Building energy saving management method, equipment and medium

Abstract

The invention discloses an energy-saving management method and equipment for building energy and a medium, belongs to the technical field of building energy saving, and aims to solve the technical problem that an existing building energy management system has remarkable defects in the aspects of real-time sensing, self-adaptive control, multi-region coordination and multi-target balance. The method comprises the following steps: performing node monitoring processing on each independent energy consumption unit in a building to obtain node energy consumption data stored in a cloud platform; performing hierarchical storage on the node energy consumption data to obtain a building energy spatio-temporal data model; performing deviation degree calculation between actual energy consumption and dynamic reference energy consumption in the building energy efficiency level to obtain an energy efficiency deviation evaluation result; generating a hierarchical feedback strategy based on an energy efficiency deviation evaluation result; and according to the hierarchical feedback strategy, performing hierarchical optimization processing on the current building energy control strategy through a hierarchical architecture of a decision-making layer to obtain a building energy saving strategy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building energy saving, and in particular to a building energy saving management method, device and medium. BACKGROUND

[0002] The building sector is one of the main sources of global energy consumption and carbon emissions, contributing about 40% of global energy consumption and 30% of CO2 emissions. Promoting the transformation of building energy systems (BES) to high efficiency and low carbon has become a key to addressing environmental crises. Currently, building energy management systems mostly use single strategies such as rule control, model predictive control or reinforcement learning, but still face multiple technical bottlenecks in actual operation, making it difficult to balance energy efficiency improvement, user comfort and system flexibility.

[0003] The existing control method has significant technical bottlenecks: first, rule control relies on fixed thresholds (such as HVAC start-stop temperature settings), which cannot dynamically respond to uncertainties such as weather changes and personnel flow, easily leading to fluctuations in indoor environmental parameters beyond the comfort range, or falling into the dilemma of "energy saving and discomfort"; second, model predictive control (MPC) can improve adaptability through rolling optimization, but it relies too much on accurate physical models, and the control parameters are fixed, which easily leads to model mismatch when facing nonlinear characteristics such as building thermal inertia and equipment coupling, and it is difficult to handle the mixed action space of discrete-continuous in multiple areas (such as the coordination of device start-stop and power regulation); third, single-agent deep reinforcement learning (DRL) can learn through environmental interaction, but it mostly focuses on single-objective optimization (such as only reducing energy consumption), lacks multi-area coordination mechanism, and often relies on penalty functions to handle constraints during the exploration process, leading to biased negative training samples, slow convergence, and even safety risks such as battery overcharging / overdischarging and indoor CO2 exceeding standards.

[0004] Therefore, in view of the problems of multi-area coordination and mixed action space, a multi-agent decision mechanism needs to be introduced to make distributed decisions by having each agent take charge of different areas or devices, realizing hierarchical coordination of "macroscopic mode selection - microcosmic parameter adjustment", which effectively handles the mixed action space of discrete-continuous in multiple areas (such as the coordination of device start-stop and power regulation), and solves the problem of energy scheduling between areas through information interaction between agents, avoiding the adaptability limitations of centralized control and improving the response ability of the system to dynamic working conditions in multiple areas.

[0005] In addition, since the conventional method relies on a penalty function to handle hard constraints, not only does it result in a high proportion of negative samples during training, delaying convergence, but it can also prompt the agent to slightly violate the constraints in pursuit of the energy consumption target. Therefore, the application needs to introduce a safety reward engineering (SRE) mechanism to ensure that the hard constraints are met during both the training and running phases through the dual design of positive incentives and real-time action correction. On the one hand, positive incentives are given to behaviors that meet the constraints, guiding the agent to voluntarily comply; on the other hand, a safety layer is added to monitor and correct illegal actions in real time, shifting from "post-punishment" to "prevention", solving the problem of balancing safety and optimization efficiency in the conventional method.

[0006] In summary, the existing building energy management system has significant shortcomings in real-time sensing, adaptive control, multi-region coordination, and multi-objective balancing. There is an urgent need for an intelligent management scheme that combines multi-agent decision-making and safety reward engineering (SRE) mechanisms to address the shortcomings of traditional methods in mixed action space, multi-region coordination, and safety. SUMMARY

[0007] The embodiments of the application provide a building energy saving management method and device and medium, which are used to solve the technical problem that the existing building energy management system has significant shortcomings in real-time sensing, adaptive control, multi-region coordination, and multi-objective balancing.

[0008] The embodiments of the application adopt the following technical solutions: On the one hand, the embodiments of the application provide a building energy saving management method, comprising: performing node monitoring processing on each independent energy consumption unit in the building through a multi-node monitoring network to obtain node energy consumption data stored in a cloud platform; performing hierarchical storage on the node energy consumption data through a pre-set building hierarchical data structure to obtain a building energy space-time data model; calculating the deviation between the actual energy consumption and the dynamic benchmark energy consumption in the building energy efficiency level according to the building energy space-time data model to obtain an energy efficiency deviation evaluation result; generating a hierarchical feedback strategy based on the energy efficiency deviation evaluation result; and performing hierarchical optimization processing on the current building energy control strategy according to the hierarchical feedback strategy and through a hierarchical architecture of a decision layer to obtain a building energy saving strategy.

