Light storage building group multi-target collaborative optimization scheduling system based on supply and demand matching degree

Abstract

The invention provides a supply and demand matching degree-based multi-target collaborative optimization scheduling system for a light storage building group, and the system comprises a sensing layer which is used for collecting the physical operation state data of the light storage building group in real time; the network layer is used for transmitting data to the control layer; the control layer is used for generating a global scheduling plan by taking maximization of the comprehensive supply and demand matching degree as an optimization target according to the supply and demand matching degree prediction information of the first scale and the time-of-use electricity price information of the power grid; according to the supply and demand matching degree prediction information of the second scale and the physical operation state data, performing rolling correction on the global scheduling plan to generate a correction instruction; according to the supply and demand matching degree prediction information of the third scale and the physical operation state data, performing local autonomous judgment on the correction instruction, and triggering a local response when supply and demand fluctuation is detected; and the application layer is used for displaying the data acquired by the sensing layer and / or the output data of the control layer, so that the self-balancing capability of the system is fundamentally improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy system optimal scheduling, and particularly relates to a photovoltaic storage building group multi-objective collaborative optimal scheduling system based on supply-demand matching degree. BACKGROUND

[0002] At present, the scale of building renewable energy applications represented by building integrated photovoltaics and distributed roof photovoltaics is rapidly expanding. However, photovoltaic power generation has significant diurnal intermittency, seasonal differences, and minute-level short-term dramatic fluctuation characteristics, while building group loads often exhibit early and late double-peak or specific operation rules, resulting in a serious mismatch between supply and demand on both sides. The existing comprehensive energy system optimal scheduling technology focuses on system comprehensive operation cost, primary energy utilization rate or carbon emissions as a single or main optimization target, and does not define the supply-demand space-time matching degree as a core state variable into the control closed loop.

[0003] However, this approach can only passively adjust energy storage or purchase electricity from the grid when facing photovoltaic output fluctuations, and cannot fundamentally improve the self-balancing ability of the system. SUMMARY

[0004] The present application provides a photovoltaic storage building group multi-objective collaborative optimal scheduling system based on supply-demand matching degree, to solve the defects in the prior art that when facing photovoltaic output fluctuations, only passive adjustment of energy storage or purchase of electricity from the grid can be made, and the self-balancing ability of the system cannot be fundamentally improved.

[0005] The present application provides a photovoltaic storage building group multi-objective collaborative optimal scheduling system, comprising: a perception layer for real-time collection of physical operation state data of the photovoltaic storage building group; a network layer for transmitting multi-time scale supply-demand matching degree prediction information integrated from an upstream prediction system and the physical operation state data to a control layer; the multi-time scale supply-demand matching degree prediction information includes first scale supply-demand matching degree prediction information, second scale supply-demand matching degree prediction information and third scale supply-demand matching degree prediction information; a control layer adopting a cloud-edge-end layered collaborative architecture of a cloud decision center, an edge collaborative node and a terminal execution unit, for generating a global scheduling plan with the maximum comprehensive supply-demand matching degree including a time matching degree index, a space matching degree index and a comprehensive matching degree index as an optimization target according to the first scale supply-demand matching degree prediction information and grid time-of-use electricity price information; performing rolling correction on the global scheduling plan according to the second scale supply-demand matching degree prediction information and the physical operation state data to generate correction instructions; performing local autonomous judgment on the correction instructions according to the third scale supply-demand matching degree prediction information and the physical operation state data, and triggering local response when detecting supply-demand fluctuations; an application layer for presenting data collected by the perception layer and / or output data of the control layer.

[0006] According to the photovoltaic storage building group multi-objective collaborative optimization scheduling system provided by the application, the first-scale supply-demand matching degree prediction information, the second-scale supply-demand matching degree prediction information and the third-scale supply-demand matching degree prediction information all include a multi-time scale rolling prediction sequence, a supply-demand matching degree index set and flexible load adjustment potential evaluation data; the multi-time scale rolling prediction sequence includes a photovoltaic output prediction sequence and a building group load prediction sequence; and the flexible load adjustment potential evaluation data includes a flexible load index, an adjustable power range and a maximum continuous adjustment time. The time matching degree index is used for quantifying the matching degree of the supply side and the demand side in the time dimension. The space matching degree index is used for quantifying the consumption efficiency of the supply side and the demand side in the space dimension. The comprehensive matching degree index is used for evaluating the self-balancing ability of the system as a whole.

[0007] According to the photovoltaic storage building group multi-objective collaborative optimization scheduling system provided by the application, the control layer generates a global scheduling plan with the optimization target of maximizing the comprehensive supply-demand matching degree including the time matching degree index, the space matching degree index and the comprehensive matching degree index, including: maximizing the optimized comprehensive supply-demand matching degree, minimizing the comprehensive operation cost, minimizing the carbon emission and maximizing the user comfort degree are taken as four-dimensional optimization targets to construct a multi-objective optimization model; An improved NSGA-III algorithm is used to solve the multi-objective optimization model to obtain the global scheduling plan.

[0008] According to the photovoltaic storage building group multi-objective collaborative optimization scheduling system provided by the application, the improved NSGA-III algorithm is used to solve the multi-objective optimization model to obtain the global scheduling plan, including: The Das-Dennis method is used to generate reference points in the four-dimensional target space, and the ideal point-extreme point method is used for target normalization; The best time lag amount in the first-scale supply-demand matching degree prediction information is used for population initialization to obtain an initial population; The initial population is subjected to genetic operation according to the comprehensive matching degree index, and the crossover and mutation parameters are adaptively adjusted to generate a child population; projection repair or penalty function punishment is performed on individuals violating the constraint condition to ensure that the child individuals are located in the constraint feasible region; A screening score formula with matching degree priority is introduced to perform environmental selection on the child population, and individuals satisfying preset conditions in supply-demand matching degree and user comfort degree safety margin are reserved to form a new generation population. After repeated iterations until convergence, the optimal solution is selected from the Pareto optimal solution set as the day-ahead global scheduling plan using the ideal point similarity ranking method.

