharging power of energy storage, upper and lower limits of energy storage SOC, and battery degradation characteristics.

[0071] Mechanical and electrical equipment parameters: Enter the rated power, rated operating parameters, start-stop response characteristics, load adjustment range, equipment energy efficiency ratio and other basic operating parameters of all building mechanical and electrical equipment such as air conditioners, fresh air systems, lighting, water supply and drainage, and elevators.

[0072] Personnel work and rest time sequence parameters: Supports the input of personnel work and rest time sequence data by time period, floor and area throughout the year, accurately simulating the changes in heating and cooling load, lighting load and equipment power load generated by personnel activities in the building at different times.

[0073] Meteorological and environmental time series parameters: Import hourly meteorological data for the project location throughout the year, including core meteorological parameters such as hourly solar radiation intensity, ambient temperature, ambient humidity, wind speed, and radiation, and match the photovoltaic power generation output characteristics with the building's cooling and heating load variation patterns.

[0074] Electricity price time series parameters: Supports the input of hourly electricity price data for industrial and commercial peak-valley flat time-of-use electricity price, peak electricity price, demand electricity price, etc. throughout the year, providing data basis for subsequent electricity billing and generation of economically optimal control strategies.

[0075] 2. Full-time simulation computing capability: The simulation engine relies on a dual-model fusion algorithm. On the one hand, it is based on mature building thermodynamics, fluid mechanics, and power load mechanism models to ensure the physical reality of energy consumption and output simulation. On the other hand, it combines AI algorithms trained with massive amounts of building microgrid operation big data to correct the errors and defects of traditional mechanism models and improve the simulation accuracy under complex operating conditions. The system can automatically complete 8760 hours of uninterrupted full-condition simulation throughout the year, completely replicating the operating conditions of the building microgrid under different seasons, time periods, weather conditions, and personnel loads throughout the year, without manual intervention, breakpoints, or omissions.

[0076] 3. Simulation Output Results and Applications: After full-time simulation calculations, the module can output a complete set of quantitative and implementable core data, including the building's total annual electricity consumption, statistical data on individual energy consumption items, annual photovoltaic green electricity generation and absorption rate, peak load extremes and peak occurrence times of the power grid, monthly and annual electricity costs, hourly carbon emission reductions, optimal control thresholds for all equipment, and optimal strategy parameters for source-storage-load coordinated control. Simultaneously, the module features a one-click batch parameter distribution function. Maintenance personnel no longer need to manually organize, debug, or input parameters; they can directly distribute the optimal control parameters obtained from the simulation to the global edge controller and end-point intelligent controller in batches. This completely eliminates the tedious process of manual trial and error, experience-based value assignment, and repeated rectification in traditional system debugging, significantly reducing manual debugging costs and shortening the traditional system debugging cycle of several weeks or even months to several days, greatly improving project implementation efficiency and system debugging accuracy.

[0077] II. Millisecond-Level Online Virtual-Real Synchronization Submodule (Core of Real-Time Management During Operation): The online virtual-real synchronization submodule is the core carrier for realizing real-time mapping and dynamic linkage between the physical building microgrid and the digital twin model. It undertakes the core functions of real-time data acquisition, synchronous updates, 3D visualization, and multi-dimensional data comparison and analysis. The module adopts high-frequency data transmission and synchronization technology to achieve real-time alignment of physical equipment, energy consumption data, source and storage status, environmental parameters, carbon emission reduction data, and the 3D twin virtual model. This completely solves the problems of lagging data updates, unclear status displays, and invisible energy flow in traditional building systems, enabling the building microgrid's operating status to be "digitalized, visualized, and transparent," including: 1. High-Frequency Global Data Synchronization Mechanism: The module is configured with an ultra-high frequency (100 milliseconds / time) full-data synchronization update mechanism, covering all core data dimensions of the building microgrid. This includes real-time photovoltaic power output data, remaining SOC of the energy storage system, real-time charging and discharging power of the energy storage system, real-time operating power of electromechanical equipment in each zone, indoor and outdoor temperature and humidity, air quality and other environmental parameters, hourly carbon emission reductions, grid interaction power, and other comprehensive operational data. Data collected by all physical terminals can be uploaded to the twin platform in real time and synchronously updated to the 3D virtual model, achieving millisecond-level virtual-physical consistency and state synchronization between the physical entity and the digital model, with no data delay and no state deviation.

[0078] 2. Four-Level Layered 3D Visualization: The 3D twin interface adopts a four-level layered display architecture: building, floor, zone, and individual room. This enables refined and hierarchical visualization and control of the entire building's status. Managers can switch between different levels of view as needed to accurately grasp the operational status of the entire area or specific regions: the building level displays the overall energy consumption, total power generation, and total carbon emission reduction of the entire building; the floor level displays the load and energy consumption data of a single floor; the zone level enables specialized monitoring of functional zones (offices, computer rooms, public areas, etc.); and the individual room level allows for precise viewing of detailed data on equipment operation, environment, and energy consumption in a single room. The interface dynamically and intuitively presents the energy flow path of the entire building, the effect of load peak shaving and valley filling, the proportion of local photovoltaic green electricity consumption, the real-time operating status of various electromechanical equipment, and the charging and discharging switching process of the energy storage system, transforming abstract energy consumption, electricity, and carbon data into intuitive 3D dynamic visuals.

[0079] 3. Multi-dimensional data comparison and analysis function: The module has built-in data statistical analysis algorithms, supporting year-on-year, month-on-month, and time-period comparison analysis of energy consumption data, photovoltaic power generation data, and carbon emission reduction data. It can automatically generate daily, weekly, monthly, and annual data comparison reports and trend curves. Managers can view data change patterns at any time through remote terminals, accurately assess the effect of load regulation, the level of new energy consumption, and energy-saving and low-carbon benefits, providing intuitive data support for daily operation and maintenance scheduling, load optimization, and energy-saving strategy adjustment, and realizing remote unattended refined management and control.

[0080] III. Full-Equipment, Full-Link Hierarchical Fault Early Warning Submodule (Core of Security Operation and Maintenance): This submodule focuses on the secure and stable operation of all equipment and links within the building microgrid. Leveraging the real-time data acquisition capabilities of sensors, communication equipment, and controllers across the entire domain, it enables real-time monitoring, intelligent identification, hierarchical alarms, location tracing, and solution empowerment for all types of fault signals. It constructs a full-link operation and maintenance support system encompassing "pre-event early warning, in-event handling, and post-event tracing," significantly reducing the pressure of manual inspections and operation and maintenance costs, and mitigating issues such as energy waste, system downtime, and safety hazards caused by equipment failures. This includes: 1. Comprehensive Fault Signal Acquisition: The module collects fault and abnormal signals from all scenarios, all devices, and all links of the building microgrid in real time, covering five major categories of fault types: hardware, communication, operation, power grid, and energy storage. Specifically, it includes all common and special fault conditions such as sensor offline failure, communication link interruption, abnormal operation of electromechanical equipment, exceeding of operating parameters, energy storage battery failure, abnormal grid voltage / frequency, and abnormal load changes, achieving fault monitoring without blind spots or omissions.

[0081] 2. Three-tiered alarm push mechanism: Based on the scope of impact, severity, and urgency of a fault, the system classifies all faults into three levels: general, important, and urgent, enabling differentiated alarm push and priority allocation for handling. General faults are minor anomalies that do not affect the overall system operation and can be handled through routine planning; important faults are anomalies in localized equipment that will slightly affect energy consumption and operating efficiency, requiring timely handling; urgent faults are severe anomalies in core equipment, the power grid, or the energy storage system, posing safety hazards or the risk of system shutdown, requiring immediate emergency handling. The system automatically pushes corresponding alarm information according to the fault level, avoiding massive invalid alarms interfering with operation and maintenance work.

[0082] 3. Intelligent Fault-Enabling O&M Capabilities: Compared to traditional systems that only provide basic fault information, this module possesses end-to-end intelligent O&M capabilities. Alarm information is accompanied by four core components: precise fault location, intelligent fault cause analysis, standardized handling plans, and personalized O&M suggestions. It can accurately pinpoint the location of faulty equipment and fault link nodes, trace the cause of the fault through big data algorithms, generate standardized handling procedures based on industry standards and equipment operating standards, and provide personalized O&M suggestions according to on-site conditions. This effectively solves the problems of traditional O&M relying on manual experience for troubleshooting, slow fault tracing, and non-standard handling, significantly improving fault handling efficiency and reducing equipment failure rates and O&M labor costs.

[0083] IV. Platform's Autonomous Iterative Evolution Capability (Core of Full Lifecycle Optimization): The digital twin virtual-real mapping unit possesses industry-differentiated autonomous learning and continuous evolution capabilities throughout the entire lifecycle. This completely overcomes the technical shortcomings of traditional building automation systems, such as fixed logic, fixed parameters, and the inability to autonomously optimize after deployment. It achieves the evolutionary characteristic that the longer the system operates, the higher the control accuracy and the more prominent the overall benefits. During daily operation, the system automatically aggregates comprehensive big data, including building-wide equipment operation data, energy consumption data, source / storage output data, environmental data, fault data, and control records, and uploads it in real-time to the cloud-based DRL deep reinforcement learning training platform. Relying on artificial intelligence algorithms, the platform performs deep learning, model training, and iterative optimization on the massive operational data, continuously correcting parameter deviations between the simulation model and the control algorithm, and optimizing the source / storage / load collaborative control logic, fault identification model, and energy consumption optimization strategy. After completing model iteration and parameter upgrades, the system encrypts the new optimal algorithm parameters and securely distributes them to the NPU computing chips of each edge controller, achieving seamless update and upgrade of terminal device algorithm parameters without downtime, manual intervention, or impact on normal system operation. Through continuous offline training, online updates, and autonomous optimization, the system can continuously adapt to dynamic scenarios such as seasonal changes in building operating conditions, equipment aging characteristics, changes in personnel work and rest, and adjustments in electricity price policies. This continuously improves the accuracy of energy-saving control, the efficiency of photovoltaic green electricity consumption, and carbon emission reduction benefits, achieving intelligent evolution and upgrading of the building microgrid throughout its entire life cycle.

[0084] V. Core Advantages of Unit Overall Application: In summary, the digital twin virtual-real mapping unit, through the collaborative empowerment of four sub-modules and its autonomous iterative evolution capabilities, constructs a closed-loop management and control system encompassing "pre-commissioning simulation optimization, real-time control during commissioning, intelligent operation and maintenance during operation, and continuous evolution throughout the entire lifecycle." Compared to traditional building microgrid management and control systems, it possesses five core advantages: refined parameters, full-time simulation, high real-time synchronization, intelligent fault diagnosis, and system evolution. This effectively reduces system debugging cycles and operation and maintenance costs, improves building energy efficiency, maximizes the proportion of photovoltaic green electricity consumption, and precisely achieves low-carbon emission reduction in buildings, comprehensively empowering the energy-saving, intelligent, low-carbon, and digital upgrades of smart buildings.

[0085] In one embodiment, the virtual power plant aggregation and dispatch unit serves as the core hub connecting the building's integrated energy system with the upper-level electricity market, grid dispatch, and regional energy network, employing a standardized, modular, and scalable architecture. The core design goal of the unit is to break down the barriers of isolated operation of building flexible loads, enabling the transformation of building energy loads from single energy consumption terminals into dispatchable, tradable, and value-added electricity assets. By reserving multiple standardized interface types, a comprehensive interconnection channel is established, simultaneously completing the standardized aggregation, quantitative modeling, and unified dispatch of building flexible loads, fully supporting diversified electricity market transactions and grid collaborative dispatch operations, as detailed below: I. Standardized Interface Configuration and Access Capabilities for Multiple Scenarios: To adapt to the industry needs of standardized operation of the power market, refined grid dispatching, and regional energy collaborative interconnection, this aggregated dispatching unit is pre-configured with three types of standardized dedicated interface interfaces. These interfaces fully comply with the general technical specifications of the domestic power market, grid dispatching, and regional energy internet, supporting seamless adaptation to mainstream industry platforms. System access can be completed without secondary development or modification, possessing strong compatibility, versatility, and scalability, including: 1. Standardized Electricity Market Interface API: This interface strictly adheres to the official data exchange protocols of the electricity spot market, ancillary services market, and demand response market, supporting bidirectional real-time data transmission. Uplink data can upload core transaction data such as building adjustable load resource parameters, adjustable capacity, real-time available capacity, and load adjustment costs; downlink data can receive market-based information such as electricity market transaction announcements, clearing results, transaction prices, performance instructions, and settlement basis. Through this interface, direct access to provincial and municipal power grid market-based dispatch and trading platforms is possible, enabling building loads to participate autonomously and compliantly in various electricity market transactions, thus opening up channels for the market-based monetization of load assets.

[0086] 2. Power Grid Demand Response Interface: The interface is compatible with the interaction standards of power grid companies for demand response, orderly power consumption, peak-valley regulation, and emergency supply guarantee, supporting real-time linkage with the power grid dispatching side's demand response management platform. It can accurately receive information such as peak-valley regulation signals, load invitation instructions, emergency power rationing notices, and demand response compensation policies issued by the power grid, while simultaneously providing real-time feedback on building load response status, regulation completion rate, and resource availability, ensuring the efficient implementation of power grid dispatching instructions and achieving dynamic and coordinated matching between building load and power grid operating status.

[0087] 3. Regional Energy Internet Interface: Designed for regional integrated energy collaborative operation scenarios, it features a standardized interface for the regional energy internet, enabling seamless integration with upper-level systems such as regional virtual power plant platforms, urban integrated energy service platforms, and park energy management platforms. It supports the aggregation, sharing, collaborative scheduling, and mutual support of energy resources from multiple buildings and of multiple types within the region. It can undertake regional energy coordination and scheduling tasks, participate in regional load balancing, energy optimization, peak shaving, and valley filling, and contribute to the efficient and low-carbon operation of the regional energy system.

[0088] II. Standardized Resource Pool Integration System for Building Flexible Loads: This unit focuses on controllable flexible load resources across the entire building area, abandoning the traditional model of independent energy consumption management, fragmented data, and unschedulable loads. It specifically integrates three core flexible loads: the overall lighting system, central air conditioning system, and fresh air system. Through data collection, model building, standardized conversion, and unified classification, a standardized, visualized, and quantifiable distributed adjustable resource pool is constructed. During resource integration, the unit uses edge acquisition devices to collect real-time information on various load operating parameters, power data, start / stop status, adjustment thresholds, and environmental constraints. Combined with building energy consumption scenarios, human comfort constraints, and equipment operation safety standards, a mathematical model for load flexible adjustment is established. This model standardizes and normalizes various scattered and heterogeneous building loads, eliminating operational differences and data barriers between different devices and systems. Simultaneously, the unit achieves three-dimensional precise quantification of the resource pool's adjustable capacity, providing accurate data support for subsequent market transactions and grid dispatch. Specific quantification indicators are as follows: 1. Unified and quantitative adjustable capacity: Accurately calculate the maximum adjustable load capacity of lighting, air conditioning and fresh air systems under different time periods and operating conditions, distinguish between normal adjustable capacity, emergency adjustable capacity and extreme adjustable capacity, form a time-sharing and scenario-based adjustable capacity ledger, and clarify the upper and lower limits of the power that the resource pool can schedule.

[0089] 2. Unified Quantitative Adjustable Rate: Based on the adjustment response speed of various equipment, the load rise and fall rate parameters are quantified, and the adjustment capabilities of the resource pool at the millisecond, second, and minute levels are accurately identified to adapt to the scheduling needs of different response times, such as power grid frequency regulation, short-term peak shaving, and long-term load control.

[0090] 3. Unified and Quantified Continuously Adjustable Duration: Combining variables such as building energy consumption patterns, equipment operating time constraints, ambient temperature, and personnel presence status, the sustainable adjustment duration under different adjustment ranges is calculated, and short-term, medium-term, and long-term adjustable cycles are defined to ensure the stability and reliability of scheduling and transaction fulfillment.

[0091] III. Support for Multiple Types of Market-Based Power Trading and Dispatch Services: Relying on a standardized interface system and a quantitatively adjustable resource pool, the virtual power plant aggregation dispatch unit possesses comprehensive power market participation capabilities. It can autonomously adapt to various market-based and routine dispatch services of the power grid, maximizing the value of building loads. Specific supported business scenarios are as follows: 1. Peak-Valley Shaving Service: Addressing the load fluctuations of the power grid during daytime peak hours and nighttime off-peak hours, this service reduces building power consumption by lowering flexible loads such as lighting, air conditioning, and ventilation during peak hours, thus shaving off-peak loads. Conversely, it moderately increases flexible load consumption during off-peak hours to fill valleys. This effectively mitigates the peak-valley load difference, alleviates regional power grid supply pressure, and improves power grid operational stability.

[0092] 2. Grid Auxiliary Frequency Regulation Service: Relying on the precise and rapid load adjustment rate of the resource pool, it responds to the grid frequency fluctuation adjustment needs. When the grid frequency is too high or too low, it fine-tunes the building's flexible load power in real time, participates in the primary and secondary frequency regulation of the grid, provides high-quality auxiliary services to the grid, and obtains revenue from the auxiliary service market.

[0093] 3. Electricity Spot Trading: Leveraging the electricity market API interface, real-time day-ahead, intraday, and real-time electricity spot price signals are obtained. Combined with the adjustable characteristics of building load and adjustment costs, trading strategies are autonomously optimized. Adjustable load is sold off during peak electricity price periods to reduce electricity consumption; electricity consumption is increased and energy is stockpiled during off-peak electricity price periods. Market arbitrage is achieved through price differences, reducing the overall energy cost of the building.