[0009] The embodiments of the present application ensure that the indoor environmental parameters (temperature, humidity, CO2, concentration) of each region are always maintained within the healthy and comfortable range through multi-region collaborative control and refined energy scheduling. In a park-level energy system, its modular architecture supports plug-and-play and coordinated optimization of multiple types of energy devices such as photovoltaic power generation, battery energy storage, heating, ventilation and air conditioning, and significantly improves the overall energy self-sufficiency rate through source-storage-load collaborative scheduling. Meanwhile, it also provides personalized energy consumption feedback, automated energy-saving reports and intelligent voice interaction functions, which can effectively solve the problem of low user participation. For special building types such as industrial plants and data centers, the high reliability design and safety constraint protection mechanism of the system ensure that the maximum energy-saving potential is tapped on the premise of meeting strict process requirements.

[0010] In a feasible implementation, each independent energy consumption unit in the building is monitored by a multi-node monitoring network to obtain node energy consumption data stored in a cloud platform, specifically including: identifying each independent energy consumption unit in the building; deploying a monitoring node to each independent energy consumption unit through the multi-node monitoring network; collecting energy consumption data of each independent energy consumption unit through the monitoring node to obtain energy consumption collection data; wherein the energy consumption collection data at least includes: current monitoring data, voltage monitoring data, microcontroller cumulative power consumption data and energy consumption power data; protocol interconnection between the monitoring node and the cloud platform; through the cloud platform, the energy consumption collection data is normalized stored under relevant space-time alignment to obtain the node energy consumption data.

[0011] In a feasible implementation, the node energy consumption data is stored in a hierarchical manner through a preset building hierarchical data structure to obtain a building energy space-time data model, specifically including: hierarchical division of data attribute items of a cloud database of building documents to determine the building hierarchical data structure; wherein the building hierarchical data structure includes: park data, floor data, building data and device data; according to the building hierarchical data structure, the node energy consumption data is stored in a JSON document format, and the node energy consumption data is cross-device associated through the node ID and timestamp of the monitoring node to establish the building energy space-time data model for multi-source data fusion.

[0012] In an embodiment, the deviation degree between the actual energy consumption and the dynamic benchmark energy consumption in the building energy efficiency level is calculated according to the building energy characteristics in the building energy space-time data model, and an energy efficiency deviation evaluation result is obtained, which specifically includes: the historical building energy characteristics in the building energy space-time data model are grouped and divided according to the building itself characteristics and the building external environment characteristics through a K-means clustering algorithm, and a historical energy intensity region is obtained; the energy intensity index of the historical energy intensity region is calculated to obtain historical actual energy consumption data; the historical actual energy consumption data is compared with the contemporaneous building itself data and the contemporaneous building external environment data to obtain historical reference benchmark data; the actual energy consumption data in the current energy intensity region is obtained based on the current building energy characteristics in the building energy space-time data model; the actual energy consumption data at the last time and the historical reference benchmark data at the last time are calculated and processed according to a preset smoothing factor, and the dynamic benchmark energy consumption data at the current time is obtained by combining the historical reference benchmark data at the current time; the dynamic benchmark energy consumption data and the actual energy consumption data at the current time are calculated by ratio to obtain an energy consumption deviation degree; and the energy efficiency deviation evaluation result of each energy intensity region is obtained by comparing the energy consumption deviation degree with a standard threshold.

[0013] In an embodiment, a hierarchical feedback strategy is generated based on the energy efficiency deviation evaluation result, which specifically includes: the energy efficiency deviation evaluation result is classified by deviation type; if the deviation type is device-level abnormality, the hierarchical feedback strategy is a device maintenance feedback strategy; if the deviation type is behavior-level abnormality, the hierarchical feedback strategy is a behavior adjustment feedback strategy; and if the deviation type is time-level abnormality, the hierarchical feedback strategy is an automatic adjustment feedback strategy.

[0014] In an embodiment, before the current building energy control strategy is hierarchically optimized to obtain a building energy saving strategy according to the hierarchical feedback strategy and through a hierarchical architecture of a decision layer, the method further includes: a building thermal dynamic response model is constructed through a dynamic sparse training mechanism; the building thermal dynamic response model is converted through a sparse neural network and embedded into a model predictive control optimization framework to obtain a lightweight building thermal dynamic response model; the building energy characteristics in the current building energy control strategy are constrained and penalized according to a model predictive control optimization framework in the lightweight building thermal dynamic response model and based on a weight coefficient control target function of running cost, energy consumption, thermal comfort ratio and self-sufficiency ratio, and the optimization preference of the model predictive control optimization framework is dynamically adjusted according to different operating environments to obtain a reinforced adaptive model for predictive control; and the reinforced adaptive model is configured as a middle-layer architecture of the decision layer.

[0015] In an implementable embodiment, a semi-Markov decision conversion process is performed on the mixed action space of equipment start-stop and power regulation in a multi-zone building to obtain an option critical framework responsible for discrete mode selection; the option critical framework is configured as an upper layer architecture of the decision layer; a reward function configuration related to constraint incentives is performed on a decision execution agent in the decision layer, and a constraint process of positive and negative rewards is performed on building energy feature actions to obtain a safety reward engineering mechanism for safety checking and correction of the building energy feature actions; the safety reward engineering mechanism is configured as a lower layer architecture of the decision layer; wherein the hierarchical architecture of the decision layer includes the upper layer architecture, the middle layer architecture and the lower layer architecture.

[0016] In an implementable embodiment, according to the hierarchical feedback strategy, the current building energy control strategy is subjected to hierarchical optimization processing through the hierarchical architecture of the decision layer to obtain a building energy saving strategy, specifically including: if the hierarchical feedback strategy is an energy efficiency abnormal feedback strategy, the current building energy control strategy is input into the hierarchical architecture of the decision layer; the current building energy control strategy is subjected to successive execution control under the upper layer architecture, the middle layer architecture and the lower layer architecture through the hierarchical architecture of the decision layer, so that the current building energy control strategy is subjected to energy saving control to obtain the building energy saving strategy.

[0017] In a second aspect, the embodiments of the present application further provide a building energy saving management device, the device comprising: at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, so that the at least one processor can execute the building energy saving management method of any of the above-mentioned embodiments.

[0018] In a third aspect, the embodiments of the present application further provide a non-volatile computer storage medium, the storage medium being a non-volatile computer readable storage medium, the non-volatile computer readable storage medium storing at least one program, each of the programs including instructions, the instructions causing the terminal to execute the building energy saving management method of any of the above-mentioned embodiments when executed by the terminal.

[0019] The present application provides a building energy saving management method, device and medium, compared with the prior art, the embodiments of the present application have the following beneficial technical effects: The technical solution of the present application is particularly suitable for commercial buildings, park complexes and large residential communities with multi-zone regulation requirements.

[0020] 1. For large commercial office buildings, through multi-zone collaborative control and fine energy scheduling, ensure that the indoor environmental parameters (temperature, humidity, CO2, concentration) of each region are always maintained within the healthy and comfortable range.

[0021] 2. In the park-level energy system, its modular architecture supports the plug-and-play and coordinated optimization of multiple types of energy equipment such as photovoltaic power generation, battery energy storage, heating and air conditioning, and significantly improves the overall energy self-sufficiency rate through source-storage-load collaborative scheduling.

[0022] 3. At the same time, it also provides personalized energy consumption feedback, automated energy saving report and intelligent voice interaction functions, which can effectively solve the problem of low user participation.

[0023] 4. For industrial plants, data centers and other special building types, the system's high reliability design and safety constraint protection mechanism ensures that the maximum energy saving potential is tapped on the premise of meeting strict process requirements.

[0024] 5. Under the background of regional energy internet, it can also be used as the intelligent management and control core of distributed energy nodes, and through multi-agent collaboration, it provides technical support for future cross-building energy trading and sharing. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments described in the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor. In the drawings: Figure 1 A building energy saving management method flow chart is provided for the embodiments of the present application; Figure 2 A building energy saving management system architecture diagram based on multi-source data fusion and multi-agent hierarchical decision is provided for the embodiments of the present application; Figure 3 A structural schematic diagram of a building energy saving management device is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0026] In order to make the person skilled in the art better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0027] The embodiment of the application provides a building energy saving management method, which comprises the following steps: Figure 1 As shown in the figure, the building energy saving management method specifically comprises steps S101-S105: S101, through a multi-node monitoring network, each independent energy consumption unit in the building is subjected to node monitoring processing, and node energy consumption data stored in a cloud platform is obtained.

[0028] Specifically, each independent energy consumption unit in the building needs to be identified first.

[0029] Further, the monitoring nodes are deployed into each independent energy consumption unit through the multi-node monitoring network.

[0030] Further, through the monitoring nodes, energy consumption data of each independent energy consumption unit is collected, and energy consumption collection data is obtained. The energy consumption collection data at least includes current monitoring data, voltage monitoring data, microcontroller cumulative electric energy consumption data and energy consumption power data.

[0031] Further, the monitoring nodes are interconnected with the cloud platform. Then, through the cloud platform, the energy consumption collection data is normalized stored under relevant space-time alignment, and node energy consumption data is obtained.

[0032] As a feasible implementation manner, Figure 2 The building energy saving management system architecture provided by the embodiment of the application is shown in the figure. Figure 2 As shown in the figure, the building energy saving management system of the application adopts a distributed and nodal hardware architecture, each monitoring node corresponds to an independent energy consumption unit (such as a single device, an independent room or a specific power distribution circuit) in the building. The node core is a customized embedded collection module, which supports AC 90-250V wide voltage input, has a plug-and-play feature, and can be quickly deployed through a standard power supply interface. The module uses a low-power microcontroller, integrates high-precision current sensing and voltage sampling circuits, realizes non-intrusive measurement of load current and voltage fluctuation, the sampling frequency is set to one per second, but the data upload period is 15 minutes, to balance real-time and power consumption, and ensure the real-time of power calculation.