[0094] 4. Orderly Power Dispatch Execution: During periods of tight power supply and demand and power shortages in the power grid, accurately receive instructions from the power grid for orderly power consumption, staggered power consumption, and peak avoidance power consumption, orderly reduce building flexible loads in accordance with dispatch requirements, strictly implement power grid supply guarantee dispatch requirements, ensure the safe and stable operation of the regional power system, and at the same time avoid the risk of illegal power consumption assessment.

[0095] 5. Demand Response Compensation Transactions: Proactively respond to market-based demand response invitations and emergency demand response tasks from the power grid, complete load regulation according to response standards, accurately record data such as response duration, regulation capacity, and performance quality, automatically connect to the demand response settlement system, and legally obtain special compensation revenue from power grid demand response to realize the added value of load resources.

[0096] IV. Core Implementation Value: The implementation of this virtual power plant aggregation and dispatch unit completely changes the traditional status quo of building energy consumption being "passive consumption, valueless, and uncontrollable," fully realizing the resource-based, asset-based, and market-oriented transformation of building loads. At the resource level, it transforms scattered building lighting, air conditioning, and fresh air loads into quantifiable, dispatchable, and usable high-quality distributed power resources, enriching the grid's adjustable resource reserves. At the asset level, it endows building flexible loads with the asset attributes of continuous revenue generation, making them an important component of building comprehensive energy assets. At the market level, it opens up the interconnection between building loads and the electricity market, allowing building energy resources to deeply participate in electricity market transactions and grid dispatch. While ensuring building energy comfort and safety, it effectively reduces building energy costs and creates market-oriented added value, while simultaneously assisting the grid in peak shaving and valley filling, renewable energy consumption, and the safe and stable operation of the grid, achieving a win-win situation for users, the grid, and the regional energy system.

[0097] In one embodiment, the technology of the blockchain-based full lifecycle carbon data trusted storage unit is described as follows: I. Overall Unit Architecture and Core Configuration: The blockchain-based full-lifecycle carbon data trusted storage unit is the core underlying data support unit of the building low-carbon smart management system. It focuses on the core capabilities of accurate full-lifecycle carbon data accounting, trusted storage and traceability, and standardized carbon asset value transformation. Unlike traditional single-operation carbon accounting modules, it achieves closed-loop management of building carbon data from collection, accounting, storage, verification to value realization. This unit incorporates dedicated encrypted nodes on a consortium blockchain, distinct from public blockchain nodes and closed private blockchain nodes, adapting to the compliance, privacy, and collaboration needs of green and low-carbon projects in the construction industry. Consortium blockchain nodes are only open to authorized entities such as project builders, operators, third-party verification agencies, regulatory authorities, and financial institutions. All node access requires identity authentication, permission grading, and key registration to prevent unauthorized access, data theft, and malicious tampering, providing compliant underlying support for subsequent carbon data storage, verification, and reporting. Meanwhile, the unit is deeply integrated with an AI-powered intelligent accounting subroutine for building carbon footprint throughout its entire lifecycle. Leveraging deep learning algorithms, big data fitting models, and industry standards for building carbon accounting, it breaks through the limitations of traditional static coefficient calculations and periodic statistics, enabling dynamic, refined, and intelligent calculation of building carbon emissions. The subroutine includes a built-in algorithm library specific to the building industry, adapting to the accounting needs of various scenarios, including public buildings, residential buildings, and industrial supporting buildings. It can automatically adapt to carbon accounting rules for different building types, energy consumption structures, and equipment configurations.

[0098] II. Multi-dimensional Dynamic Carbon Accounting Coefficient Integration System: To address the industry pain points of traditional building carbon accounting coefficients being singular, static, and prone to large deviations in calculation results, this system innovatively integrates four core dynamic correction coefficients. Combined with real-time building operation data, it achieves hourly, item-by-item, millisecond-level accurate carbon emission accounting, completely breaking away from the traditional monthly and quarterly extensive statistical model, including: 1. Regional Power Grid Time-of-Use Carbon Emission Factor: Abandoning the nationally unified fixed power grid carbon emission factor, this system connects to real-time dispatch data from provincial and municipal regional power grids. Based on dynamic parameters such as peak-valley and off-peak periods, the proportion of renewable energy connected to the grid, and the proportion of thermal power output, the regional time-of-use carbon emission factor is updated in real time. A high coefficient is used during peak grid periods and periods with a high proportion of thermal power, while a low coefficient is used during peak periods of renewable energy connections such as wind, solar, and hydropower. This accurately matches the carbon emission intensity of real-time building electricity consumption, ensuring that the electricity consumption carbon emission calculation closely reflects the actual carbon emission status of the power grid.

[0099] 2. Photovoltaic Zero-Carbon Offset Coefficient: A dynamic zero-carbon offset calculation mechanism is established for zero-carbon power generation equipment such as building-distributed photovoltaics (PV), rooftop PV, and curtain wall PV. The photovoltaic zero-carbon offset coefficient is dynamically calculated based on parameters such as real-time PV power generation, self-consumption, surplus electricity fed into the grid, solar irradiance, and equipment efficiency degradation. The carbon emissions from self-consumed PV power are fully offset, and grid-connected electricity is included in the regional carbon reduction ledger. This accurately calculates the offsetting effect of the PV system on building operation carbon emissions, avoiding the calculation distortion caused by fixed coefficient offsets.

[0100] 3. Carbon Conversion Factors for Energy Storage Charging and Discharging: Adapted to the carbon accounting needs of new energy storage systems such as building-integrated energy storage power stations, lithium battery energy storage, and flywheel energy storage, differentiated conversion factors are set for energy storage charging, discharging, and static states. During the charging period, the implicit carbon emissions of energy storage are calculated based on the real-time carbon factor of the power grid. During the discharging period, the carbon emission reduction is calculated based on the purpose of discharging (building self-use, power grid peak shaving). At the same time, dynamic correction factors are considered for parameters such as energy storage cycle charging and discharging efficiency, self-loss rate, and service life decay, solving the industry problem of double counting and omission of carbon emissions from energy storage in traditional accounting.

[0101] 4. Equipment Loss Carbon Correction Coefficient: To address the hidden carbon emissions from long-term aging, energy efficiency degradation, and maintenance losses of building electromechanical equipment, a life-cycle loss correction system is established. Based on operational data such as light decay of lighting fixtures, energy efficiency degradation of air conditioning compressors, losses of fresh air fans, energy consumption losses of pipelines, and losses from equipment start-up and shutdown due to malfunctions, combined with parameters such as equipment service life, maintenance frequency, and ambient temperature and humidity, a dynamic equipment loss carbon correction coefficient is generated. This coefficient accurately corrects the basic carbon emission accounting results, covering all dimensions of hidden carbon emissions from equipment operation.

[0102] III. Real-time Refined Carbon Emission Calculation: Relying on the four dynamic coefficient systems mentioned above and AI intelligent calculation subroutines, the system realizes real-time carbon emission calculation for core building energy systems on an hourly, item-by-item, and precise quantitative basis. It further subdivides three core energy consumption scenarios to achieve refined data granular control. Lighting System Carbon Emission Calculation: The system collects real-time data on the number of public area lighting, indoor lighting, and landscape lighting systems, their operating duration, and power load. Combined with lighting equipment type, energy efficiency rating, light decay coefficient, and grid time-of-use carbon factor, it calculates the real-time carbon emissions of the lighting system every second. It distinguishes between carbon emission data for different scenarios such as normal lighting, emergency lighting, and holiday lighting, enabling itemized statistics and precise source tracing. Air Conditioning System Carbon Emission Calculation: The system links with the operating data of central air conditioning, split air conditioning, and integrated fresh air conditioning units, collecting parameters such as equipment start / stop status, cooling / heating power, operating load, indoor and outdoor temperature and humidity, and energy efficiency ratio. Combined with seasonal operating conditions, grid time-of-use carbon factor, and equipment energy efficiency decay coefficient, it dynamically calculates the real-time carbon emissions of the air conditioning system, while also distinguishing the differentiated carbon emission characteristics of the cooling season, heating season, and transitional season. Carbon emission accounting for fresh air systems: Data such as operating power, air volume, operating time, and filter wear of fresh air fans, exhaust equipment, and air purification equipment are collected. Combined with the carbon correction coefficient for equipment loss and the dynamic carbon factor of the power grid, the real-time carbon emissions corresponding to the operating energy consumption of the fresh air system are calculated, taking into account the balance between the rigid demand for ventilation and the carbon emissions from energy consumption.

[0103] IV. Full Life Cycle Implicit Carbon Overlay Accounting and Carbon Asset Ledger Construction: The system breaks through the limitations of traditional methods that only calculate carbon emissions during the building operation phase. It constructs a four-stage full life cycle carbon accounting model encompassing "production-construction-operation-decommissioning." Based on real-time operational carbon emissions, it comprehensively overlays implicit carbon emissions from various electromechanical equipment throughout the entire process, forming a complete and closed-loop building carbon asset data system, including: 1. The full-process implicit carbon accounting coverage includes: Equipment production stage: collecting carbon data on the production and manufacturing of all electromechanical equipment such as building lighting, air conditioning, fresh air, elevators, water supply and drainage, energy storage, and photovoltaics, including the implicit carbon emissions of the entire process from raw material mining, parts processing, equipment assembly, and factory energy consumption. The carbon emissions of a single piece of equipment are accurately quantified based on equipment model, parameters, and production batch. Equipment transportation stage: combining parameters such as equipment production location, transportation distance, transportation mode (road, rail, waterway), and energy consumption of transportation vehicles, the carbon emissions of equipment transportation from the production plant to the project construction site are calculated and included in the building's full life cycle carbon ledger. Equipment installation stage: collecting carbon emissions such as mechanical energy consumption, manual auxiliary energy consumption, and auxiliary material loss during equipment installation and construction, and accurately calculating the implicit carbon emissions of the installation stage based on data such as construction period and construction equipment power. Equipment scrapping stage: based on parameters such as equipment design service life, scrapping and dismantling energy consumption, waste material recycling rate, and solid waste treatment carbon emissions, the carbon emissions and carbon reduction and recovery volume of the equipment during the retirement and scrapping stage are pre-calculated to achieve closed-loop accounting of the entire life cycle.

[0104] 2. Construction of a Full Lifecycle Carbon Asset Ledger: Based on four-stage, multi-dimensional carbon accounting data, the system automatically constructs a full lifecycle carbon asset ledger for buildings. The ledger data is dynamically updated, archived in real-time, and categorized for statistical analysis. The core of the ledger comprises four modules: 1) Implicit carbon ledger for equipment production, categorizing and collecting carbon emission data from the production of all electromechanical equipment; 2) Carbon ledger for engineering construction, summarizing carbon emission data from the construction, installation, and transportation stages; 3) Dynamic carbon ledger for building operation, updating in real-time carbon emissions and zero-carbon deduction data from the operation of systems such as lighting, air conditioning, and ventilation; and 4) Pre-calculated carbon ledger for equipment decommissioning, storing carbon data for the equipment scrapping stage. The ledger supports itemized queries, time-period queries, equipment-dimensional queries, and full-lifecycle summary queries, providing comprehensive data support for carbon asset management, low-carbon retrofitting, and carbon value conversion.

[0105] V. Consortium Blockchain Trusted Evidence Storage Mechanism (Immutable, Fully Traceable): To address the industry pain points of easily tampered, difficult-to-trace, and insufficient credibility of building carbon data, the unit builds a comprehensive data on-chain evidence storage system to achieve secure encryption, hash-based evidence storage, full traceability, and cross-verification of all core carbon data. The system encrypts and stores all raw data, covering all core dimensions of data, including real-time building energy consumption data, photovoltaic self-generated power, energy storage charging and discharging, grid power purchase and sales data, building load regulation data, indoor and outdoor environmental monitoring data, and comprehensive carbon accounting results. After data collection, it is first encrypted using a national-level asymmetric encryption algorithm to generate a unique encryption key and data fingerprint. Then, it is uploaded to the blockchain in real time through dedicated nodes, automatically generating a unique hash value. Each batch of data corresponds to a unique hash identifier, and data blocks are linked together in chronological order. This evidence storage mechanism has four core characteristics: First, it is tamper-proof. Once the data on the chain is uploaded, no entity can modify, delete, or overwrite it. Even the slightest change to the original data will invalidate the hash value, allowing for rapid identification of tampering. Second, it is non-deletable. All data uploaded to the chain is permanently archived and stored in a chain-like manner, with no permission to clear it or any expiration date. Third, it is fully traceable. It supports one-click tracing of the original data, calculation process, on-chain records, and operating entities by time, device, data type, and calculation batch. Fourth, it allows for cross-verification. Regulatory agencies, third-party verification units, and financial institutions can cross-verify the original data, calculation results, and evidence storage records through authorized nodes on the consortium chain, ensuring the authenticity, compliance, and authority of the carbon data.

[0106] VI. Automatic Generation of Standardized Carbon Verification Certificates for Multiple Scenarios: The unit incorporates multi-system green and low-carbon evaluation standard algorithm templates, which can automatically adapt to mainstream domestic and international green building evaluation, carbon asset declaration, ESG disclosure, carbon trading, and green finance declaration systems according to project needs. It generates standardized, compliant, and directly reusable carbon verification certificates with one click, eliminating the need for manual secondary calculation and processing, significantly reducing project declaration and verification costs. Comprehensive scenario coverage: Green building three-star certification verification, LEED international green building certification, BREEAM green building assessment, corporate annual ESG report carbon data disclosure, CCER forestry / construction carbon sink project development declaration, green credit carbon verification, green bond issuance carbon asset filing, and other scenarios. All generated verification certificates are linked to on-chain hash storage records, accompanied by data traceability QR codes and verification codes. Third-party institutions can directly scan the codes to verify the data authenticity, and the certificates have officially recognized compliance validity.

[0107] VII. Dynamic Carbon Emission Reduction Calculation and Carbon Trading Revenue Calculation: The system incorporates a real-time dynamic carbon trading price model, connecting in real-time with dynamic data from the national carbon market, local carbon market carbon quota trading prices, CCER carbon sink trading prices, and green certificate trading prices. Combined with the project's real-time carbon accounting results, it achieves accurate calculation of the project's annual carbon emission reduction and dynamic calculation of potential carbon trading revenue. The model can automatically calculate the project's annual certified carbon emission reduction based on core data such as building baseline carbon emissions, actual operating carbon emissions, photovoltaic zero-carbon credit, energy storage peak-shaving emission reduction, and energy-saving retrofit emission reduction. Simultaneously, it dynamically updates carbon asset valuation, annual potential carbon trading revenue, and long-term carbon asset appreciation potential based on carbon market price fluctuation trends. This provides accurate data support for project carbon asset revitalization, carbon trading market entry, and green investment and financing decisions, realizing the quantification, visualization, and market-based transformation of the low-carbon value of buildings.

[0108] In one embodiment, the system relies on a deeply integrated architecture of distributed microgrid source-load-storage collaborative units and flexible control units for lighting, air conditioning, and fresh air, breaking down the technical barriers of independent operation of equipment, data silos, and fragmented strategies in traditional building energy systems. By building a fully interconnected data transmission system and an intelligent collaborative algorithm engine, it achieves an integrated operation mode with complete data interoperability at the bottom layer, deep linkage of top-level algorithms, and coordinated matching of operating conditions and strategies. Based on the actual energy consumption scenarios of buildings, photovoltaic output characteristics, grid pricing mechanisms, energy storage operation status, and grid demand response commands, the system accurately identifies four typical extreme operating conditions of building operation. It then executes refined, differentiated, and intelligent collaborative scheduling strategies targeting the energy consumption pain points, resource characteristics, and control objectives of different operating conditions, comprehensively improving building energy utilization efficiency, energy economy, grid adaptability, and green and low-carbon levels. The entire control system possesses core capabilities such as second-level full-domain perception, real-time computing power analysis, dynamic strategy iteration, and precise load control. It can dynamically adapt the three core flexible loads—building lighting, air conditioning, and fresh air—to photovoltaic and energy storage microgrid units. Simultaneously, it can be standardized and integrated with virtual power plant dispatch systems, providing core technical support for smart building energy use, lean grid dispatch, and energy market trading. The detailed collaborative dispatch strategies for various typical extreme operating conditions are explained below: I. Operating Condition 1: Solar PV Midday Surplus Generation Coordinated Dispatch Strategy. This condition primarily occurs during the midday period on sunny days, when solar radiation intensity is high, distributed photovoltaic (PV) module output reaches its daily peak, and real-time PV power generation far exceeds the building's basic electrical load. The system experiences a large surplus of green electricity, representing a crucial window for on-site PV consumption, improving green electricity utilization, and reducing dependence on grid power. For this condition, the AI ​​intelligent dispatch engine activates a special strategy to maximize on-site green electricity consumption. Through flexible load control and energy storage linkage, nearly 100% on-site PV power generation is achieved. The specific control logic is as follows: First, based on a real-time data sensing module, the system synchronously collects core data such as photovoltaic output data for each zone, lighting power consumption for each floor and area of ​​the building, indoor temperature and humidity parameters, air conditioning load, and fresh air operation parameters. Through AI algorithms, it accurately matches surplus photovoltaic power with the building's flexibly adjustable load. On this basis, the AI ​​engine proactively and hierarchically adjusts the lighting power of different zones, abandoning the traditional one-size-fits-all dimming mode. It differentiates the output power of the lights based on the lighting conditions, occupancy status, and basic lighting needs of different areas. By appropriately reducing the operating load of the lighting fixtures, it effectively reduces infrared heat radiation and surface heat dissipation generated by prolonged full-power operation, reducing the increase in indoor environmental heat caused by light heat dissipation at the source, and reducing the additional cooling consumption of air conditioning due to indoor heat accumulation, forming a closed-loop energy-saving system of "lighting load reduction—indoor heat reduction—air conditioning consumption reduction." Second, the system links with the central air conditioning control system, appropriately relaxing the reasonable temperature control range of the air conditioning while strictly ensuring basic indoor comfort standards, avoiding excessive cooling and redundant cooling waste under traditional fixed temperature thresholds. Conventional building air conditioning systems typically use fixed temperature settings, leading to resource mismatch issues during peak solar power generation periods at midday, where surplus solar power generates energy while air conditioning consumes excessive amounts. This system addresses this by employing a dynamic temperature threshold algorithm, incorporating multi-dimensional parameters such as outdoor ambient temperature, solar radiation intensity, indoor occupancy density, and wall insulation temperature rise. This dynamically expands the adjustable temperature range for air conditioning, eliminating ineffective cooling energy consumption and maximizing the absorption of local solar power. Finally, for surplus solar power that cannot be absorbed locally even after flexible load adjustment, the system automatically triggers an energy storage charging strategy. It precisely controls the energy storage battery bank to absorb the surplus green electricity smoothly at optimal charging power, completing the local storage of green energy. The stored green electricity can be released and used during periods of insufficient solar power output or peak grid electricity prices, achieving a spatial and temporal transfer of green energy resources. Through this dual control mode of "flexible load absorption + energy storage as a backup," the system completely solves the problem of midday solar curtailment, achieving nearly 100% local absorption of solar power generation, significantly increasing the proportion of green energy in buildings, and reducing building carbon emissions.