[0033] As a feasible implementation, the current monitoring can directly measure the load current by clamping the wire. The voltage monitoring adopts resistance voltage division and signal conditioning circuit, and outputs an analog signal proportional to the grid voltage. The microcontroller calculates the instantaneous active power by true RMS algorithm, and accumulates the energy consumption by time integration. All raw data are filtered by sliding average and outliers before transmission, effectively suppressing the measurement jump caused by instantaneous interference, ensuring measurement accuracy. The collected current, voltage and power data are uploaded to the cloud database through Wi-Fi, providing real-time energy consumption basic data for the data processing layer, and providing state input for the intelligent decision-making layer.

[0034] As a feasible implementation, in view of the fluctuation of building power grid and long-term operation demand, the node module is built-in two-stage voltage stabilizing circuit: the front end adopts linear voltage stabilizing chip to convert 7-12V direct current input into 5V, and the rear stage generates 3.3V system power supply through low voltage difference stabilizer, and integrates reverse connection protection and overvoltage protection function. The design ensures that the module works stably within the voltage fluctuation range of ±15%, avoiding data interruption or hardware damage caused by power grid noise.

[0035] As a feasible implementation, the monitoring node establishes a long connection with the cloud platform through Wi-Fi, supports TCP / UDP protocol transmission, the average communication delay is less than 10ms, and the single data packet capacity is greater than 1KB. The module is built-in configurable network manager, and the user can quickly configure the Wi-Fi parameters through the local webpage. The data is packaged in JSON format, including device ID, timestamp, voltage, current, power and cumulative power, and is uploaded to the cloud database at a period of 15 minutes. Each node has a unique identifier, supports dynamic addition and deletion, and forms a flexible expansion multi-node monitoring network.

[0036] As a feasible implementation, all the data collected by the nodes are normalized and stored in the cloud, ensuring the spatio-temporal alignment of multi-source heterogeneous data. The transmission layer adopts WPA2 encryption protocol, the data packets are stored in the form of documents in the non-relational database, and the timestamp and node ID are used to realize fast retrieval and correlation analysis.

[0037] S102, store the node energy consumption data by preset building hierarchical data structure to obtain building energy spatio-temporal data model.

[0038] Specifically, it is also necessary to divide the cloud database of building documents into hierarchical data attributes, and determine the building hierarchical data structure. The building hierarchical data structure includes: park data, floor data, building data and equipment data.

[0039] Further, in combination with the building hierarchical data structure, the node energy consumption data is stored in the JSON document format, and the node energy consumption data is cross-device associated by monitoring the node ID and timestamp of the node, and a building energy space-time data model for multi-source data fusion is established.

[0040] As a feasible implementation, the monitoring node continuously collects voltage, current and power data, and after true effective value calculation and sliding average filtering, the data is packaged and uploaded to the cloud database every 15 minutes. At the same time, external data such as outdoor temperature and humidity, solar radiation intensity, and electricity price signals provided by BAS are received. Then a document-based cloud database needs to be established, using a four-level data structure of "park-building-floor-device": park archives (industry type, building area, tenant number), building information (building age, air conditioning system type), floor nodes (node ID, main energy-using equipment), and equipment-level energy consumption data (power, running time, time-of-use electricity price). Each data point is stored in JSON document format, and cross-device association is achieved through node ID and timestamp to establish a unified space-time data model. The data access layer supports MQTT / HTTP dual protocol access, performs outlier rejection and format standardization on input data to ensure input quality. Case data shows that the system can accurately capture daily energy consumption fluctuation patterns, with a peak power monitoring error of less than 2%.

[0041] S103、According to the building energy characteristics in the building energy space-time data model, the deviation degree between the actual energy consumption in the building energy efficiency level and the dynamic reference energy consumption is calculated to obtain the energy efficiency deviation evaluation result.

[0042] Specifically, first, the historical building energy characteristics in the building energy space-time data model are grouped and divided by the K-means clustering algorithm, and according to the characteristics of the building itself and the external environment of the building, the historical building energy characteristics in the building energy space-time data model are grouped and divided, and the historical energy consumption intensity region is obtained.

[0043] Further, the energy consumption intensity index of the historical energy consumption intensity region is calculated to obtain the historical actual energy consumption data.

[0044] Further, the historical actual energy consumption data is compared with the contemporaneous building itself data and the contemporaneous building external environment data to obtain the historical reference benchmark data. And based on the current building energy characteristics in the building energy space-time data model, the actual energy consumption data in the current energy consumption intensity region is obtained.

[0045] Further, according to the preset smoothing factor, the actual energy consumption data of the previous moment and the historical reference benchmark data of the previous moment are calculated and processed, and combined with the historical reference benchmark data of the current moment, the dynamic reference energy consumption data of the current moment is obtained.

[0046] Furthermore, based on the current moment, the ratio between the dynamic baseline energy consumption data and the actual energy consumption data is calculated to obtain the energy consumption deviation. Then, the energy consumption deviation is compared with a standard threshold to obtain the energy efficiency deviation assessment result for each energy intensity region.

[0047] In one embodiment, such as Figure 2 As shown, a dynamic energy efficiency baseline based on similarity clustering is introduced. First, using the K-means clustering algorithm, historical energy consumption data is divided into multiple groups based on features such as building type, area, time period, and outdoor temperature. The system calculates the energy intensity index for each region in real time and compares it with the dynamic baseline energy consumption based on historical data from the same period and under the same meteorological conditions. Energy efficiency level is expressed as actual energy consumption E. real With dynamic reference energy consumption E bench The deviation is quantified and represented, and a diagnostic process is automatically triggered when the deviation consistently exceeds a threshold. The formula is: Among them, E bench (t): The dynamic baseline energy consumption value at the current time t (dynamic baseline energy consumption data); μ history (t): Reference baseline calculated based on historical data; E real (t-1): The actual energy consumption at the previous moment. α A smoothing factor (0 ≤ α ≤ 1, typically 0.1-0.3) is used to control the rate at which the baseline value adapts to recent sudden changes. Then, based on... The energy consumption deviation η is obtained. When η>0, it indicates that the energy efficiency is better than the benchmark of similar scenarios, and when η<0, it indicates that the energy efficiency is lower. That is, the energy consumption deviation is compared with the standard threshold to obtain the energy efficiency deviation assessment result for each energy intensity region. Among them, the dynamic energy efficiency benchmark is used to evaluate the energy efficiency deviation of each region in real time. When the deviation exceeds the threshold, the intelligent decision layer is triggered to adjust the control strategy.

[0048] S104. Based on the energy efficiency deviation assessment results, generate a graded feedback strategy.

[0049] Specifically, the energy efficiency deviation assessment results also need to be categorized by deviation type. If the deviation type is an equipment-level anomaly, the tiered feedback strategy is an equipment maintenance feedback strategy. If the deviation type is a behavior-level anomaly, the tiered feedback strategy is a behavior adjustment feedback strategy. If the deviation type is a time-level anomaly, the tiered feedback strategy is an automatic adjustment feedback strategy.

[0050] As a feasible implementation method, the dynamic energy efficiency baseline is updated every 24 hours, using... α=0.2. The energy efficiency evaluation results are used to generate a hierarchical feedback strategy. When a district energy consumption deviates from the baseline by more than 15% for more than 2 hours, a secondary warning is triggered: first, a device maintenance suggestion is pushed (e.g., "East district air conditioner power anomaly, please check the filter"), and if there is no improvement within 1 hour, the control strategy is automatically adjusted (e.g., reduce the supply air setpoint). If it is a device-level anomaly (e.g., air conditioner running at high power continuously), a device maintenance feedback strategy is pushed. If it is a behavior-level anomaly (e.g., high energy consumption during peak hours), a behavior adjustment feedback strategy is generated.

[0051] As a feasible implementation, after generating the hierarchical feedback strategy, the feedback content is also dynamically generated by a template engine, considering the user's historical response rate to adjust the push frequency. At the same time, a user preference model is established, and users with low feedback acceptance are automatically switched to a threshold alarm mode.

[0052] At the same time, the system calculates real-time indoor environmental quality indicators, such as predicted percentage of dissatisfaction (PPD): ; where PMV (predicted mean vote) is determined by factors such as indoor temperature and humidity, and indoor activity intensity. When PPD (predicted percentage of dissatisfaction) exceeds the comfort threshold (e.g., 10%), the system adjusts the supply air or temperature setpoint on the premise of ensuring energy saving. At the same time, CO2 concentration (threshold 800 ppm) and relative humidity (range 40%-60%) are monitored to ensure indoor environmental health and comfort.

[0053] In addition, an automated report generation system can be designed, with report content dynamically generated based on templates, including daily / monthly / yearly energy consumption summary, comparison with the same period last month, energy saving achievements quantification, and customized improvement suggestions. It supports integration with smart home platforms, and users can view energy efficiency data in real time, receive abnormal alarms, and remotely adjust control strategies through mobile terminals. Voice assistant functions are integrated, and users can query energy consumption information and receive personalized prompts through natural language.

[0054] S105, according to the hierarchical feedback strategy, and through the hierarchical architecture of the decision layer, the current building energy control strategy is hierarchically optimized to obtain a building energy saving strategy.

[0055] It should be noted that, as shown in Figure 2 , the intelligent decision layer adopts a hierarchical architecture: the upper layer is responsible for discrete mode selection (e.g., HVAC start-stop, battery operation mode) by the Option-Critic framework, the middle layer performs continuous control (e.g., power regulation, supply air) by the Multi-Agent Deep Reinforcement Learning (MADRL), and the bottom layer performs safety verification and correction of actions through the Safety Reward Engineering (SRE) mechanism.

[0056] Specifically, a building thermal dynamic response model is constructed using a dynamic sparse training mechanism. Then, a sparse neural network is used to transform the building thermal dynamic response model into a mixed-integer linear programming constraint, and it is embedded into a model predictive control optimization framework to obtain a lightweight building thermal dynamic response model.

[0057] As a feasible implementation method, to address the high complexity of building thermal response modeling in commercial parks, a dynamic sparse training mechanism is adopted to construct a building thermal dynamic response model. During training, the k connections with the smallest absolute weight values ​​are periodically pruned: k = rpr(t)·(1-sl)·Nl; where: k is the number of connections pruned; rpr(t) is the pruning rate at time t; sl is the sparsity rate; and Nl is the total number of connections. Simultaneously, the k connections with the largest absolute gradient values ​​are regenerated to maintain network sparsity. The trained SNN is equivalently converted to mixed-integer linear programming (MILP) constraints using the Big-M method: a l i ≥ z l i a l i ≤ z l i - M(1-ψ l i ), a l i ≤ Mψ l i , ψ l i ϵ{0, 1}; where ψ is a binary auxiliary variable and M is a sufficiently large constant. During training, unimportant connections (with small absolute weights) are periodically pruned and new connections (with large absolute gradients) are regenerated, significantly reducing the number of model parameters. The trained SNN model is equivalently converted into mixed-integer linear programming constraints using the Big-M method, and embedded into the MPC optimization framework for efficient solution.