[0109] II. Operating Scenario 2: Collaborative Dispatch Strategy for Peak Grid Electricity Prices and High Demand. This scenario primarily occurs during peak daytime and evening electricity consumption periods in urban power grids. It is characterized by high grid electricity prices, significant grid supply pressure, and stringent demand billing. Under traditional building energy consumption models, large-scale electricity purchases during peak hours result in high electricity costs. Simultaneously, excessively high instantaneous peak loads in buildings can inflate the maximum demand declaration value for the grid, easily leading to demand default penalties and significantly increasing building operation and maintenance energy costs. To address this economic challenge, the system initiates a special peak load shaving and consumption reduction strategy. Through a combination of energy storage replenishment and flexible load reduction using multiple devices, it comprehensively improves the economic efficiency of building energy consumption. At the energy storage regulation level, the microgrid source-load-storage collaborative unit accurately monitors grid electricity price periods, real-time electricity load, and energy storage SOC (state of charge) status. During peak grid periods, it controls the energy storage battery pack to replenish building loads in a high-power, stable, and continuous discharge mode. It replaces the grid's high-priced peak electricity with low-priced off-peak electricity stored in the energy storage and local photovoltaic green electricity, directly reducing peak electricity costs from the energy procurement end and reducing the building's dependence on grid peak power. At the flexible load collaborative regulation level, the system synchronously links lighting, air conditioning, and fresh air systems to achieve stepless and smooth load reduction. For the lighting system, stepless dimming technology is employed to smoothly reduce the output power of the entire lighting area based on indoor lighting conditions and occupant activity, avoiding instantaneous power fluctuations and reducing lighting load without affecting basic office and living lighting needs. For the air conditioning system, precise cooling control is implemented, dynamically optimizing the operating frequency of the air conditioning unit, the speed of the terminal fans, and the chilled water flow. This precisely reduces redundant cooling load while ensuring a relatively stable indoor temperature and no significant difference in occupant comfort. For the fresh air system, the fresh air supply speed and air exchange frequency are adaptively reduced based on indoor occupant density and air quality parameters, lowering the energy consumption of the fresh air equipment. Through deep linkage between energy storage replenishment and stepless flexible load reduction across multiple devices, the system can smoothly reduce the building's instantaneous peak electrical load, effectively smoothing out the building's electricity consumption curve. On the one hand, it significantly reduces electricity costs during peak grid periods, thereby reducing monthly and annual electricity costs for buildings; on the other hand, it effectively reduces the maximum demand declaration value of the power grid, avoids the risk of penalties for exceeding demand, and comprehensively improves the energy economy of buildings from both the electricity price and demand dimensions. At the same time, it effectively alleviates the peak power supply pressure on the regional power grid and has good grid compatibility value.

[0110] III. Operating Condition 3: Collaborative Scheduling Strategy for Insufficient Photovoltaic Output, Weak Light Due to Cloudy Weather, and Low SOC of Energy Storage. This condition often occurs during periods of weak light, such as cloudy weather, fog, or overcast skies, and at night when there is no photovoltaic output. Its core characteristics include a sharp drop in distributed photovoltaic output, even to zero; energy storage batteries at low SOC levels; insufficient green energy resources available; and the building's operational electricity consumption relying entirely on grid power, significantly increasing grid purchase costs. To minimize grid power consumption and control energy costs, the system initiates a precise load management strategy. Through AI-adaptive constraint control, it strictly controls the building's total electricity load while ensuring a basic living experience. The AI ​​intelligent engine, based on real-time collected photovoltaic output data, energy storage SOC values, grid electricity prices, indoor and outdoor environmental parameters, and occupancy scenarios, adaptively tightens indoor living comfort constraint thresholds, abandoning the usual lenient comfort control standards. It optimizes equipment operating parameters with "basic guarantee and extreme energy saving" as the core objective. Regarding air conditioning system control, the system precisely compresses the temperature and humidity adjustment range, eliminating redundant fine-tuning functions. Within a comfortable range acceptable to the human body, it moderately adjusts the air conditioning operating conditions, reducing the continuous energy consumption of the air conditioning unit and terminal equipment, and eliminating ineffective energy consumption in low-light, low-energy-storage scenarios. Regarding lighting system control, the system strictly limits the maximum brightness output threshold of lighting based on indoor lighting levels and the areas where personnel are on duty, turning off redundant auxiliary lighting, landscape lighting, and backup lighting, retaining only basic functional lighting, and precisely controlling the overall power load of the lighting system to prevent excessive energy waste from excessive lighting. Regarding fresh air system control, the system combines indoor carbon dioxide concentration, air quality, and the number of people to reasonably reduce the basic fresh air supply volume and air exchange frequency, ensuring indoor air quality meets standards and eliminates stuffiness while minimizing the power consumption of continuous fresh air equipment operation. The entire control strategy comprehensively reduces redundant building power load through precise, hierarchical, and adaptive management of three high-energy-consuming devices: air conditioning, lighting, and fresh air systems. It also precisely controls the total building power consumption, minimizes unnecessary grid power purchases, effectively reduces grid power costs during off-peak hours, and achieves efficient, economical, and stable building energy use under harsh conditions such as low light and low energy storage, balancing the dual needs of human living experience and energy conservation.

[0111] IV. Operating Scenario 4: Virtual Power Plant Grid Load Downshoring Invitation and Demand Response Trigger Scenario. With the continuous advancement of the new power system construction, virtual power plants have become the core carriers for grid load regulation, supply and demand balance, and renewable energy consumption. The grid will issue load downshoring invitations and demand response dispatch instructions to adjustable load resources within its jurisdiction based on the overall power supply and demand situation. For this operating scenario, this system can quickly respond to grid dispatch needs, accurately output standardized, controllable, and stable flexible load adjustment resources, and efficiently support grid supply and demand balance dispatch. Upon receiving the virtual power plant grid load downshoring invitation and demand response trigger signal, the system immediately activates the overall flexible load resource survey and calculation mechanism, constructing a three-dimensional dynamic adjustable load capacity matrix for lighting, air conditioning, and fresh air systems in real time. The matrix covers all dimensions of core parameters for the three types of equipment, including real-time operating power, rated adjustable upper limit, instantaneous adjustable rate, continuous adjustment duration, minimum operating guarantee power, and adjustment response delay. Through big data algorithms, the adjustable load potential of each type of equipment and each region is accurately calculated, forming a visualized, quantifiable, and dispatchable load resource database. Based on the target power reduction values, response times, and adjustment accuracy requirements issued by the power grid, the AI ​​optimal allocation algorithm intelligently decomposes and dynamically optimizes the adjustment ratios of the three types of flexible equipment. It differentiates the allocation of adjustment tasks by considering the adjustment response speed, load adjustment potential, and impact on residential areas of different equipment. Priority is given to adjusting the loads of equipment with minimal impact on residential experience, fast response speed, and large adjustable capacity. Load reduction operations are executed step-by-step, layered, and orderly to avoid environmental fluctuations and experience degradation caused by large adjustments to a single equipment, while ensuring that the overall adjustment power accurately matches the power grid dispatch target. Throughout the demand response process, the system continuously outputs continuous, stable, and controllable standardized adjustable load resources. The adjustment accuracy, response speed, and continuous stability fully meet the refined requirements of virtual power plant power grid dispatch, effectively helping the power grid to smooth load fluctuations, balance regional power supply and demand, and alleviate grid voltage regulation pressure.

[0112] V. A comprehensive load dynamic statistics and virtual power plant ledger support system: To achieve standardized, market-oriented, and traceable management of building flexible load resources, the system is equipped with a second-level comprehensive data statistics and encrypted reporting module, enabling 24 / 7 dynamic monitoring and ledger updates of building adjustable load resources. The system provides real-time statistics on the total adjustable load capacity, executed adjustment capacity, and remaining adjustable load capacity across the entire building area at a second-level frequency. It accurately records core data such as adjustment duration, adjustment power, response accuracy, and energy consumption changes for each demand response, forming a fully traceable, statistically verifiable data chain. Simultaneously, the system, adhering to the data specifications and security standards of the virtual power plant dispatch platform, periodically encrypts and compliantly encapsulates statistical data, stably uploading it to the superior virtual power plant dispatch platform to dynamically update the building flexible load resource ledger. This ledger possesses the core characteristics of dynamic updates, data authenticity, full traceability, and tradable settlement, transforming the scattered flexible electricity loads of buildings into standardized and large-scale grid adjustable resources. This ledger system can effectively support core businesses such as regional unified load dispatching of the power grid, electricity market trading, and precise settlement of demand response subsidies. It allows building flexible loads to deeply participate in the operation of the new power system, realizing both refined and autonomous management of building energy consumption and empowering efficient grid dispatching. This helps buildings achieve multiple value enhancements, including energy conservation, economic revenue generation, and green and low-carbon development.

[0113] In one embodiment, the system addresses industry pain points such as high energy consumption, difficulty in carbon emission control, low level of intelligence, and poor scenario adaptability in modern buildings, innovatively adopting an integrated architecture of eight core modules. All hardware devices, embedded software programs, intelligent algorithm logic, and standardized data interfaces corresponding to all modules of the system achieve complete decoupling of hardware and software across all dimensions, and a standardized plug-and-play modular design. This architecture completely overturns the shortcomings of traditional building energy-saving systems, such as deep hardware and software binding, fixed and rigid functions, difficulty in later expansion, single application scenarios, and high upgrade and renovation costs. Each functional unit possesses the core characteristics of independent operation, no interference, free selection, flexible deletion, and iterative upgrades, and can accurately adapt to various differentiated engineering scenarios such as energy-saving renovation of traditional existing buildings, new zero-carbon buildings, near-zero energy buildings, and green and low-carbon public buildings. At the same time, the system reserves ample space for functional expansion and data interaction interfaces, possessing strong capabilities for later functional iteration, equipment expansion, and scenario expansion, as detailed below: I. Modular and Decoupled Design Characteristics of Core Functional Units: The system is divided into four value-added functional units and four basic support units. All eight modules adhere to the design principles of "standardized interfaces, independent operation, modular combination, and pluggable deployment," forming the core foundation for the system's multi-scenario adaptability, high scalability, and high flexibility. All modules adopt industry-standardized communication interfaces, control protocols, and data formats. Modules interact and link commands via a standardized bus, eliminating proprietary binding protocols and achieving complete decoupling of hardware and software from the underlying layer. The four value-added functional units include a distributed microgrid source-load-storage coordination unit, a digital twin virtual-real mapping unit, a virtual power plant aggregation and dispatch unit, and a blockchain-based full-lifecycle carbon data trusted storage unit, serving as the carriers of the system's high-end, low-carbon, intelligent, and market-oriented core capabilities. Each of the four units has a built-in independent embedded computing chip, data processing subroutine, command output control module, and data encryption transmission module. They do not rely on fixed hardware hosts or dedicated software systems and can be deployed individually, in combination, or shut down when idle, depending on project construction standards and actual needs, fully supporting pluggable deployment. In practical engineering, low-standard renovation projects can directly eliminate value-added modules, retaining only basic functions; high-standard zero-carbon projects can deploy the entire system to achieve full functional coverage, effectively avoiding the problems of functional redundancy, resource waste, and excessive costs associated with traditional systems. The four basic support modules serve as the core foundation for system operation, specifically including a multi-dimensional sensing terminal covering the entire domain, a cloud-edge-device full-domain AI adaptive computing architecture, an integrated flexible load control unit for lighting / air conditioning / fresh air, and a three-layer heterogeneous redundant communication network of 5G+WiFi+LoRa. These basic modules provide the underlying guarantee for system energy-saving control, data acquisition, computing power support, and communication transmission. They can independently support basic building energy-saving and efficiency-enhancing functions, while seamlessly linking and deeply coupling with the four value-added modules to achieve a comprehensive upgrade of system functions from basic energy saving to zero-carbon management, market-oriented operation, and digital visualization. The completely decoupled hardware and software design allows system hardware to be replaced and upgraded as needed based on technological iterations, and software algorithms to be remotely iterated and customized through the cloud, without requiring the dismantling and reconstruction of the entire system. This significantly reduces the initial construction cost, mid-term renovation cost, and later operation and maintenance upgrade cost of building energy-saving systems.

[0114] II. Modular Adaptation Solution for Differentiated Scenarios: (I) Energy-saving retrofitting scenarios for existing buildings without photovoltaics or energy storage: Currently, a large number of existing buildings in China are of traditional design and lack photovoltaic power generation and energy storage equipment. Their electromechanical equipment operates in a crude manner, lacks energy consumption management, and has extremely low levels of intelligence, resulting in serious energy waste, high operation and maintenance costs, and low comfort for living and working. The core requirements for retrofitting such buildings are low-cost implementation, lightweight renovation, rapid energy-saving and efficiency improvement, no need for large-scale civil engineering construction, and no need to build high-end functions such as source-load-storage coordinated scheduling, carbon data traceability and certification, and grid market linkage. Based on the core characteristics of pluggable modularity and on-demand configuration, a simplified configuration scheme can be adopted for this type of scenario. Only four basic support units need to be deployed to achieve intelligent and low-cost energy-saving upgrades of traditional existing buildings, adapting to the energy-saving retrofitting needs of most old residential buildings, old office buildings, and old public buildings, as detailed below: 1. Comprehensive Multi-Dimensional Sensing Terminals: The system can precisely deploy comprehensive multi-dimensional sensing terminals at key locations such as office areas, public corridors, underground computer rooms, equipment floors, and indoor functional areas, based on the building's functional layout. Terminal types include indoor and outdoor temperature and humidity sensors, natural light intensity sensors, infrared human presence sensors, energy consumption metering sensors for electromechanical equipment, and indoor PM2.5 and carbon dioxide concentration air quality sensors. All terminals adopt a low-power design, support rapid installation without wiring, and can continuously and accurately collect dynamic data on the building environment, operational status data of electromechanical equipment, and hourly energy consumption data 24 / 7. Through the accumulation of massive real-time data, it completely solves the industry pain points of traditional energy consumption management in existing buildings, such as lack of data support, blind equipment control, and experience-based operation and maintenance management, providing a core data foundation for AI-based precise energy-saving control.

[0115] 2. Cloud-Edge-End Global AI Adaptive Computing Architecture: The architecture adopts a three-tiered collaborative model of terminal edge computing power, local edge node computing power, and cloud supercomputing power to construct a global, hierarchically scheduled AI computing power system. Terminal edge computing power is responsible for the initial cleaning, filtering, and execution of local commands from the perceived data; local edge node computing power is responsible for the coordinated control of devices within the region, short-term data modeling, and load prediction and analysis; cloud supercomputing power is responsible for global data aggregation, long-term energy consumption pattern modeling, algorithm iteration and optimization, and cross-regional overall scheduling. Through the collaboration of these three levels of computing power, the system can perform real-time analysis, dynamic modeling, and intelligent judgment on massive amounts of collected data. It can autonomously learn the building energy consumption patterns and environmental change patterns under different seasons, time periods, and personnel conditions, and adaptively optimize equipment operation strategies without manual intervention, achieving unmanned, intelligent, and precise energy-saving management.

[0116] 3. Integrated Flexible Load Control Unit for Lighting / Air Conditioning / Fresh Air: This unit establishes an integrated flexible load control system for the three core high-energy-consuming electromechanical equipment components: building lighting, central air conditioning, and fresh air systems. It connects the independent control links of various devices, enabling coordinated operation and unified scheduling and management of multiple devices. Based on sensor data and AI computing power analysis results, combined with multi-dimensional parameters such as indoor occupant density, indoor and outdoor ambient temperature, natural light conditions, grid load during certain periods, and building energy consumption baseline, the system dynamically and adaptively adjusts the lighting switch status and brightness, the air conditioning operating temperature and humidity and start / stop mode, and the fresh air system's ventilation frequency and air volume. This effectively eliminates waste problems such as equipment idling, overloading, and ineffective energy output, accurately matching the actual usage needs of the building, maximizing the energy-saving potential of traditional existing buildings, and achieving refined management of equipment energy consumption.

[0117] 4. 5G+WiFi+LoRa Three-Layer Heterogeneous Redundant Communication Network: To ensure stable data transmission and timely command response, a three-layer heterogeneous redundant communication architecture with full coverage, high reliability, and low latency is established. The LoRa network, with its advantages of long range, low power consumption, and wide coverage, is primarily responsible for low-speed data collection and transmission from sensing terminals across the entire area, adapting to scenarios with large-scale, multi-point terminal access in buildings. The WiFi network, with its high bandwidth and low latency, is responsible for high-speed interaction of indoor short-range device control commands and high-definition data. The 5G network, with its wide-area coverage, high-speed transmission, and low-latency remote scheduling capabilities, is responsible for system cloud data synchronization, remote operation and maintenance command transmission, and cross-scenario data interaction. The three networks serve as redundant backups for each other. When a single network experiences lag, disconnection, or signal fluctuations, it can automatically switch to the backup network, comprehensively ensuring the real-time and stable transmission of building sensing data and device control commands. This completely avoids problems such as control failure and device malfunction caused by network failures, ensuring the system's safe and stable operation around the clock.

[0118] Through the lightweight deployment of the four basic modules mentioned above, traditional buildings can be rapidly upgraded to intelligent energy-saving systems without requiring any civil engineering modifications to existing buildings or the addition of photovoltaic and energy storage equipment. This solution can effectively reduce the overall energy consumption of buildings by 15%-25%, significantly optimize the comfort of indoor living and office environments, and has the advantages of short renovation cycles, low construction difficulty, low investment costs, and stable and significant energy-saving effects, making it highly suitable for large-scale energy-saving renovation projects of existing buildings.

[0119] (II) New Zero-Carbon / Near-Zero Energy Building Scenarios: New zero-carbon and near-zero energy buildings represent the core development direction of current green and low-carbon buildings. These buildings have extremely high standards for intelligent management and control, efficient energy utilization, precise carbon emission control, positive interaction with the power grid, and digital and visualized operation. Core construction needs focus on all-dimensional zero-carbon management and control, efficient synergy between energy sources, loads, and storage, optimal building energy consumption configuration, full-cycle carbon data traceability, and market-based arbitrage participation in the power grid. Leveraging the modular and fully expandable characteristics of the system, new zero-carbon buildings can be fully equipped with eight core functional modules, achieving comprehensive coverage of building energy conservation and consumption reduction, carbon emission control, new energy consumption, power grid linkage, visualized operation and maintenance, and carbon asset management. This fully meets the high standards for acceptance and operation of national and industry-standard zero-carbon and near-zero energy buildings. Based on the four basic support modules that solidify the system's underlying operational capabilities, four high-end value-added functional modules are deployed to comprehensively empower the zero-carbon, intelligent, and market-oriented operation of new buildings. The distributed microgrid source-load-storage coordination unit can coordinate and regulate building photovoltaic power generation systems, energy storage systems, and building electricity loads, matching the power generation of new energy sources with the building's electricity demand in real time. It dynamically achieves a balance and optimal scheduling among the three elements, significantly improving the local utilization rate of distributed new energy, reducing dependence on public grid electricity, and lowering building carbon emissions and energy consumption from the energy supply side. The digital twin virtual-real mapping unit, based on the building's BIM model and real-time operational data, builds a 1:1 three-dimensional virtual simulation model of the building. This enables synchronized and visualized real-time monitoring of building equipment operating status, real-time energy consumption data, indoor and outdoor environmental parameters, carbon emission data, and new energy operation data, supporting fault tracing, energy consumption analysis, and operation and maintenance prediction. The virtual power plant aggregation and dispatch unit can aggregate building flexible load resources into the grid, participating in market-based transactions such as grid peak-valley regulation, demand response, and ancillary services. It leverages peak-valley electricity price differences to achieve market arbitrage, effectively reducing long-term building energy costs and enabling buildings to transform from "passive electricity consumption" to "active participation in grid dispatch." The blockchain-based full-lifecycle carbon data trusted storage unit can collect and store carbon emission and carbon reduction data on the blockchain in real time throughout the entire lifecycle of building materials production, construction, operation, and use. Leveraging the immutability and traceability of blockchain, it ensures the authenticity and reliability of carbon data, directly supporting the implementation of zero-carbon building certification, green building rating, carbon asset accounting, and carbon trading. Eight modules operate in a comprehensive and coordinated manner, constructing a full-chain management system covering building energy, energy consumption, carbon emissions, operation and maintenance, and market-based operation, fully meeting the high-standard construction and long-term operation needs of newly built zero-carbon and near-zero-energy buildings.

[0120] III. A Six-Element Synergistic Passive-Active Energy-Saving System: To further break through the energy-saving bottleneck of traditional single electromechanical control in buildings, deeply explore the potential of passive energy saving in buildings, and make up for the shortcomings of active control energy saving, the system reserves dedicated hardware control interfaces and software data interaction interfaces. It innovatively constructs a six-element synergistic energy-saving system of photovoltaics, energy storage, electric shading, lighting, air conditioning, and fresh air, realizing the deep integration and multi-dimensional linkage of passive natural energy saving and active intelligent control, and comprehensively improving the overall energy efficiency of buildings. The system is separately configured with a dedicated control interface and standardized data interaction interface for electric shading louvers, upgrading the traditional electric shading louvers, which only have a shading function, into an important passive energy-saving flexible and adjustable resource for buildings, and uniformly connecting to the edge control platform for overall scheduling. Unlike the traditional crude mode of fixed shading structures and manual adjustment of louvers, this system relies on AI intelligent algorithms, combined with multi-dimensional parameters such as outdoor real-time light intensity, ambient air temperature, seasonal climate characteristics, real-time building power consumption, indoor occupant load, and equipment operating status, to achieve dynamic, precise, and fully automatic adjustment of louver opening angle, lifting range, and start / stop status, without manual intervention. Under high temperatures and intense sunlight in summer, the system automatically adjusts the louvers to precisely block strong outdoor sunlight and solar radiation, preventing solar heat from entering the interior at the source. This significantly reduces the heat gain of the building envelope, effectively reducing the basic cooling load of the central air conditioning system and preventing the air conditioning from operating at full load and high power for extended periods, thus significantly reducing building cooling energy consumption in summer. During the gentler sunlight periods of spring and autumn, and the periods of higher lighting demand in winter, the system adaptively adjusts the opening angle of the louvers to maximize the introduction of natural light, replacing artificial lighting equipment and effectively reducing building lighting energy consumption. Simultaneously, the system can dynamically balance shading, lighting, ventilation, and temperature control based on outdoor wind speed, air quality, and indoor temperature and humidity conditions, linking the louver status with the fresh air system's operating mode to adapt to the energy-saving needs of different seasons, weather conditions, and usage scenarios. In this six-element synergistic system, photovoltaic and energy storage systems constitute the active energy regulation capability for new energy supply and peak shaving; lighting, air conditioning, and fresh air systems constitute the active optimization capability for indoor load; and motorized shading louvers constitute the passive energy-saving capability for source-based energy reduction. The six major resources work together and complement each other to form a six-dimensional integrated energy-saving management system that integrates "source reduction, energy optimization, load regulation and intelligent adaptation". This system reduces ineffective energy consumption in buildings at each level, avoids the limitations of single energy-saving methods, and significantly improves the overall energy efficiency of buildings, providing core technical support for buildings to achieve near-zero energy consumption and zero-carbon normalized operation.

[0121] IV. Core Advantages of Modular System Design: 1. Highly adaptable to various scenarios, covering all building types: Based on pluggable modular design and decoupled hardware and software design, it can flexibly adapt to various scenarios such as lightweight and low-cost renovation of old existing buildings and high-standard construction of new zero-carbon / near-zero energy buildings with full functionality. It takes into account the needs of inclusive renovation and high-end construction, and is suitable for various types of buildings such as residential buildings, office buildings, public buildings, and park buildings.

[0122] 2. Excellent scalability and iterative capabilities, with an ultra-long lifespan: The system has no hardware or software binding restrictions, supports the addition of new functions, expansion of hardware devices, remote iteration of intelligent algorithms, and upgrade of scenario functions in the future, without the need to completely dismantle and reconstruct the original system. It can adapt to the iterative updates of future low-carbon policies, smart technologies, and power grid models, greatly extending the system's lifespan.

[0123] 3. Comprehensive energy-saving measures with significant consumption reduction: Breaking away from the traditional single-equipment energy-saving model, it integrates passive shading natural energy saving with active electromechanical control, new energy synergy, and full-cycle carbon data management, achieving refined management of building energy consumption from the source, process, and end, with energy-saving and consumption-reducing effects far exceeding those of traditional building energy-saving systems.

[0124] 4. Outstanding economic benefits throughout the entire life cycle, with significant cost reduction and efficiency improvement: Selecting modules on demand can effectively avoid functional redundancy and reduce the initial construction investment cost of the project; AI-based precise energy saving and six-element collaborative energy saving continuously reduce the daily operation and maintenance energy consumption cost of the building; additional revenue can be realized by relying on the market-based trading of virtual power plants, comprehensively improving the economic benefits of the building throughout its entire life cycle.

[0125] 5. High system stability and convenient operation and maintenance: The three-layer heterogeneous redundant communication architecture ensures stable data transmission, and the modular independent operation mode can realize independent isolation of single module failures without affecting the overall system operation, greatly reducing the probability of system failure and the difficulty of later operation and maintenance, and adapting to the long-term stable operation requirements of buildings.

[0126] In one embodiment, with the rapid advancement of smart parks, green buildings, and zero-carbon parks, the energy consumption patterns of modern industrial parks, business parks, and integrated commercial and residential parks are becoming increasingly complex, exhibiting characteristics such as a wide variety of equipment, diverse energy consumption scenarios, strong randomness in load fluctuations, and significant differences in spatial functions. Traditional building energy management often adopts a centralized, one-size-fits-all, and extensive management model, which generally suffers from a series of pain points, including coarse spatial management granularity, poor equipment linkage, difficulty in controlling peak loads, an inability to balance human comfort and energy conservation, fragmented carbon energy consumption data statistics, and a lack of global coordination capabilities in energy dispatch. These issues make it difficult to meet the current needs of low-carbon, intelligent, refined, and efficient operation and maintenance management in parks. To address these industry pain points, this system has independently developed and adopted a six-level hierarchical refined distributed management architecture: park-multiple buildings-single building-floor-independent zone-independent room, completely overturning the traditional single-layer centralized management model. The system is designed with "layered decoupling, distributed computing, hierarchical control, and global collaboration" as its core principles, achieving independent operation of six-level architecture, real-time data exchange across levels, and intelligent collaborative linkage of devices across the entire domain. Through a three-dimensional control model of cloud-based global coordination, edge-level hierarchical scheduling, and precise terminal execution, it ensures both personalized living comfort in the smallest spatial unit and optimal energy scheduling for the entire building and park, achieving the comprehensive operational goals of lowest energy consumption, lowest carbon emissions, lowest maintenance costs, and highest energy benefits. The system architecture design, hierarchical functions, technical principles, control mechanisms, core advantages, and applications are described below: I. Overall System Architecture and Core Features: This system's six-level hierarchical, finely-defined distributed control architecture is a typical "cloud-edge-device" integrated smart energy management system. From top to bottom, it consists of six levels: the park cloud platform layer, the multi-building cluster collaboration layer, the single-building edge centralized control layer, the floor-level domain control layer, the functional zone finely-defined control layer, and the independent room terminal execution layer. Each level has clear responsibilities and division of labor, adopting a distributed independent computing + full-domain collaborative operation mode, fundamentally solving the shortcomings of traditional centralized control systems such as insufficient computing power, slow response, poor fault tolerance, weak scalability, and insufficient fine-grained control. The system as a whole possesses three core features, providing underlying technical support for full-domain smart energy management. First, each level of the architecture operates logically independently, significantly improving system fault tolerance and stability. Each control unit in the six-level architecture is configured with independent computing modules, control programs, logic policy libraries, and device execution permissions, achieving hierarchical decoupling and independent autonomy. Single-point problems such as park cloud platform failures, network interruptions, single-building equipment anomalies, and single-floor module failures will not affect the normal operation of other levels or other areas. The lower-level edge control unit can operate independently offline, autonomously completing energy consumption regulation, equipment management, load balancing, and fault protection within its region. This completely avoids the drawbacks of traditional centralized systems where "a single fault leads to the entire region's paralysis," greatly improving the stability and reliability of the park's energy system operating around the clock. Secondly, it enables real-time data interoperability across the entire region, eliminating information silos. The system builds an industrial-grade low-latency IoT communication system, opening up data interaction channels across all six levels of the architecture, achieving bidirectional and peer-to-peer data transmission across the entire region. Information across all dimensions, including equipment operating status, electrical parameters, environmental sensing data, personnel presence / absence data, real-time energy consumption, load power, control commands, fault alarms, and carbon accounting data, is updated synchronously in milliseconds, uploaded, and distributed in real-time. All levels of data use standardized protocols, unified data ledgers, and unified time traceability, completely solving problems such as data lag, data gaps, inconsistent statistical standards, and severe information silos in traditional systems, providing a precise, real-time, and complete data foundation for global intelligent algorithm calculations. Thirdly, it enables global collaborative linkage across the entire region, balancing personalization and optimization. The system supports top-down strategy batch distribution, hierarchical breakdown, and tiered implementation, while also supporting bottom-up data aggregation, load feedback, and anomaly reporting. Furthermore, it enables load coordination, equipment linkage, and fault assistance among buildings, floors, and adjacent zones at the same level. Through multi-level intelligent algorithm coupling, it achieves the dual goals of "personalized autonomous control in small areas and global collaborative optimization in large areas," ensuring optimal energy resource allocation for the entire building and the entire park while guaranteeing the comfort of each individual space.

[0127] II. Detailed Explanation of the Six-Level Hierarchical Control Unit's Step-by-Step Refined Functions 1. Park-level: Cloud-based intelligent platform for centralized and intensive management across the entire group. The park's cloud platform serves as the highest management and data hub of the entire system, providing centralized, visualized, digitalized, and intelligent energy management services for the entire park's multi-building complex and group-wide park clusters. Positioned as the core carrier for group-level coordination, comprehensive data aggregation, global strategy control, and integrated benefit accounting, it covers all dimensions of management and control for all buildings, floors, functional zones, and individual rooms within the park. Regarding centralized monitoring, the platform aggregates real-time operating data and status information for all electrical, HVAC, lighting, ventilation, energy storage, photovoltaic, and charging equipment in all buildings within the park. It also accesses information such as environmental parameters, personnel status, energy consumption data, load curves, control records, and fault logs. Through a 3D visualized park screen, digital ledgers, and real-time charts, it achieves 24 / 7 uninterrupted and comprehensive monitoring of the park's energy operation. It supports precise source tracing queries at multiple levels, including park, building, floor, zone, and room, accurately locating high-energy-consuming equipment, abnormal operation points, and load fluctuation nodes, providing intuitive and accurate visualization support for operation and maintenance management. Regarding batch policy control, the platform has the capability for group-level batch policy configuration and hierarchical distribution. Managers can formulate differentiated energy consumption control strategies, equipment operation specifications, load limit standards, and energy-saving control thresholds based on different building types (office buildings, conference centers, guest rooms / apartments, public supporting buildings), different usage periods (weekdays, weekends, holidays, day and night), and different operation and maintenance standards within the park. It supports one-click batch distribution to the edge control units of all buildings in the park, while also supporting customized policy fine-tuning for individual buildings and areas, balancing standardized unified control with personalized adaptation needs, and fully adapting to diverse and scenario-based operation and maintenance management models within the park. In terms of carbon energy consumption data aggregation and analysis, the platform has built a standardized carbon energy consumption accounting and analysis model, automatically collecting various energy consumption data such as water, electricity, gas, cooling and heating, and fresh air throughout the park. Combining multi-dimensional factors such as building area, personnel density, usage time, equipment configuration, and business type attributes, it conducts energy consumption statistics, energy consumption benchmarking, year-on-year and month-on-month energy consumption comparisons, energy consumption anomaly diagnosis, and energy-saving potential analysis, accurately identifying energy-wasting links and inefficiently operating equipment. Simultaneously, in accordance with national carbon emission accounting standards, it automatically completes the aggregation, statistics, ledger generation, and visualization of carbon emissions, carbon reduction, and clean energy substitution emission reduction data for the entire park, providing authoritative data support for low-carbon acceptance, green building rating, carbon asset inventory, and energy-saving renovation project initiation. Regarding unified statistics of comprehensive benefits, the platform constructs a comprehensive energy benefit accounting system, automatically calculating various economic benefits such as photovoltaic self-consumption revenue, energy storage peak-valley arbitrage revenue, demand response subsidy revenue, energy conservation and consumption reduction revenue, and energy trading revenue.The system can automatically generate daily, monthly, quarterly, and annual energy operation and maintenance reports and revenue analysis reports, enabling digital, refined, and visual management of the park's energy operation and maintenance benefits, and providing data-driven decision-making support for the iteration of park energy operation strategies, equipment upgrades and renovations, and optimization of operation and maintenance models.