[0058] Furthermore, based on the model predictive control optimization framework in the lightweight building thermal dynamic response model, and using objective functions of operating cost, energy consumption, thermal comfort ratio, and self-sufficiency rate under weighted coefficient control, constraints and penalties are imposed on the building energy characteristics in the current building energy control strategy. The optimization preferences of the model predictive control optimization framework are dynamically adjusted according to different operating environments to obtain a reinforced adaptive model for predictive control. This reinforced adaptive model is then configured as the mid-level architecture of the decision-making layer.

[0059] In one embodiment, an adaptive model predictive control module employing a reinforcement learning-enhanced MPC framework is also required to address the uncertainties of the system model. Its core is a parameterized finite-time optimization problem, with the objective function as follows: ; wherein, OC t is the operation cost, EC t is the energy consumption, TCR t is the thermal comfort ratio, SSR t is the self-sufficiency ratio, cw , ew , tw , gw are the corresponding weight coefficients, respectively. These weight coefficients are not fixed but are dynamically adjusted by the upper reinforcement learning agent according to the system state, forming part of the action space: a t = [cw t , ew t , tw t , gw t , N p,t ], the optimization problem needs to satisfy the system dynamics and control constraints, i.e.: where the reward function r t of the RL agent integrates multiple objectives such as operation cost T in t , energy consumption u t , thermal comfort w t , and imposes a penalty term based on the logarithmic function for constraint violations: Then through continuous interaction, the RL agent learns to dynamically adjust the optimization preferences of the MPC according to different operating scenarios (such as high photovoltaic output on sunny days, peak electricity prices), ultimately obtaining a reinforcement adaptive model for predictive control.

[0060] Further, the mixed action space of equipment start-stop and power regulation in multi-zone buildings is processed through semi-Markov decision-making conversion to obtain an option critic framework responsible for discrete mode selection. The option critic framework is configured as the upper architecture of the decision layer.

[0061] As a feasible implementation, to solve the problem of mixed action space of equipment start-stop (discrete) and power regulation (continuous) in multi-zone buildings, the present application proposes a hierarchical multi-agent decision framework, which is formalized as a semi-Markov decision process (SMDP). The upper layer: the option critic framework is responsible for macro mode selection. Selecting "options" with extended duration, such as battery operation mode or HVAC start-stop. The value function of the option is defined as: where ω represents an option, π, ω, θ are the internal strategies under this option. The bottom layer is executed by multiple agents performing continuous actions such as battery power, air supply adjustment, etc. The agents optimize the overall performance through collaborative strategies. This "option-behavior" hierarchical structure (option critic framework) effectively decouples strategic decision-making and tactical control.

[0062] Further, the decision-making agent in the decision-making layer also needs to be configured with a reward function related to constraint incentives, and the constraint processing of positive and negative rewards for building energy feature actions is needed to obtain a safety reward engineering mechanism for safety verification and correction of building energy feature actions; the safety reward engineering mechanism is configured as the lower layer architecture of the decision-making layer.

[0063] As a feasible implementation, the reward function gives positive incentives for constraint satisfaction and logarithmic penalties for violation behavior to design double security. The reward function is designed as: , wherein OC t is the operation cost, PR t is the segmented reward for each constraint. When the action satisfies the constraint (such as the battery SOC being within the safe range, the indoor PMV being comfortable), a positive reward ζ1 is given; when the constraint is violated, a logarithmic penalty function is used: , wherein R t is the penalty reward at time t; ζ1 is the positive reward coefficient; X t is the variable value at time t; X max and X min are the upper and lower limits of the variable, respectively. At the same time, the mechanism contains a safety layer that will correct the action at to the nearest action at that satisfies the constraint when the action at violates the hard constraint: * t : The above design changes the post-penalty into pre-prevention, limits the exploration within the safe range, greatly reduces the risk of system out-of-control caused by random exploration at the initial stage of training, and accelerates the convergence.

[0064] The hierarchical architecture of the decision-making layer includes: an upper layer architecture, a middle layer architecture, and a lower layer architecture.

[0065] Further, if the hierarchical feedback strategy is an energy efficiency abnormal feedback strategy, the current building energy control strategy is input into the hierarchical architecture of the decision-making layer; Further, the current building energy control strategy is sequentially executed and controlled under the upper layer architecture, the middle layer architecture, and the lower layer architecture through the hierarchical architecture of the decision-making layer, so that the current building energy control strategy performs energy-saving control to obtain a building energy-saving strategy.

[0066] As a feasible implementation, the building energy-saving management system of the present application can run according to the "option selection-continuous adjustment-safety verification" process: first, select the discrete mode according to the state, then decompose it into each device set value, and finally verify the action feasibility after the SRE layer verification and execution. For example, ensure that the room temperature is maintained at 22-26℃, and CO2<800ppm. Experimental results show that this control strategy reduces the daily average power consumption of office buildings and the length of discomfort.