[0128] 2. Multi-Building Level: The multi-building level serves as the core relay and collaboration layer between the park's cloud platform and the edge of individual buildings. It undertakes key functions such as strategy decomposition and adaptation, group load coordination, multi-building collaborative peak shifting, cross-building fault linkage, and cluster data aggregation. It is the core mid-level control unit for achieving overall load balance across the park and preventing peak energy consumption from exceeding limits. This level receives global energy consumption control indicators, load limits, energy-saving strategies, and safety regulations from the park's cloud platform in real time. Combining this with the building structure, functional positioning, equipment capacity, energy consumption characteristics, and historical load patterns of each building within the park, it decomposes the park-level macro-strategy into implementable, refined strategies tailored to individual buildings, achieving micro-level implementation of macro-goals. Simultaneously, this level aggregates real-time load, energy consumption status, equipment operating conditions, and fault information for all buildings, constructing a big data model of the park's building cluster load to accurately predict overall load fluctuation trends. To address issues such as soaring peak electricity consumption, excessive demand, surging peak electricity costs, and excessive grid connection pressure caused by the simultaneous start-up and shutdown of high-power equipment and simultaneous high-load operation of multiple buildings in the park, the multi-building management unit uses a cluster collaboration algorithm to coordinate the staggered start-up and shutdown, load complementarity, and time-sharing operation of each building, rationally allocate the load quota of each building, smooth out the peak electricity consumption of the park, achieve optimal energy synergy operation of the building group, and comprehensively improve the overall energy efficiency and grid adaptability of the park.

[0129] 3. Single Building Level: Edge Control Core, Building-wide Microgrid Security and Source-Load Balance Management. The single building edge controller is the localized intelligent management core of a single building, possessing core capabilities such as offline autonomous operation, edge intelligent computing, local policy execution, and full-domain scheduling and management. It can independently complete the energy management of the entire building without relying on a cloud platform. It is a key core level connecting upper-level cluster management and lower-level floor-zone refined management, mainly responsible for three core tasks: building microgrid source-load-storage power balance, full-domain load optimization scheduling, and equipment security protection and management. In terms of microgrid source-load-storage power balance management, the edge controller collects data from various energy supply ends such as distributed photovoltaics, energy storage systems, the mains power grid, charging piles, and DC loads in real time at high frequency, while simultaneously collecting data from all power load ends such as air conditioning, lighting, fresh air, water supply and drainage, office equipment, and smart terminals. Through real-time dynamic calculations using localized edge intelligent algorithms, the system accurately matches energy supply and load demand, dynamically adjusting the charging and discharging timing and power of the energy storage system, the proportion of clean energy consumption, and the grid power quota. This achieves second-level dynamic balance between supply and demand in the building microgrid system, maximizing the local consumption of photovoltaic clean energy, reducing grid power dependence, lowering electricity purchase costs, and improving the building's energy self-sufficiency rate. Regarding overall load optimization and scheduling, the central controller coordinates real-time load data from all floors, independent zones, and rooms of the entire building, constructing a dynamic load model for the entire building. This model monitors load fluctuations in real time, predicts peak nodes, and identifies load redundancy and gaps. Through a global intelligent scheduling strategy, the system dynamically allocates available load quotas for each floor and zone, coordinates the operating power and start-up / shutdown sequence of equipment in each area, and controls the load ceiling from the overall building level, preventing disorderly local load spikes and achieving balanced load and optimal energy consumption for the entire building. In terms of equipment safety management, the edge controller possesses comprehensive electrical safety and equipment operation safety protection capabilities, supporting multiple safety mechanisms such as line overload detection, short circuit protection, overvoltage and undervoltage monitoring, equipment overheating alarm, abnormal energy consumption fluctuation identification, and illegal electricity use alarm. Once abnormal operating conditions are detected, the system can automatically execute operations such as load limiting, power reduction, power outage protection, fault isolation, audible and visual alarms, and information reporting, effectively avoiding the risk of electrical safety accidents and equipment damage, and ensuring the long-term safe, stable, and reliable operation of the building's energy equipment.

[0130] 4. Floor-level: Domain-based subset control enables refined spatial domain scheduling. Each floor is independently configured with a domain-based subset control unit, receiving scheduling commands from the building's edge controller and independently responsible for the zoned control, load management, status monitoring, and collaborative scheduling of the entire space on its floor. The system breaks away from the traditional, extensive model of unified floor-wide control. Based on the building's functional layout and usage attributes, it finely divides the floor space, dividing different functional spaces such as open-plan large office areas, private offices, large and small meeting rooms, reception rooms, guest rooms, public corridors, lobbies, and equipment auxiliary areas into several independent and non-interfering finely managed zones, constructing a refined control pattern of "one floor, multiple zones, independent domain control, and overall collaborative management." The core of the floor-level domain-based subset control unit plays a crucial role in regional-level control, summarizing environmental data, equipment operation data, real-time energy consumption, load power, control status, and personnel usage of all zones on the floor in real time, and simultaneously uploading this data to the building controller and the park's cloud platform; it also precisely issues load control commands, scenario strategies, and equipment adjustment parameters downwards, coordinating the operational logic of each zone. It is also responsible for coordinating load conflicts, optimizing timing, and balancing power in multiple zones within a floor, effectively avoiding the problem of excessively high local peak values ​​and serious energy waste caused by multiple zones in the same floor starting high-power equipment and simultaneously surging high loads, thus achieving refined and collaborative management of floor-level energy.

[0131] 5. Independent Zonal Level: AI-Adaptive Closed-Loop Management Enables Scenario-Based Personalized Precise Control. Independent functional zones are the core refined management units for achieving a two-way balance between human comfort and energy conservation, and are also the core implementation level of AI intelligent regulation. Each subdivided functional zone is equipped with a dedicated and independent sensing network and flexible control terminal, forming a zone-specific "sensing-analysis-decision-control-feedback" closed-loop management system. It does not share equipment and strategies with other zones, ensuring the independence and precision of management. The zone-specific sensing network consists of high-precision sensors for temperature, humidity, light intensity, carbon dioxide concentration, air quality, human presence sensing, and micro-motion sensing. It can continuously and frequently collect environmental parameter changes, human presence / absence status, and space usage dynamics within the zone 24 / 7, providing comprehensive and high-precision data support for intelligent regulation. The zone's flexible control terminal fully connects to all controllable living equipment within the area, such as air conditioning, fresh air systems, lighting, curtains, exhaust systems, humidifiers, and dehumidifiers. It supports stepless fine adjustment, flexible power control, and precise start / stop management, eliminating the comfort fluctuations and energy waste caused by the crude on / off control of traditional equipment. Each independent zone possesses fully autonomous AI adaptive closed-loop control capabilities, enabling intelligent control of the entire process without intervention from upper-level units. Users can customize diverse and exclusive control rules based on space usage needs and personal comfort habits, adapting to various usage scenarios: First, users can freely set personalized human comfort thresholds such as temperature, humidity, light, and air quality. The system's AI algorithm automatically and dynamically fine-tunes the operating status of equipment around the comfortable range to ensure a constant and comfortable environment. Second, it supports customizable unattended delay strategies, combining human body sensing data to achieve precise operation and maintenance when manned and delayed shutdown when unattended, preventing long-term energy consumption in unattended spaces. Third, it supports one-click switching between multiple scenario modes, adapting to exclusive operating logic for different scenarios such as office, meeting, reception, rest, energy saving, and duty. Fourth, it allows for precise configuration of equipment adjustment accuracy, refining the accuracy of air conditioning temperature control, lighting brightness, and fresh air volume, achieving micron-level fine control while balancing ultimate comfort and high energy efficiency.

[0132] 6. Independent Room Level: Ultimate Terminal Control, Precise Adaptation to the Smallest Space Unit. The independent room serves as the smallest execution unit in the system's six-level control system, enabling highly refined, point-to-point, and customized control for independent, enclosed small spaces (independent single offices, business rooms, small private offices, cubicles, etc.). Compared to zone-level large-scale control, independent room-level control offers finer granularity, faster response, and stronger adaptability, completely resolving the shortcomings of traditional large-area unified control models, such as poor experience and low energy consumption adaptability in small spaces. Each independent room is equipped with a dedicated independent sensing terminal and intelligent control module, accurately capturing subtle environmental changes, personnel entry and exit status, and usage habits. It supports independent start / stop of equipment, independent adjustment of operating parameters, independent energy consumption statistics, and independent configuration of control strategies for each room. The system can customize exclusive control schemes for each room's users, usage time, and functional positioning, achieving highly refined control with "one strategy per room, one adaptation per room." This maximizes the comfort experience of individual spaces while precisely reducing ineffective energy consumption in individual rooms, achieving efficient energy utilization of the smallest unit.

[0133] III. System-wide Collaborative Control and Operation Mechanism and Core Technologies: This system, through a six-level hierarchical architecture with vertical integration and horizontal linkage, constructs a three-dimensional, multi-layered intelligent energy dispatching system. This system enables autonomous and precise control of room zones, overall balanced management of floors and buildings, and optimal scheduling of the entire park cluster. It fundamentally resolves the industry contradiction of traditional systems where "energy saving inevitably leads to a decrease in user experience, while maintaining user experience inevitably leads to high energy consumption," achieving a win-win situation for both comfort and energy conservation. At the personalized control level, the system fully delegates control authority to lower levels. Each independent zone and room, relying on an AI adaptive closed-loop algorithm, can autonomously optimize its operation based on its own scenario needs and user habits. It flexibly adapts to the usage needs of different spaces, different groups of people, and different time periods, retaining maximum personalized control freedom. This completely abandons the traditional, uniform, and rigid extensive control model of buildings, significantly improving the overall living, working, and residential experience of the park. At the global optimization level, a three-tiered overall scheduling system is formed, consisting of floor-level sub-control units, building-level edge controllers, and the park's cloud platform. This system aggregates load, energy consumption, and equipment operation data from all subdivided spaces across the entire area in real time. Through big data analysis and intelligent collaborative optimization algorithms, it dynamically coordinates the start-up and shutdown sequence, operating power, and load distribution ratio of equipment in each zone and room. Intelligent strategies such as off-peak operation, time-sharing start-up and shutdown, load diversion, and peak load mitigation effectively address a series of problems caused by simultaneous start-up of equipment in multiple areas and simultaneous load surges, including excessive instantaneous peak loads, demand exceeding limits, high peak electricity costs, and large equipment impact loads. Based on a complete set of refined, hierarchical, and collaborative management and control mechanisms, the system achieves three core value upgrades in three dimensions: First, a refined upgrade in the living experience, enabling customized comfort management for different functional spaces and usage scenarios, ensuring constant and balanced environmental parameters to meet the needs of various users; second, an upgrade in efficient energy utilization, effectively smoothing peak electricity consumption, reducing overall energy consumption, and increasing the local consumption rate of clean energy, significantly reducing ineffective energy consumption and energy waste; and third, an upgrade in maximizing operation and maintenance benefits, achieving the lowest overall energy consumption cost, the lowest carbon emissions, and the highest comprehensive energy benefits in the park through optimal energy scheduling across the entire area, helping the park achieve its high-quality operation and maintenance goals of being green, low-carbon, intelligent, efficient, economical, and safe.

[0134] In one embodiment, an AI-powered intelligent control method applied to the aforementioned system constructs a continuous, uninterrupted, and self-iterative closed-loop execution system. The system comprises seven standardized, continuous closed-loop execution steps. The entire process relies on intelligent algorithms, edge computing power, cloud-based big data, and blockchain technology to achieve 100% automated operation. No manual monitoring, intervention, or parameter adjustment is required throughout the process. It can adapt to the intelligent control needs of various scenarios, including building microgrids, energy storage systems, electromechanical equipment, grid interaction, and carbon asset management. This method completely solves the industry pain points of traditional building energy control, such as lagging performance, high energy consumption, weak security, poor adaptability, unreliable carbon data, and insufficient market adaptability. The seven steps of this method form a complete closed-loop chain, from bottom-level data acquisition, mid-level AI intelligent computation, and upper-level resource collaborative scheduling, to end-device execution, grid market interaction, digital twin iterative optimization, and blockchain carbon data storage. These progressively layered, mutually constraining, and continuously iterating steps achieve full-dimensional intelligent closed-loop operation, including microgrid collaborative operation, AI-based pre-judgment, integrated flexible control of multiple devices, electricity market interaction, and reliable carbon data management. The seven standardized continuous closed-loop steps are as follows: S1. Millisecond-level multi-source and multi-dimensional synchronous data acquisition: This step forms the underlying data foundation of the entire intelligent control system. It adopts a parallel high-frequency synchronous acquisition architecture, abandoning the inefficient mode of traditional time-sharing acquisition and batch uploading, and realizes millisecond-level acquisition and real-time transmission of full-dimensional time-series data. It builds a high-precision, high-timeliness, and full-coverage underlying data foundation, providing comprehensive data support for subsequent AI algorithm optimization, safety constraint control, and precise load scheduling. The system synchronously collects multi-dimensional core data, including eight categories of time-series data, covering six dimensions: energy production, energy storage operation, electricity load, grid status, living environment, and personnel status. First, it collects real-time photovoltaic power generation data, frequently monitoring the instantaneous output of photovoltaic modules, power generation fluctuation trends, and cumulative power generation over time periods, accurately capturing the intermittent and fluctuating output characteristics of photovoltaics. Second, it collects a complete set of energy storage BMS operating parameters, including remaining SOC power, individual battery voltage, overall charging and discharging current, real-time battery temperature, and battery balancing status, comprehensively monitoring the operating conditions of energy storage batteries. Third, it collects instantaneous load data from charging piles, capturing real-time instantaneous charging power, charging time, and load fluctuations of new energy vehicles, accurately calculating the dynamic electricity demand of charging piles. Fourth, it collects time-of-use electricity price data, simultaneously capturing electricity price standards and dynamic adjustments for peak, valley, and flat periods. Fifth, it monitors grid voltage, frequency, and power quality parameters in real time to assess grid operational stability and predict grid fluctuation risks. Sixth, it uses millimeter-wave human posture point cloud data to accurately collect indoor human posture, location, and real-time number of people via millimeter-wave radar, enabling dynamic perception of living environments. Seventh, it provides multi-dimensional indoor environmental data, including indoor illuminance, air temperature and humidity, carbon dioxide concentration, PM2.5, TVOC, and formaldehyde concentration, comprehensively monitoring indoor living environment quality. Eighth, it collects energy storage safety parameters continuously around the clock, including a full set of safety fault parameters such as overcharging, over-discharging, overheating, abnormal single-cell voltage, and excessive voltage difference. In terms of data transmission, the system adopts a LoRa / WiFi / 5G three-link redundant transmission architecture, with multiple links working in parallel and serving as backups for each other. All collected data is uploaded to the edge NPU inference engine in real time, eliminating data transmission delays, packet loss, and disconnection issues. Meanwhile, relying on the real-time collection of all-dimensional safety parameters of energy storage, multiple safety constraint logics are embedded into the AI ​​algorithm, and equipment safety and system safety are taken as prerequisites for regulation. This ensures the inherent safety of system operation from the data source and completely avoids problems such as regulation distortion, over-limit operation, and equipment damage caused by data loss or data lag. S2. Edge-Cloud Layered AI Multi-Objective Fusion Optimization Computation: This step adopts a three-layer progressive computing architecture: terminal pre-processing, edge local real-time optimization, and daily cloud-based iterative upgrades. This achieves end-to-end AI intelligent computing, encompassing "front-end prediction, mid-end execution, and back-end optimization," differing from traditional single-cloud computing and lagging control modes. It balances real-time control, scenario adaptability, and algorithm iteration and evolution capabilities. In the terminal pre-processing stage, a lightweight CNN convolutional neural network and an LSTM long short-term memory temporal model are used to achieve collaborative prediction between the two models. The CNN network is specifically used to accurately identify human posture, real-time indoor personnel numbers, and personnel activity areas, accurately matching human comfort needs and avoiding idle energy consumption when unattended and imbalanced control when manned. The LSTM model constructs a 72-hour ultra-long sliding window temporal prediction mechanism, continuously collecting historical environmental data, photovoltaic output data, and grid data to dynamically predict future environmental parameter fluctuations, photovoltaic output changes, and load fluctuation trends. This enables pre-judgment of operating conditions and proactive equipment adjustment, fundamentally solving the problems of energy waste and comfort fluctuations caused by passive control and parameter mutations in traditional systems. In the edge-local real-time optimization stage, relying on the dedicated computing power of the edge NPU, and equipped with a self-developed octet DRL deep reinforcement learning algorithm, the algorithm simultaneously integrates eight core parameters: microgrid operating status, grid time-of-use pricing, energy storage safety constraints, electric shading operating status, indoor environmental parameters, personnel status, equipment load parameters, and grid quality. Through real-time dynamic calculation, it solves for stepless and precise power combinations of lighting, air conditioning, and fresh air systems from 0% to 100%, achieving refined and flexible control of equipment. The system also supports one-click switching between three operating modes: energy-saving priority, comfort priority, and economy priority. Furthermore, it allows for customized weight allocation based on building usage scenarios and user needs, adapting to the control requirements of various scenarios such as office buildings, residences, parks, and commercial complexes. All customized scenario strategies and algorithm parameters are permanently stored in the edge's local flash memory, ensuring that the algorithm's execution effect is indistinguishable from network-connected conditions under extreme operating conditions such as offline network outages and isolated operation, guaranteeing control stability. In the cloud-based iteration and upgrade phase, a fully automated daily iteration and optimization mechanism is adopted. The cloud platform automatically aggregates real-time, multi-dimensional time-series big data to complete offline training of the algorithm model, parameter weight optimization, and defect correction. The optimal algorithm parameters after iteration and upgrade are automatically sent to the flash memory of the edge device for permanent solidification, continuously improving the algorithm's prediction accuracy, optimization efficiency, and scene adaptability, and realizing the autonomous evolution of the algorithm model. S3. Microgrid Source-Load-Storage and Six-Element Flexible Resource Coordinated Scheduling: This step relies on AI calculations to achieve integrated coordinated scheduling of microgrid source, load, and storage core resources with building flexible energy-saving resources. The system can automatically identify five core operating conditions in real time and adaptively match differentiated optimal control logic for different conditions, eliminating the problem of poor adaptability of single control strategies. The five operating conditions accurately identified by the system include: photovoltaic power generation surplus output, grid peak high electricity price and high cost, energy storage low SOC and power loss, microgrid islanding fault off-grid operation, and virtual power plant grid power invitation operation. Based on intelligent operating condition identification, the system constructs a six-element integrated coordinated energy-saving system of photovoltaic-energy storage-electric shading-lighting-air conditioning-fresh air, integrating the dual advantages of passive energy saving and active control to achieve all-dimensional energy-saving optimization. Under extreme conditions of strong sunlight and high temperatures, the system synchronously adjusts the opening of electric shading louvers, the output brightness of indoor lighting fixtures, the cooling load of air conditioning, and the frequency and volume of fresh air exchange. By blocking strong outdoor sunlight radiation and heat from entering through the shading system, the building's heating and cooling load is reduced at the source. Combined with precise power adjustment of electromechanical equipment, the overall energy consumption of the building is significantly reduced. Compared with the traditional single-equipment control mode, the energy efficiency is significantly improved. For microgrid islanding fault conditions, the system has a built-in load level graded smooth load reduction logic. Following the principle of prioritizing important loads and gradually reducing unnecessary loads, it achieves stepless smooth load reduction, completely avoiding drastic fluctuations in grid voltage and frequency, and eliminating equipment overload, tripping, and shutdown faults. When the grid fault is eliminated and grid parameters return to normal, the system adopts a phased gradient load recovery strategy to gradually restore the operating power of various equipment, achieving smooth grid connection without impact, surge, or risk of reconnection, comprehensively ensuring the operational safety and stability of the microgrid, grid, and terminal equipment. S4. Local Three-Layer Redundant Offline Flexible Closed-Loop Execution: This step is the implementation phase of the algorithm control strategy. Relying on the edge controller, it achieves real-time issuance of optimal control commands and precise execution by the devices, ensuring smooth, shock-free, and stepless power adjustment for all end devices. This avoids energy consumption and equipment damage caused by device start-up, shutdown, and sudden power fluctuations. Simultaneously, the system establishes a three-layer, full-scenario fault redundancy protection mechanism, covering all extreme and abnormal operating conditions, ensuring that control is never interrupted, the strategy never fails, and data is never lost. The three-layer redundancy protection mechanism consists of three layers: First, cloud-based disconnection redundancy. When the network between the edge device and the cloud platform is interrupted or data is lost, the local edge computing power can independently run the entire set of algorithm strategies, autonomously completing data collection, AI computation, load scheduling, and equipment control without cloud instructions. Second, wireless link failure redundancy. When the main WiFi / 5G communication link fails or the signal is interrupted, the system automatically and seamlessly switches to the backup LoRa communication link to ensure uninterrupted data transmission and command issuance. Third, public grid power failure redundancy. When the main grid loses power and the system enters islanded operation mode, the microgrid can autonomously complete source-load-storage coordinated scheduling, prioritizing the stable power supply to core loads and achieving safe and stable operation in off-grid conditions. The system has second-level full-state monitoring capabilities, real-time inspection of the operating status of all sensing devices, terminal electromechanical equipment, communication links, and energy storage devices. It can accurately locate faulty devices, fault locations, and fault types, realize graded fault alarms, and automatically push standardized fault handling solutions, significantly reducing the pressure of manual operation and maintenance inspections and the time for fault handling. Meanwhile, after the fault is recovered, the system can automatically retransmit all the original data collected during the offline running period, algorithm operation logs, and equipment control records, so as to achieve no data loss, no operational gaps, and no policy interruptions throughout the process, ensuring the integrity and continuity of the closed-loop operation of the system. S5. Two-way Flexible Load Market Interaction with Virtual Power Plants: This step achieves deep market-based adaptation between building energy systems and the power grid and virtual power plants, breaking away from the traditional passive electricity consumption model of building energy systems. It constructs a two-way flexible market-based interaction system, deeply adapting to modern power grid time-of-use pricing, demand response, electricity spot trading, and virtual power plant power invitation mechanisms, maximizing cost reduction and maximizing energy revenue. The cloud platform monitors the two-way power invitation commands from the power grid dispatch center and virtual power plants in real time, supporting high-precision flexible adjustment of load increases and decreases. During peak power consumption periods, in response to peak load reduction invitations, the system intelligently disassembles the adjustable load capacity of three types of flexible equipment—air conditioning, ventilation, and lighting—to accurately, stably, and orderly reduce peak building power load, responding to power grid dispatch needs and obtaining demand response market subsidies. During off-peak power consumption periods, the system appropriately relaxes the human comfort threshold, reasonably increases the operating power of electromechanical equipment, maximizes the absorption of surplus power during off-peak hours, and simultaneously controls the energy storage system to fully charge and store electricity during off-peak hours, achieving peak-valley price arbitrage. The system dynamically adapts to the grid's time-of-use pricing mechanism, consistently implementing the optimal economic operation strategy of "peak-load reduction, valley-load storage, and flat-load optimization," continuously reducing building electricity purchase costs. Simultaneously, the system possesses automated market-based data statistics capabilities, accurately calculating the adjusted load, adjustment duration, and response accuracy for each demand response. This provides precise data support for electricity market subsidy calculations and electricity transaction settlements, enabling a shift in building energy consumption from "passive consumption" to "active market-based value-added services." S6. Digital Twin Simulation and AI Model Lifecycle Iterative Evolution: This step relies on digital twin technology to construct a virtual-real integrated simulation iteration system, achieving optimal parameter pre-configuration before system commissioning and autonomous evolution throughout the entire lifecycle after commissioning. This completely solves the drawbacks of traditional equipment control parameters being fixed and unable to adapt to changes in operating conditions. In the offline simulation phase, comprehensive basic parameters such as building structure parameters, terminal equipment performance parameters, annual meteorological data, personnel activity patterns, electricity price time-series data, and power grid operation parameters are input to complete 8760 hours of full-time, full-condition simulation throughout the year. Optimal control parameters for different scenarios and operating conditions are pre-generated through massive simulation calculations and batch-fixed and distributed to edge devices, achieving optimal operation from system commissioning without manual debugging or parameter trial and error, significantly shortening the project debugging cycle and reducing commissioning costs. In the online operation phase, a 100ms ultra-high frequency virtual-real synchronization mechanism is constructed to synchronize all operational data of the physical building, equipment, power grid, and environment to the digital twin virtual model in real time, achieving real-time mapping and precise synchronization between the virtual model and the physical entity. The cloud-based training platform summarizes all actual operating data of the system daily, completes iterative training, parameter optimization, and defect correction of AI algorithm models, and continuously improves the model's operating condition prediction accuracy, multi-objective optimization capability, and scenario adaptation capability. It enables the system to learn autonomously and evolve continuously throughout its entire life cycle, ensuring that the control strategy always adapts to changes in the long-term operating conditions of buildings, adjustments to power grid policies, and changes in human living needs. S7. Blockchain-based Full Lifecycle Carbon Accounting and Trusted Evidence Storage: This step constructs an intelligent, refined, and trustworthy building carbon asset management system. Relying on AI carbon accounting algorithms and blockchain evidence storage technology, it achieves accurate, tamper-proof, and standardized compliant output of building energy consumption, carbon emissions, and carbon assets throughout the entire process, adapting to various green building and low-carbon operation needs. The AI ​​carbon accounting unit enables precise carbon accounting by time, item, level, and grade. It can statistically analyze real-time energy consumption and carbon emission data for each electromechanical device, functional area, floor, and building by hour, day, month, and year. Simultaneously, it integrates official national carbon emission conversion factors and implicit carbon data throughout the equipment's lifecycle to automatically generate a refined carbon asset ledger, accurately quantifying carbon emissions and energy savings during the building operation and equipment operation phases. In terms of data storage, the system encrypts all original collected data, AI-controlled operation data, energy consumption data, and carbon accounting data before storing them on the blockchain. Leveraging the decentralized, immutable, and fully traceable characteristics of blockchain, it achieves reliable storage and cross-verification of all core data, eliminating human tampering and data falsification, and ensuring the authenticity, legality, and validity of carbon data. The system features automatic export of standardized carbon reports, capable of outputting multi-dimensional and multi-format carbon verification reports. It fully adapts to various compliance scenarios, including green building star-rating certification, domestic and international green building system certification, corporate ESG information disclosure, carbon sink project development, green credit, and low-carbon park applications, achieving one-stop traceability, quantitative statistics, compliant application, and value transformation of building carbon data. The system continuously executes the above seven closed-loop steps, forming a seamless, self-optimizing, highly secure, high-yield, and traceable fully closed-loop intelligent operation system. This comprehensively achieves the core objectives of stable microgrid collaborative operation, precise AI-based pre-judgment control, flexible energy saving through multi-device integration, value-added interaction in the electricity market, and reliable and compliant carbon data management.

[0135] In one embodiment, based on the aforementioned AI intelligent control method, the corresponding steps incorporate multi-level, full-process, and multi-dimensional security protection, energy-saving optimization, scenario adaptation, and market-oriented adaptation subdivided control logic to achieve refined, standardized, and intelligent management and control of each control link. The specific subdivided logic is as follows: S1. Data Acquisition Stage: This stage constructs a seven-dimensional, full-volume synchronous data acquisition system, covering microgrid operation data, equipment operating status data, indoor environmental data, personnel status data, grid electricity price data, grid quality data, and energy storage safety data. This achieves comprehensive, complete, and timely data acquisition. Through multi-dimensional, full-coverage data acquisition, a refined and highly accurate underlying data foundation is built, completely solving the problems of single data dimensions, delayed acquisition, and missing data in traditional energy regulation systems. This eliminates problems such as AI calculation distortion, regulation strategy failure, and energy consumption management imbalance caused by data deviation and missing data from the source. It provides comprehensive and high-precision data support for upper-level AI multi-objective optimization calculations, multi-dimensional safety constraint regulation, and refined energy-saving scheduling. S2, AI computation stage detailed control logic: This stage follows the core control logic of prioritizing safety, predictive pre-judgment, and offline equivalence. First, it incorporates multiple hard constraint mechanisms for energy storage safety, prioritizing the safety of energy storage devices and microgrid systems. The algorithm dynamically identifies various risks such as overcharging, over-discharging, overheating, excessive voltage difference between individual cells, and abnormal temperatures in real time. Once a safety threshold is triggered, the system immediately and automatically reduces flexible loads such as air conditioning, fresh air systems, and lighting, while simultaneously suspending energy storage charging and discharging, eliminating safety hazards caused by energy storage exceeding limits and ensuring the inherent safety of the system. Second, it is equipped with a continuously iterating LSTM time-series prediction model, which continuously learns historical operating data through a 72-hour rolling window to predict scenarios such as photovoltaic output fluctuations, changes in ambient temperature, and sudden changes in electricity load. This allows for pre-adjustment of equipment in advance, completely solving the problems of energy waste and drastic fluctuations in human comfort caused by the passive control and delayed response of traditional systems. Third, it achieves equivalent adaptation of extreme operating condition strategies. All user-defined scenario strategies and optimal control parameters are permanently stored locally, ensuring that the system algorithm execution logic and control effect are completely consistent with normal network conditions under extreme operating conditions such as offline, network outage, and isolated conditions, guaranteeing the stability and reliability of system operation. S3. Detailed Control Logic of the Coordinated Scheduling Phase: This phase focuses on three core logics: source energy conservation, grid security and stability, and smooth operation. First, there is a dedicated passive energy-saving linkage logic, with a built-in electric shading louver linkage energy-saving sub-process. This deeply integrates passive energy-saving shading technology with active electromechanical control technologies for air conditioning, fresh air systems, and lighting. By prioritizing the blocking of outdoor heat and strong sunlight intrusion through the shading system, it reduces building heating and cooling loads and daylighting energy consumption at the source. Combined with precise control of electromechanical equipment, it maximizes energy-saving effects. Second, there is an islanding operation safety control logic. For grid outage islanding scenarios, it strictly implements a smooth, stepless load reduction strategy according to the load importance level to avoid system oscillations caused by a sudden drop in overall load. Third, there is a grid connection safety recovery logic. After grid fault recovery, a phased gradient load recovery mode is adopted to completely avoid problems such as closing impact, voltage surges, harmonic disturbances, and equipment tripping, comprehensively ensuring the operational safety of terminal equipment, microgrid systems, and the public power grid. S4. Execution Stage Subdivided Control Logic: This stage relies on triple redundancy, full-scenario fault tolerance, and intelligent operation and maintenance subdivided logic to achieve highly reliable operation of the control and execution stage. The three-layer fault redundancy mechanism comprehensively covers all scenarios of faults, including network outages, cloud disconnection, power grid failures, communication failures, and equipment anomalies, ensuring uninterrupted control, policy effectiveness, and data integrity under any fault scenario. Simultaneously, the system is equipped with intelligent fault operation and maintenance logic, which can accurately locate fault points, fault types, and fault impact ranges within seconds, automatically match standardized fault handling solutions, and push them to maintenance personnel. This significantly shortens fault investigation and handling time, eliminates the need for manual inspection and troubleshooting, greatly reduces system manual operation and maintenance costs, improves system fault self-healing and emergency response capabilities, and ensures the integrity and stability of the closed-loop execution system. S5. Market-Oriented Interaction Detailed Control Logic: This section deeply adapts to the power market reform mechanism, constructing a detailed control logic for bidirectional flexible adjustment, maximizing revenue, and minimizing costs. The system accurately adapts to diverse market scenarios such as grid time-of-use pricing, demand response invitations, electricity spot trading, and virtual power plant cluster dispatching. It can adaptively optimize equipment load operation strategies based on real-time grid status and electricity price dynamics. During peak grid periods, it precisely reduces peak load to obtain demand response subsidies and reduce peak-price electricity costs; during off-peak periods, it maximizes the absorption of surplus power and completes arbitrage energy storage during off-peak hours, continuously improving off-peak power utilization, peak-valley arbitrage revenue, and market subsidy revenue, achieving continuous reduction in building lifecycle electricity costs and maximizing the market value of building flexible loads. S7. Carbon Accounting Detailed Control Logic: This section adapts to refined management, group-level adaptation, and full-scenario compliance detailed logic. First, it supports multi-level hierarchical carbon asset management, enabling independent statistics and ledger export of energy consumption and carbon emission data for single rooms, single floors, single buildings, and multi-building clusters within a group, meeting the refined carbon management needs of large commercial parks, industrial parks, and group building complexes. Second, it achieves full-dimensional compliance adaptation; all blockchain-based data and carbon accounting ledgers are fully compatible with domestic green building evaluation standards, international green building certification systems, corporate ESG disclosure guidelines, carbon sink development, green credit, and other compliance scenarios. Third, it realizes carbon value transformation; relying on credible, traceable, and verifiable carbon data, it completes the quantitative statistics, compliant declaration, and market value transformation of building carbon assets, constructing a credible management and control system for the entire process of low-carbon building operation.

[0136] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made without departing from the spirit and scope of the invention, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents. The embodiments of the present invention are for reference only and should not be considered as a basis for limiting the scope of protection of the present invention. Any modifications, equivalent substitutions, or improvements made based on the spirit and principles of the present invention should be considered as included within the scope of protection of the present invention.