[0067] The core innovation of the application is to construct a hierarchical optimization framework deeply integrating intelligent decision-making and security guarantee. Compared with the existing single controller or centralized architecture, the building energy saving management system of the application innovatively solves the problem of coordination of discrete-continuous mixed action space in building energy systems by introducing an option critic architecture, decouples strategic mode selection (such as device start-stop) from tactical power regulation, and realizes the organic unity of decisions at different time scales. More breakthrough is that the designed security reward engineering mechanism surpasses the traditional constraint processing method relying on penalty function, and through the double protection of "positive incentive-security correction", it changes the constraint guarantee from passive post-punishment to active pre-prevention, fundamentally improving the safety and learning efficiency of the agent in the exploration process.

[0068] In addition, the embodiment of the application also provides a building energy saving management device, as shown in the figure Figure 3 The building energy saving management device 300 specifically comprises: at least one processor 301; and a memory 302 connected in communication with the at least one processor 301; wherein the memory 302 stores instructions executable by the at least one processor 301, so that the at least one processor 301 can execute: Through the multi-node monitoring network, each independent energy consumption unit in the building is subjected to node monitoring processing, and node energy consumption data stored in the cloud platform is obtained; Through the pre-set building hierarchical data structure, the node energy consumption data is stored hierarchically, and a building energy space-time data model is obtained; According to the building energy space-time data model, the deviation degree between the actual energy consumption and the dynamic benchmark energy consumption in the building energy efficiency level is calculated, and an energy efficiency deviation evaluation result is obtained; Based on the energy efficiency deviation evaluation result, a hierarchical feedback strategy is generated; According to the hierarchical feedback strategy, and through the hierarchical architecture of the decision layer, the current building energy control strategy is subjected to hierarchical optimization processing, and a building energy saving strategy is obtained.

[0069] The embodiments of the present application ensure that the indoor environmental parameters (temperature, humidity, CO2 concentration) of each region are always maintained within the healthy and comfortable range through multi-region collaborative control and refined energy scheduling. In the park-level energy system, the modular architecture supports plug-and-play and coordinated optimization of multiple types of energy equipment such as photovoltaic power generation, battery energy storage, heating, ventilation and air conditioning, and significantly improves the overall energy self-sufficiency rate through source-storage-load collaborative scheduling. Meanwhile, personalized energy consumption feedback, automated energy-saving reports and intelligent voice interaction functions are also provided, which can effectively solve the problem of low user participation. For special building types such as industrial plants and data centers, the high reliability design and safety constraint protection mechanism of the system ensure that the maximum energy-saving potential is tapped on the premise of meeting strict process requirements.

[0070] Each of the embodiments in the present application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiments.

[0071] The device and medium provided by the embodiments of the present application are one-to-one corresponding to the method, so the device and medium also have similar beneficial technical effects as the method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the device and medium will not be described here.

[0072] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0073] The present application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks.

[0074] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0075] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0076] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0077] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory. The memory is an example of computer-readable media.

[0078] Computer-readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media, such as modulated data signals and carrier waves.

[0079] It is also to be noted that the terms "comprising", "including", and any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0080] The above description is merely illustrative of the application, and not restrictive. Various modifications and changes can become apparent to those skilled in the art. Incorporating any modification, equivalent substitution, improvement, etc. within the spirit and principle of the application, shall be included in the scope of the description.

Claims

1. A building energy conservation management method, characterized in that, The method includes: Through a multi-node monitoring network, node monitoring processing is performed on each independent energy consumption unit in the building to obtain node energy consumption data stored in the cloud platform. By using a pre-defined building classification data structure, the node energy consumption data is stored in a hierarchical manner to obtain a building energy spatiotemporal data model. Based on the building energy characteristics in the building energy spatiotemporal data model, the deviation between the actual energy consumption in the building energy efficiency level and the dynamic benchmark energy consumption is calculated to obtain the energy efficiency deviation assessment result. Based on the energy efficiency deviation assessment results, a graded feedback strategy is generated; Based on the hierarchical feedback strategy and through the hierarchical architecture of the decision layer, the current building energy control strategy is optimized in a hierarchical manner to obtain a building energy-saving strategy.

2. The building energy conservation management method according to claim 1, characterized in that, Through a multi-node monitoring network, node monitoring processing is performed on each independent energy consumption unit within the building to obtain node energy consumption data stored in the cloud platform, specifically including: Identify each of the aforementioned individual energy-consuming units within the building; The monitoring nodes are deployed in each of the independent energy consumption units through the multi-node monitoring network. Through the monitoring node, energy consumption data is collected for each of the independent energy consumption units to obtain energy consumption data; wherein, the energy consumption data includes at least: current monitoring data, voltage monitoring data, microcontroller cumulative power consumption data, and energy consumption power data; The monitoring nodes are interconnected with the cloud platform via a protocol. The energy consumption data collected is normalized and stored under spatiotemporal alignment through the cloud platform to obtain the node energy consumption data.