Claims

1. A building microgrid-based, AI-adaptive, integrated intelligent flexible energy load control system for lighting, air conditioning, and fresh air systems, characterized in that: The system constructs a standardized, layered architecture from top to bottom: a cloud-based digital twin scheduling layer, an edge-controlled computing power layer, a flexible load equipment layer, a global sensing layer, and a microgrid source and storage equipment layer. These five layers are vertically independent and decoupled, yet horizontally interconnected, enabling full-domain linkage of computing power, load, sensing, source and storage, and scheduling. The system integrates eight mutually decoupled, independently selectable, and modularly pluggable core functional modules. It also adds an auxiliary flexible linkage interface for electric louvers and reserves flexible expansion interfaces for electric sunshades, smart curtains, and building water supply and drainage, forming a multi-source coupled smart energy control system encompassing photovoltaics, energy storage, sunshades, lighting, air conditioning, and fresh air systems. The eight modules include: The distributed microgrid source-load-storage collaborative unit is equipped with a photovoltaic grid-connected inverter, a multi-type energy storage BMS management host, a park AC / DC charging pile controller, a millisecond-level microgrid on-grid / off-grid switching switch, and a source-storage-load synchronous sampling module. The unit incorporates source-load-storage dynamic matching subroutines, power smoothing subroutines, and islanding fault-tolerant control subroutines, and is compatible with three mainstream energy storage devices: lithium iron phosphate, lead-carbon, and vanadium redox flow batteries, as well as energy storage systems of different voltage levels. It collects real-time data on photovoltaic instantaneous power generation, real-time energy storage SOC, individual cell voltage, charging / discharging current, battery temperature, and charging pile dynamics. The system autonomously manages the entire process, including load, equipment operating status, grid time-of-use peak and off-peak electricity prices, grid voltage, frequency, and quality signals, to maximize local photovoltaic power consumption, utilize energy storage for peak-valley charging and discharging arbitrage, manage and control isolated loads with graded fault tolerance, restore grid fault recovery without impact and smooth grid connection, and gradually backtrack power after grid connection. The entire microgrid scheduling logic, security constraint logic, and power optimization logic are all embedded in the edge local flash memory, requiring no continuous cloud-based instructions or cloud computing power support. It can achieve localized, fully functional, autonomous, and stable operation even under grid outages, cloud disconnection, and abnormal grid conditions. The cloud-edge-device full-domain AI adaptive computing architecture constructs a three-tiered distributed heterogeneous computing power system: lightweight inference at the terminal, real-time optimization at the edge, and iterative training in the cloud. The bottom layer consists of a lightweight AI inference unit built into the sensing terminal, responsible for pre-sensing and predicting the environment, personnel, and operating conditions. The middle layer is the NPU local inference engine mounted on the intelligent flexible controller, handling millisecond-level real-time optimal control solutions. The top layer is a cloud-based reinforcement learning iterative training platform, responsible for accumulating large-scale data samples, iterating models, optimizing parameters, and upgrading strategies. This three-tiered computing power system is uniformly connected to a standardized multi-source time-series database, integrating real-time microgrid output, energy storage operation status, grid time-of-use pricing, millimeter-wave human multimodal point cloud data, indoor temperature and humidity, carbon dioxide, PM2.5, TVOC, natural light intensity, outdoor weather time series, and building load. The system performs global collaborative optimization calculations using time-series multi-dimensional parameters, completely eliminating the problems of isolated computing power, data fragmentation, and isolated strategies in traditional building systems. The intelligent flexible central controller serves as the core computing power and control carrier at the system edge, locally embedding a complete set of microgrid collaborative algorithms, multi-load linkage adjustment algorithms, DRL deep reinforcement learning multi-objective optimization algorithms, and fault-tolerant redundancy algorithms. It is equipped with a large-capacity power outage protection flash memory and a power failure data retention mechanism, and has millisecond-level local offline fully autonomous control capabilities. When single or multiple overlapping link failures occur, such as external fiber optic cable disconnection, cloud server disconnection, public power grid power failure, or single communication link interruption, the system can independently execute all flexible control logic in a closed loop, continuously ensuring uninterrupted and uncontrolled adaptive and refined adjustment of lighting, air conditioning, and fresh air, as well as microgrid energy storage collaborative scheduling and load optimization allocation. The integrated flexible load control unit for lighting, air conditioning, and fresh air is equipped with a stepless dimming LED driver module, a variable frequency air conditioning continuous cooling and heating adjustment module, and a fresh air stepless speed regulation airflow adjustment module. The three types of equipment output a unified and standardized 0~100% continuous analog control signal with a minimum adjustment step size accurate to 1%, without fixed levels, graded jumps, or start / stop dead zones. The system unifies the modeling, aggregation, and management of all flexible adjustable loads of lighting, air conditioning, and fresh air throughout the building, integrating them into a standardized, quantifiable, dispatchable, and tradable flexible load resource pool. It outputs continuously smooth and adjustable power, and can accurately match the millisecond-level fine-grained dispatching commands of the power grid and the demand response commands of the virtual power plant, realizing the upgrade of building load from rigid fixed load to flexible adjustable resource. The all-domain multi-dimensional sensing terminal is uniformly equipped with a frequency-modulated continuous wave high-precision millimeter-wave radar, a wide-range illumination acquisition probe, a high-precision temperature and humidity sensor, and a four-in-one air quality detection module for carbon dioxide / PM2.5 / TVOC / formaldehyde. All components utilize a low-power lithium battery power supply and a LoRa long-range wireless communication solution. Retrofitting existing buildings requires no trenching or wiring, no laying of high-voltage control lines, and no modification to existing equipment wiring, making it suitable for rapid deployment in older buildings without modification. The terminal incorporates a lightweight CNN convolutional neural network and a 72-hour sliding window LSTM time-series prediction submodule: the CNN network can accurately identify four core postures: standing while working, sitting while working, lying down while resting, and in an unoccupied space, providing precise output. Real-time parameters for the number of people, density, and duration of stay in each zone; the LSTM model predicts fluctuations in natural light intensity, indoor pollutant concentration trends, and photovoltaic power generation fluctuations based on long-term historical data, enabling proactive prediction and control of environmental and operational changes, eliminating control lag issues; all AI sensing, recognition, and prediction calculations at the terminal are completed locally, without needing to upload to the cloud computing power, and the function is completely unaffected by network conditions such as isolated offline or network outages; the cloud backend supports custom configuration and batch distribution of parameters such as unattended latency, sensitivity thresholds, and comfort zone parameters, which can accurately adapt to different personnel stay patterns and different energy consumption scenarios such as office buildings, hotels, campuses, shopping malls, hospitals, and industrial parks; A three-layer heterogeneous redundant communication network of 5G+WiFi+LoRa is deployed in parallel with industrial 4G or 5G modules, dual-band 2.4G / 5G WiFi modules, and LoR star-shaped networking modules. The three types of communication links have built-in real-time fault monitoring, automatic identification, and seamless switching mechanisms, with no delay and no packet loss during link switching. Among them, LoRa is responsible for low-power, long-distance, and interference-resistant data collection from massive sensor terminals at the bottom layer; WiFi is responsible for local high-speed data interaction and policy synchronization between floor central control and sub-controllers and terminal devices; 5G industrial modules carry long-distance high-speed and reliable transmission of real-time synchronous data of digital twin 3D, millisecond-level scheduling commands of virtual power plants, and power grid demand response interaction commands. The three-layer network has a clear division of labor, serves as a backup for each other, and is collaboratively redundant, comprehensively ensuring the real-time performance and reliability of data transmission at different levels. The digital twin virtual-real mapping unit integrates four sub-modules: offline full-condition simulation before commissioning, millisecond-level online virtual-real synchronization after commissioning, hierarchical fault early warning for all equipment and the entire link, and visualization analysis of operational data. The offline simulation can input multi-dimensional parameters such as building unit structure, window-to-wall ratio, heat transfer coefficient of the building envelope, photovoltaic and energy storage installed capacity, rated power of electromechanical equipment, annual personnel work and rest schedules, hourly weather, and hourly electricity prices. It automatically simulates the microgrid output, load conditions, and energy consumption characteristics for 8760 hours throughout the year, outputting optimal collaborative control thresholds and strategy parameters. The online synchronization module provides 100m... The system synchronizes microgrid power generation, energy storage SOC, real-time power of zoned equipment, environmental parameters, and carbon emission reduction data into a 3D twin model, visually and dynamically presenting the building's energy flow path, load peak shaving and valley filling amplitude, photovoltaic green electricity consumption ratio, and equipment operating status. The fault early warning module collects all types of fault signals in real time, including sensor offline, communication interruption, equipment abnormality, parameter exceeding limits, energy storage fault, and grid abnormality, and pushes alarms in three levels: general, important, and urgent, with precise location, fault cause analysis, standard handling plan, and operation and maintenance suggestions, significantly reducing the pressure of manual inspection and operation and maintenance costs. The virtual power plant aggregation and dispatch unit reserves standardized API interfaces for electricity market access, grid demand response access, and regional energy internet access, enabling seamless integration with regional virtual power plant platforms, grid market-based dispatch platforms, and integrated energy service platforms. It integrates flexible loads such as building-wide lighting, air conditioning, and fresh air into a standardized distributed adjustable resource pool, uniformly quantifying adjustable capacity, adjustable rate, and continuously adjustable duration. It supports participation in various market-based electricity transactions, including grid peak-valley shaving, auxiliary frequency regulation, electricity spot trading, orderly electricity consumption, and demand response compensation, thereby realizing the resource-based, asset-based, and market-oriented management of building loads. This blockchain-based, full-lifecycle carbon data trusted storage unit features a built-in dedicated encrypted node on the consortium blockchain and an AI-powered intelligent accounting subroutine for the carbon footprint of buildings throughout their entire lifecycle. The system integrates regional power grid time-of-use carbon emission factors, photovoltaic zero-carbon offset coefficients, energy storage charging and discharging carbon conversion coefficients, and equipment loss carbon correction coefficients. It accurately calculates the real-time carbon emissions of lighting, air conditioning, and fresh air systems, overlaying the implicit carbon from the entire process of production, installation, transportation, and scrapping of various electromechanical equipment. This constructs a complete full-lifecycle carbon asset ledger covering equipment production, engineering construction, building operation, and equipment decommissioning. All original energy consumption, power generation, energy storage, load regulation, environmental monitoring, and carbon accounting data are asymmetrically encrypted to generate unique hash values ​​for on-chain hash storage. The data is tamper-proof, undeletable, fully traceable, and cross-verifiable. It can automatically generate standardized carbon verification certificates compatible with Green Building Three-Star, LEED, BREEAM, corporate ESG annual reports, CCER carbon sink development, green credit, and green bond applications. Furthermore, it automatically calculates the project's annual carbon emission reduction and potential carbon trading revenue through a built-in real-time carbon trading price dynamic model. The system's eight modules employ a fully decoupled, pluggable modular design for both hardware and software. The distributed microgrid source-load-storage coordination unit, digital twin virtual-real mapping unit, virtual power plant aggregation and scheduling unit, and blockchain-based full-lifecycle carbon data trusted storage unit can all be selected, removed, or expanded as needed to adapt to different renovation and new construction scenarios. For energy-saving renovations of existing buildings without photovoltaics or energy storage, energy saving can be achieved simply by deploying a full-domain multi-dimensional sensing terminal, a cloud-edge-device full-domain AI adaptive computing architecture, a flexible load control unit integrating lighting / air conditioning / fresh air, and a 5G+WiFi+LoRa three-layer heterogeneous redundant communication network. Effectiveness: For newly constructed zero-carbon and near-zero-energy buildings, a complete set of full-function modules can be installed to achieve full-dimensional capabilities such as source-load-storage synergy, AI self-adaptation, digital twin visualization, reliable carbon data management and control, and grid market participation; The system also reserves a dedicated control interface and data interaction interface for the linkage of electric shading louvers, allowing the electric louvers to be connected to the edge control as a passive energy-saving flexible resource, and constructing a six-element synergistic energy-saving system of photovoltaic-energy storage-electric shading-lighting-air conditioning-fresh air. By dynamically adjusting the louver angle in conjunction with light, temperature, and operating conditions, the solar radiation heat gain of the building is reduced from the source, significantly reducing the basic load of air conditioning in summer.

2. The system according to claim 1, characterized in that: The distributed microgrid source-load-storage collaborative unit is equipped with a dedicated microgrid on-grid / off-grid switching controller. It incorporates a hardware-level millisecond-level islanding detection circuit and a voltage and frequency distortion identification algorithm, achieving an islanding voltage and frequency anomaly detection latency of ≤10ms. This allows for accurate identification of real islanding, instantaneous grid fluctuations, and load surges, preventing misjudgments and incorrect switching. When a public grid failure causes a power outage or voltage / frequency exceeding limits, leading to islanding mode, the cloud-edge-device full-domain AI computing power architecture automatically activates a preset tiered flexible load reduction strategy, classifying the building load into a three-tiered load system: Level 1 basic guarantee load, Level 2 office routine load, and Level 3 non-essential public area load. Level 1 basic guarantee load operates continuously at full power, ensuring basic lighting, computer room temperature control, basic ventilation, and core security monitoring loads. Uninterrupted operation; Level 2 and 3 loads are not directly cut off or instantaneously shut down, but are smoothly and steplessly reduced in power and load by integrated flexible units according to priority, achieving shock-free load reduction; surplus photovoltaic power generation is preferentially stored in energy storage batteries for local consumption; when the energy storage SOC is lower than the system's preset safety threshold, the AI ​​engine automatically and adaptively tightens the indoor temperature, humidity, illuminance, and air quality comfort range, further reducing equipment operating power and strictly controlling energy consumption while ensuring basic living experience; after the grid voltage, frequency, and phase parameters are restored to grid connection standards, the system gradually and gradually increases the load power in multiple stages, simultaneously completing precise phase matching of the microgrid, with no voltage impact or current surge throughout the process, and no need for manual on-site operation, achieving fully automatic and imperceptible grid connection restoration; The integrated flexible load unit for lighting / air conditioning / fresh air adopts a smooth stepless continuous power regulation mode from 0 to 100% throughout the entire process. This completely avoids the instantaneous high current impact caused by the frequent switching of relays, repeated start-stop of compressors, and fan speed jumps in traditional equipment, as well as the harmonic disturbances of the power grid and wear and tear on the mechanical structure. It effectively extends the service life of core equipment such as LED lighting fixtures, air conditioning compressors, and fresh air fans, and significantly reduces the energy consumption and failure probability caused by the frequent start-stop of central air conditioning units. The system is equipped with an intelligent operation and maintenance early warning module in the cloud, supporting real-time alarm pushes on both mobile app and PC web management terminal. All fault information is accompanied by a precise building, floor, and room location QR code and a unique equipment code. Operation and maintenance personnel can scan the code to accurately locate the fault point, retrieve equipment files and historical operation and maintenance records, and completely reduce ineffective inspections and blind troubleshooting. The system supports customizable fault push blocking time periods and fault level push rules. Low-priority faults at night and on holidays can be pushed and handled centrally on weekdays, balancing the rest rights of operation and maintenance personnel with the timeliness and standardization of fault handling. The cloud-based backend of the full-domain multi-dimensional millimeter-wave AI sensing terminal supports batch configuration and one-click distribution of unattended latency parameters, sensing sensitivity, and human recognition thresholds. It can set differentiated departure latency strategies for different functional scenarios such as hotel rooms, independent office areas, open office areas, conference rooms, and corridors, accurately adapting to diverse human activity patterns. All LoRa sensing terminals have built-in intelligent low-power sleep and wake-up mechanisms, with high-frequency data collection when people are present and low-frequency sleep when people are absent, significantly extending device battery life and reducing terminal maintenance frequency while ensuring sensing accuracy.

3. The system according to claim 1, characterized in that: The intelligent flexible controller's control logic engine incorporates a DRL (Deep Reinforcement Learning) octet optimization algorithm specifically designed for complex scenarios involving strong coupling of source, load, and storage in building microgrids, multiple interferences, and nonlinearity. Unlike standard reinforcement learning algorithms, it specifically adds source-storage constraints, equipment lifespan constraints, human comfort constraints, and grid quality constraints. The algorithm constructs a four-fold weighted constraint reward function: a positive economic reward for minimizing monthly electricity purchase costs, a negative penalty for exceeding indoor environmental parameter thresholds, a positive reward for extending equipment lifespan without frequent start-stop cycles, and a low-carbon positive reward for improving photovoltaic green energy consumption. The four reward weights are adjustable online and offline, and can be fixed in effect locally. Users can switch between three core operating modes—priority energy saving, priority human comfort, and priority green energy consumption—with a single click. Custom weight allocation is also supported to form personalized operating strategies. The DRL algorithm has the ability to iterate online and train offline around the clock: the cloud-based digital twin platform automatically summarizes the big data samples of the previous day's full microgrid power generation, energy storage operation, personnel flow, environmental time series, electricity price fluctuations, and load changes, and completes offline model iterative training and neural network weight optimization. The optimized neural network weight parameters are automatically encrypted and sent to the edge NPU inference engine of each floor for permanent fixation, continuously reducing the building's overall energy consumption, increasing the photovoltaic self-consumption rate and energy storage arbitrage income. The longer the algorithm iteration cycle and the more data accumulated, the more significant the system's global optimization effect, adaptability, and energy-saving benefits. The algorithm incorporates six standardized preset scenario templates: weekday office work, lunch break, nighttime security, holiday hibernation, meeting reception, and hotel room sleep, covering all normal operating conditions of the building at all times. It also supports drag-and-drop visual customization, editing, adding, modifying, and deleting exclusive scenario strategies. Custom scenario logic is permanently stored in edge flash memory, and can be called and executed normally and stably even when the cloud is disconnected, the network is interrupted, or the microgrid is isolated mode, without being affected by the status of the upper-level system. The algorithm also incorporates a scene intelligent recognition subroutine, which can automatically match the optimal scenario strategy based on time, personnel, lighting, and load, and achieve unattended adaptive switching.

4. The system according to claim 1, characterized in that: The distributed microgrid source-load-storage coordination unit and the lighting / air conditioning / fresh air flexible control unit achieve complete interconnection of underlying data, deep linkage of top-level algorithms, and coordinated matching of operating conditions and strategies, and execute refined and differentiated coordinated scheduling strategies for four typical extreme operating conditions of building operation. Operating Condition 1: Solar power generation surplus at midday; The AI ​​engine actively and synchronously reduces the power of zoned lighting to reduce the indoor heat increase caused by the heat generated by the lamps, thereby reducing the air conditioning cooling load from the source; The reasonable range of air conditioning cooling temperature is appropriately relaxed to avoid excessive cooling and waste, maximizing the local consumption of real-time solar power generation. The surplus electricity that cannot be consumed locally is automatically stored in the energy storage battery to complete green electricity storage, achieving nearly 100% local consumption of solar power generation. Operating Condition 2: Peak electricity prices and high demand conditions in the power grid; the microgrid unit controls the energy storage to discharge and replenish energy at high power levels, replacing the high-priced electricity in the power grid; synchronously linking stepless dimming and power reduction of lighting, precise cooling of air conditioning, and speed reduction of fresh air ventilation, steplessly and smoothly reducing the instantaneous peak load of buildings, effectively reducing electricity costs during peak periods, lowering the maximum demand declaration value of the power grid, avoiding demand penalties, and significantly improving the energy economy of buildings; Operating Condition 3: Insufficient photovoltaic output, cloudy and rainy weather with weak light, and low SOC and low power consumption of energy storage; The AI ​​engine automatically and adaptively tightens the indoor comfort constraint threshold, compressing the air conditioning temperature and humidity adjustment range, limiting the maximum brightness threshold of lighting, and reducing the basic fresh air supply volume while ensuring the basic living experience, accurately and strictly controlling the total building power load, minimizing the amount of electricity purchased from the grid, and reducing the grid electricity cost outside of off-peak hours; Operating Condition 4: Virtual power plant grid load reduction invitation and demand response triggering condition; The system constructs a three-dimensional dynamic adjustable load capacity matrix of lighting, air conditioning and fresh air in real time, accurately calculates the adjustable upper limit, adjustable rate and sustainable adjustment duration of the three types of equipment, and intelligently and optimally allocates the adjustment ratio of the three types of equipment according to the grid target power reduction, and outputs continuous, stable and controllable standardized adjustable load resources to meet the grid demand response accuracy requirements; The system provides real-time statistics on the adjustable load capacity, adjusted capacity, and remaining adjustable capacity across the entire building area within seconds, and periodically reports these statistics to the virtual power plant dispatch platform with encryption. This forms a standardized, flexible load resource ledger that is dynamically updated, traceable, and tradable, supporting regional unified dispatch of the power grid, market-based trading, and demand response subsidy settlement.

5. The system according to claim 1, characterized in that: The digital twin virtual-real mapping unit is configured as follows: The offline simulation submodule supports custom input of multi-dimensional refined parameters such as building envelope heat transfer coefficient, window shading coefficient, wall insulation parameters, photovoltaic energy storage installation parameters, electromechanical equipment rated parameters, personnel work and rest patterns, hourly electricity price, and annual meteorological and solar radiation time series. The simulation engine is based on a fusion algorithm of building energy consumption mechanism and AI big data model to automatically complete 8760 hours of hourly dynamic simulation of all working conditions throughout the year. It accurately outputs a complete set of quantitative data, including the building's annual power consumption, photovoltaic green electricity absorption rate, grid peak load, monthly electricity cost, hourly carbon emission reduction, and optimal equipment control threshold. Operation and maintenance personnel can send the optimal simulation parameters to all edge controllers and terminal controllers in batches with one click, eliminating the tedious process of traditional manual debugging, repeated trial and error, and experience parameter settings, and significantly shortening the system debugging cycle. The online virtual-real synchronization submodule completes a full synchronization update of equipment, energy consumption, energy storage, environment, and carbon data every 100ms. The 3D twin interface displays real-time operating power of equipment, real-time output of photovoltaic power generation, remaining energy storage capacity and charging / discharging status, indoor environmental parameters, and hourly carbon emission reduction data in four hierarchical levels: building, floor, zone, and individual room. It supports year-on-year, month-on-month, and time-period comparative visualization analysis of energy consumption, power generation, and carbon data. Managers can remotely, intuitively, and accurately grasp the energy flow path, load control effect, and energy-saving and low-carbon benefits of the entire building. The digital twin platform automatically aggregates big data from the entire operation every day and sends it to the cloud-based DRL training platform to complete offline iterative optimization and parameter upgrades of the model. The updated optimal algorithm parameters are automatically encrypted and sent to the NPU computing chips of each central controller, enabling the system to learn autonomously throughout its entire life cycle, continuously evolve, and become better with use. The longer the building is in operation, the more prominent the system's energy-saving effect, green electricity absorption efficiency, and comprehensive carbon emission reduction benefits become, completely breaking through the technical drawbacks of traditional building automation systems, such as fixed logic, fixed parameters, and no evolutionary ability.

6. The system according to claim 1, characterized in that: The cloud-based AI-powered carbon footprint calculation unit can dynamically calculate real-time carbon emissions during the building's operation phase at multiple time scales, including minutes, hours, days, and months. It automatically and accurately distinguishes between carbon emissions from grid-connected thermal power purchases, zero-carbon electricity offsets from photovoltaic power, carbon conversions from energy storage charging and discharging, and additional carbon emissions from equipment wear and tear. Simultaneously, it supports batch input of implicit carbon data from the entire process of manufacturing, logistics, engineering installation, and waste disposal of various electromechanical equipment, overlaying annual carbon reduction data during the building's long-term operation period to construct a carbon asset ledger covering the entire lifecycle, all dimensions, and all time series from equipment production to building decommissioning. The system is equipped with a dedicated consortium blockchain encrypted storage node. All minute-level raw energy consumption, photovoltaic power generation, energy storage charging and discharging, environmental monitoring, load regulation, and carbon accounting details are asymmetrically encrypted to generate unique, non-repeatable hash values ​​and stored on the blockchain in real time. The data is tamper-proof, non-deletable, fully traceable, and cross-verifiable, possessing judicial, financial, and certification-level credibility. The cloud-based system supports the automatic generation and export of standardized PDF carbon verification reports and carbon asset statements by year, quarter, month, and custom time period, accurately adapting to five core application scenarios: green building star rating certification, LEED international green building certification, corporate ESG annual report disclosure, CCER carbon sink project development, green finance credit, and green bond application. It also features a built-in real-time carbon trading average price dynamic calculation model, which can automatically and accurately calculate a project's annual carbon emission reduction, potential carbon trading revenue, and energy-saving and carbon-reduction economic value based on real-time prices in the national carbon market, providing comprehensive quantitative data support for the refined, market-oriented, and value-driven operation of building carbon assets.

7. The system according to claim 1, characterized in that: The intelligent flexible controller serves as the core hardware carrier at the system edge. It highly integrates LoRa long-distance communication modules, dual-band WiFi high-speed modules, 5G full-network industrial modules, multi-core NPU high-speed computing chips, large-capacity power failure protection flash memory, multiple high-precision analog / digital input / output interfaces, and Ethernet wired backup communication interfaces, forming a 5G+WiFi+LoRa+wired Ethernet quadruple communication redundancy guarantee system to prevent system control failure and data interruption caused by a single communication failure. The central controller's large-capacity power-loss protection flash memory can permanently store custom scenario strategies, algorithm weight parameters, historical energy consumption data, sensor monitoring data, microgrid operation raw data, and fault log data. It will not lose or fail when power is off. In the event of any communication link failure, cloud disconnection, or network fluctuation, it can independently and completely run the entire set of AI algorithms, microgrid scheduling strategies, and load flexible control logic locally, and continuously complete millisecond-level high-precision flexible adjustment. After the communication link is restored, the system automatically and intelligently identifies the offline period and completely retransmits all offline raw data and operation logs to the cloud platform and blockchain node, without data loss, computational gaps, or policy interruptions. A single intelligent flexible controller can concurrently and stably connect up to 200 LoRa sensor terminals and end devices, with a full-domain data polling cycle of ≤100ms and a local NPU multi-target AI inference and optimization operation latency of ≤50ms. It fully meets the ultra-high timeliness requirements of microgrid on-grid and off-grid switching, virtual power plant millisecond-level scheduling, and real-time building load optimization. It can efficiently adapt to networking and control scenarios with massive points and wide coverage in large industrial parks, commercial complexes, and multi-building groups.

8. The system according to claim 1, characterized in that: The system adopts a six-level hierarchical, finely distributed management and control architecture: park-multiple buildings-single building-floor-independent zone-independent room. Each level of the architecture operates independently, with real-time data exchange and global collaborative linkage. The park's cloud platform enables centralized monitoring of multiple buildings, batch policy distribution, group-level carbon data aggregation, unified energy consumption analysis, and unified statistics of trading revenue. The edge controller of each building is responsible for the power balance of the microgrid's source, load, and storage, global load optimization, and security management of the entire building. Floors are configured with domain-specific sub-control units, dividing open-plan office areas, independent offices, meeting rooms, guest rooms, and public corridors into independent management and control zones. Each sub-zone is equipped with a dedicated sensing network and flexible control terminals to achieve fine-grained management and control of the zones. Each independent zone can autonomously execute AI adaptive closed-loop control, supporting customized exclusive living comfort thresholds, unattended delay strategies, scene operation modes, and equipment adjustment precision; the floor and building central controllers coordinate the overall load balance, intelligently coordinate the staggered operation and staggered start-up and shutdown of each zone, effectively eliminating the problem of simultaneous load surges and excessive instantaneous peaks in multiple zones, and achieving optimal energy scheduling, lowest energy consumption, and maximum benefits for the entire building while ensuring personalized living comfort in each zone.

9. An AI intelligent control method applied to the system described in any one of claims 1 to 8, characterized in that: This method is a continuous, uninterrupted, self-iterative closed-loop execution system that operates around the clock, comprising seven standardized continuous closed-loop steps. It runs fully automatically without human intervention. S1. Millisecond-level multi-source multi-dimensional data synchronous acquisition: The system collects real-time photovoltaic power generation, a complete set of energy storage BMS operating parameters (SOC, cell voltage, charging and discharging current, battery temperature, equalization status), instantaneous load of charging piles, grid time-of-use electricity price, grid voltage frequency quality, millimeter-wave human posture point cloud data, indoor illuminance, temperature and humidity, carbon dioxide, PM2.5, TVOC, formaldehyde multi-dimensional time-series data in parallel at high frequency, and uploads them to the edge NPU inference engine in real time through LoRa / WiFi / 5G redundant links; it also collects a complete set of safety parameters for energy storage overcharging, over-discharging, over-temperature, and voltage difference anomalies around the clock, providing multiple safety constraints for AI algorithms and ensuring the inherent safety of system operation; S2, Edge-Cloud Layered AI Multi-Objective Fusion Optimization Computation: Divided into three progressive processes: terminal pre-processing, edge local real-time optimization, and cloud daily iterative upgrades; The terminal CNN network accurately identifies human posture and real-time personnel count, and the LSTM 72-hour sliding window time series model predicts environmental parameter fluctuations and photovoltaic output changes, achieving pre-condition prediction and proactive adjustment; The edge NPU loads a dedicated eight-tuple DRL deep reinforcement learning algorithm, integrating multi-dimensional parameters such as microgrid status, time-of-use electricity price, energy storage safety constraints, and electric shading status, to solve for the optimal stepless power combination of lighting, air conditioning, and fresh air from 0 to 100% in real time, supporting one-click switching of three operating modes and custom weight adaptation; The cloud automatically summarizes the full amount of time series big data daily to complete offline model training and weight optimization, and the upgraded algorithm parameters are automatically sent to the edge flash memory for permanent fixation, ensuring stable and consistent execution of strategies under offline network outage and isolated operating conditions. S3, Microgrid Source-Load-Storage and Six-Element Flexible Resource Co-scheduling: The algorithm automatically identifies five core operating conditions in real time: high photovoltaic power generation, high grid peak electricity price, low energy storage SOC, microgrid islanding fault, and virtual power plant power invitation, and adaptively executes differentiated optimal control logic; it links electric shading louvers to construct a six-element active and passive integrated collaborative energy-saving system of photovoltaic-energy storage-shading-lighting-air conditioning-fresh air, and synchronously adjusts the opening of shading louvers, the output brightness of lamps, and the cooling load of air conditioning under strong light and high temperature conditions, reducing building heat gain and cooling energy consumption from the source; in islanding mode, it smoothly and steplessly reduces load according to load level, avoiding grid voltage and frequency fluctuations and equipment tripping; After the grid parameters are restored, the load will gradually increase in stages to achieve a smooth grid connection without impact, surge, or risk of closing. S4. Local three-layer redundant offline flexible closed-loop execution: The edge controller sends the optimal stepless adjustment command obtained from the solution to the end device in real time, realizing smooth and shock-free power adjustment of the device; The system has a built-in three-layer fault redundancy protection mechanism: When the cloud connection is lost, the complete set of algorithms and strategies will run independently locally; Wireless link failure will automatically and seamlessly switch to the backup LoRa link; Microgrid will autonomously coordinate and schedule power supply in the power outage island mode of the public power grid; The system monitors the operating status of all devices, sensors and communications in real time at the second level, and accurately locates abnormal faults, issues graded alarms and pushes handling solutions in an instant; After the fault is recovered, all original data and operation logs of the offline period will be automatically and completely retransmitted, with no data loss, no computational gaps and no strategy interruptions throughout the process; S5, Virtual Power Plant Two-Way Flexible Load Market Interaction: The cloud monitors the two-way power invitation commands between the power grid and the virtual power plant in real time, supporting two-way flexible adjustment of load increase and load decrease; when the power grid invites peak loads, it intelligently disassembles the adjustable capacity of three types of equipment to stably and accurately reduce the peak load of buildings. During off-peak hours, the threshold for human comfort is appropriately relaxed and the operating power of equipment is reasonably increased to maximize the absorption of surplus electricity during off-peak hours. Simultaneously, the energy storage system is fully charged during off-peak hours to complete arbitrage storage. The system dynamically matches the grid's time-of-use pricing strategy, optimizes peak load reduction, off-peak energy storage, and flat load, continuously reduces the cost of electricity purchase for buildings, and automatically calculates the demand response adjustment amount and duration for accurate market-based subsidy calculation. S6. Digital Twin Simulation and AI Model Full Lifecycle Iterative Evolution: Through offline simulation, all parameters of buildings, equipment, weather, personnel, and electricity prices are entered to complete 8760 hours of full-time-series operating condition simulation throughout the year. The optimal control parameters are pre-generated and distributed to edge devices in batches to achieve optimal performance upon commissioning. In online mode, all data is synchronized between virtual and real systems every 100ms. The daily summary of operating data is sent to the cloud training platform to complete model iteration and optimization, continuously improving the algorithm's prediction accuracy and optimization capabilities, and realizing the system's autonomous learning and continuous evolution throughout its entire lifecycle. S7. Blockchain-based full lifecycle carbon accounting and trusted evidence storage: The AI ​​carbon accounting unit accurately counts the energy consumption data of each device and each zone in a time-sharing and item-by-item manner, integrates the official carbon conversion factor and the implicit carbon data of the device's full lifecycle, and generates a refined carbon asset ledger. All original operational data, control data, and carbon data are encrypted and stored on the blockchain to ensure immutability, full traceability, and cross-verification. The system supports the automatic export of standardized carbon verification reports in multiple dimensions and formats, fully meeting various compliance requirements for green building certification, ESG disclosure, carbon sink development, and green credit. The system executes the above seven steps in a 24 / 7 loop to achieve fully closed-loop intelligent operation, including microgrid collaboration, AI-based pre-judgment, integrated flexible control of the three devices, market-based interaction, and reliable carbon data management.

10. The AI ​​intelligent control method according to claim 9, characterized in that: The corresponding steps incorporate multi-level, end-to-end, and multi-dimensional security protection, energy-saving optimization, scenario adaptation, and market-specific control logic, as detailed below: S1. The data acquisition process enables the simultaneous collection of seven dimensions of data, including microgrid, equipment, environment, personnel, electricity price, grid quality, and energy storage safety. This constructs a complete, detailed, and timely underlying data foundation, providing comprehensive data support for upper-level AI optimization and safety constraint control, and preventing control distortion caused by data gaps. The S2 AI computing stage incorporates multiple hard constraint mechanisms for energy storage safety, dynamically identifying risks such as overcharging, over-discharging, overheating, and excessive individual cell pressure differences in real time. Once a safety threshold is triggered, it prioritizes automatically reducing flexible loads and suspending energy storage charging and discharging, prioritizing equipment and system safety. The LSTM time-series model continuously iterates and predicts operating conditions, performing pre-adjustment actions in advance, completely resolving energy waste and fluctuations in human comfort caused by lagging control and parameter mutations in traditional systems. All custom scenario strategies are permanently stored locally, ensuring consistent performance under offline, offline, isolated, and other extreme conditions compared to networked conditions. S3, the collaborative scheduling process incorporates a dedicated energy-saving subprocess linked to electric shading louvers, achieving a deep integration of passive shading energy saving and active electromechanical equipment control, reducing building heating and cooling loads from the source; through islanded load tiered smooth load reduction and grid-connected phased logic, it effectively avoids grid voltage fluctuations, equipment tripping, closing impacts, and harmonic disturbances, ensuring the safety of equipment and grid operation in all aspects. S4. The execution phase has triple fault redundancy and seamless switching capabilities, covering all fault scenarios such as network outage, disconnection, power grid failure, and equipment malfunction, ensuring uninterrupted control, unbroken strategies, and no data loss; combined with accurate fault location and standardized handling plan push, it significantly reduces the cost of manual operation and maintenance inspection and the time for fault handling. S5, the market-oriented interaction link supports two-way high-precision power regulation of virtual power plants, deeply adapts to the grid time-of-use pricing, demand response, and electricity spot trading mechanism, dynamically optimizes load operation strategy, maximizes the off-peak power consumption rate, peak-valley arbitrage income and demand response subsidy income, and continuously reduces the electricity cost of the entire life cycle of buildings. S7, the carbon accounting process supports the export and statistics of multi-level hierarchical carbon ledgers for single rooms, single floors, single buildings, and multiple buildings in a group, adapting to the refined carbon asset management needs of group-based building clusters; All on-chain data is fully compatible with domestic green building standards, international green certification, ESG disclosure, carbon sink development, and green finance credit compliance requirements, enabling one-stop trusted traceability, quantitative statistics, compliant declaration, and value transformation of building carbon data.

Images