3. The building energy conservation management method according to claim 1, characterized in that, By using a pre-defined building hierarchy data structure, the node energy consumption data is stored hierarchically to obtain a building energy spatiotemporal data model, specifically including: The cloud database of building documents is divided into hierarchical data attribute items to determine the hierarchical data structure of the building; wherein, the hierarchical data structure of the building includes: park data, floor data, building data and equipment data; Based on the building classification data structure, the node energy consumption data is stored in a JSON document format, and the node energy consumption data is associated across devices by monitoring the node ID and timestamp of the node to establish the building energy spatiotemporal data model for multi-source data fusion.

4. The building energy conservation management method according to claim 1, characterized in that, Based on the building energy characteristics in the aforementioned building energy spatiotemporal data model, the deviation between the actual energy consumption and the dynamic benchmark energy consumption in the building energy efficiency level is calculated to obtain the energy efficiency deviation assessment result, which specifically includes: Using the K-means clustering algorithm, and based on the characteristics of the building itself and the characteristics of the external environment, the historical building energy characteristics in the spatiotemporal data model of building energy are grouped to obtain the historical energy consumption intensity regions. Calculate the energy intensity index of the historical energy intensity area to obtain historical actual energy consumption data; The historical actual energy consumption data is compared with the data of the building itself and the external environment of the building during the same period to obtain historical reference benchmark data. Based on the current building energy characteristics in the building energy spatiotemporal data model, the actual energy consumption data in the current energy intensity area is obtained; Based on the preset smoothing factor, the actual energy consumption data of the previous moment and the historical reference data of the previous moment are calculated and processed, and combined with the historical reference data of the current moment to obtain the dynamic reference energy consumption data of the current moment. Based on the current moment, the ratio of the dynamic benchmark energy consumption data to the actual energy consumption data is calculated to obtain the energy consumption deviation. The energy consumption deviation is compared with the standard threshold to obtain the energy efficiency deviation assessment result for each energy consumption intensity region.

5. The building energy conservation management method according to claim 1, characterized in that, Based on the energy efficiency deviation assessment results, a tiered feedback strategy is generated, specifically including: The energy efficiency deviation assessment results are classified into different deviation types. If the deviation type is a device-level anomaly, then the hierarchical feedback strategy is a device maintenance feedback strategy. If the deviation type is a behavioral anomaly, then the hierarchical feedback strategy is a behavioral adjustment feedback strategy. If the deviation type is a time-level anomaly, then the hierarchical feedback strategy is an automatic adjustment feedback strategy.

6. The building energy conservation management method according to claim 1, characterized in that, Before obtaining a building energy-saving strategy by performing hierarchical optimization of the current building energy control strategy according to the hierarchical feedback strategy and through the hierarchical architecture of the decision layer, the method further includes: A building thermal dynamic response model is constructed through a dynamic sparse training mechanism; the building thermal dynamic response model is transformed into a mixed integer linear programming constraint through a sparse neural network and embedded into a model predictive control optimization framework to obtain a lightweight building thermal dynamic response model. Based on the model predictive control optimization framework in the lightweight building thermal dynamic response model, and based on the objective functions of operating cost, energy consumption, thermal comfort ratio and self-sufficiency rate under weighted coefficient control, the building energy characteristics in the current building energy control strategy are constrained and penalized. According to different operating environments, the optimization preference of the model predictive control optimization framework is dynamically adjusted to obtain an enhanced adaptive model for predictive control. Configure the enhanced adaptive model as a mid-level architecture of the decision layer.

7. A building energy conservation management method according to claim 6, characterized in that, The mixed action space of equipment start-up and power regulation in multi-zone buildings is transformed by semi-Markov decision-making to obtain the option critique framework responsible for discrete mode selection. Configure the option critique framework as the upper-level architecture of the decision-making layer; The decision-making execution agent in the decision layer is configured with a reward function for relevant constraints and incentives, and the building energy characteristic actions are subjected to positive and negative reward constraint processing to obtain a safety reward engineering mechanism for the safety verification and correction of the building energy characteristic actions. Configure the security reward engineering mechanism as a lower-level architecture of the decision-making layer; The hierarchical architecture of the decision-making layer includes: the upper-layer architecture, the middle-layer architecture, and the lower-layer architecture.

8. The building energy conservation management method according to claim 1, characterized in that, Based on the aforementioned hierarchical feedback strategy, and through the hierarchical architecture of the decision-making layer, the current building energy control strategy is optimized hierarchically to obtain a building energy-saving strategy, specifically including: If the hierarchical feedback strategy is an energy efficiency anomaly feedback strategy, then the current building energy control strategy will be input into the hierarchical architecture of the decision-making layer. Through the hierarchical architecture of the decision layer, the current building energy control strategy is sequentially executed and controlled at the upper, middle, and lower levels to enable the current building energy control strategy to perform energy-saving control, thereby obtaining the building energy-saving strategy.

9. A building energy conservation management device, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform a building energy conservation management method according to any one of claims 1-8.

10. A non-volatile computer storage medium, characterized in that, The storage medium is a non-volatile computer-readable storage medium that stores at least one program, each program including instructions that, when executed by a terminal, cause the terminal to perform a building energy conservation management method according to any one of claims 1-8.

Images