[0009] According to the multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes provided by the present invention, the control layer is further used for: Integrate data on the assessment of flexible load regulation potential from upstream forecasting systems; The data on the assessment of flexible load adjustment potential is mapped to the decision variable constraints in the optimization model, and the control and scheduling instructions are kept within the physical limits of the building complex and the boundaries of user comfort.

[0010] According to the present invention, a multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes is provided, wherein the control layer performs rolling corrections on the global scheduling plan, including: Model predictive control is adopted, with the goal of tracking the global scheduling plan, and with the penalty terms of suppressing power fluctuations of energy storage devices and ensuring indoor temperature comfort, to solve the optimization problem within a rolling window and generate correction instructions. The model predictive control, based on the second-scale supply-demand matching degree prediction information and physical operation status data, smoothly corrects the energy storage output and air conditioning set temperature in each rolling cycle.

[0011] According to the present invention, a multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes includes a control layer that performs local autonomous judgment on the correction instructions based on the supply-demand matching degree prediction information of the third scale and the physical operation status data, comprising: Based on the supply and demand matching degree prediction information of the third scale, when the predicted value of supply and demand matching degree within the future preset time window is detected to be lower than the preset threshold, the autonomous smoothing mode is triggered, and the energy storage device is controlled to switch to the discharge preparation state in advance or reserve power support capacity by reducing non-critical loads. Furthermore, during the power scheduling process of executing the correction command, the local bus voltage and frequency are monitored in real time. When a voltage sag or frequency shift is detected, the photovoltaic inverter or bidirectional converter is controlled to automatically adjust its output according to the preset droop characteristic curve to provide millisecond-level transient support.

[0012] According to the multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes provided by the present invention, the control layer is subject to the following constraints when generating a global scheduling plan or correction instructions: The system's real-time power balance constraint is used to ensure that the total power on the supply side equals the total power on the demand side at any given time. Energy storage system operation constraints include upper and lower limits of state of charge, charging and discharging power constraints, and energy capacity constraints and charging and discharging power constraints of cold or heat storage devices; Building passive energy storage constraints are based on building thermal inertia and a first-order thermal network model to quantify dynamic changes in indoor temperature, combined with user-defined upper and lower limits for temperature comfort. Flexible load dispatch boundary constraints include instantaneous adjustment power upper and lower limit constraints and total adjustment power constraints within the dispatch window. The instantaneous adjustment power upper and lower limit constraints and total adjustment power constraints are determined based on the flexible load index and potential assessment results.

[0013] According to the multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes provided by the present invention, the physical operation status data collected by the sensing layer includes: Photovoltaic panel temperature, irradiance, and output power on the photovoltaic side; The energy storage side includes the battery's state of charge, health status, charging and discharging power, voltage and temperature, as well as the heat exchange fluid flow rate and inlet and outlet temperatures of the cold or heat storage equipment. The load side includes indoor temperature, carbon dioxide concentration, occupancy rate, and sub-item electrical loads, which include lighting load, socket load, power load, and air conditioning load.

[0014] According to the multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes provided by the present invention, the application layer is further used for: The execution result of the control layer is fed back to the control layer; The control layer dynamically calibrates the parameters of the system model based on the deviation between the execution result and the actual response value. The parameters include the building's equivalent thermal resistance and thermal capacity parameters, the capacity decay coefficient and charging and discharging efficiency of the energy storage device, and the user comfort preference weight. The network layer is also used to trigger a hierarchical takeover strategy when communication is interrupted, including switching the edge side to an independent optimization mode to maintain operation based on locally stored historical typical daily data and the most recent valid daily plan when communication between the control layer and the cloud is interrupted, and switching the terminal to a local autonomous mode to maintain local voltage and frequency stability and device safety as constraints for autonomous control when communication between the terminal and the edge side is interrupted.

[0015] The present invention provides a multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes based on supply and demand matching degree. The control layer can perform hierarchical optimization scheduling based on supply and demand matching degree prediction information at multiple time scales. Through the supply and demand matching degree prediction information at the first scale, a global scheduling plan is generated with maximizing the comprehensive supply and demand matching degree as the optimization objective. The global scheduling plan is rolled over and corrected according to the supply and demand matching degree prediction information at the second scale and physical operation status data. According to the supply and demand matching degree prediction information at the third scale, the correction instructions are judged locally autonomously. When supply and demand fluctuations are detected, a local response is triggered, realizing the transformation from passive response to active matching. This fundamentally solves the problem of source and load spatiotemporal misalignment and improves the local photovoltaic consumption rate and system self-balancing capability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the principle of the multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes based on supply and demand matching degree provided by the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] Figure 1 This is a schematic diagram of the principle of the multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes based on supply and demand matching degree provided by the present invention.

[0020] like Figure 1 As shown in the figure, this embodiment provides a multi-objective collaborative optimization scheduling system for photovoltaic-storage building clusters based on supply and demand matching degree, including: Perception layer 1 is used to collect real-time physical operation status data of the photovoltaic-storage building complex.

[0021] Specifically, the perception layer 1 consists of a series of distributed physical sensing devices. It adopts a high-frequency and high-precision acquisition method to complete the acquisition of physical quantities from the photovoltaic side, energy storage side and load side. The collected raw data will be aggregated to the edge gateway to provide a real and real-time physical state basis for optimized scheduling and avoid the distortion of scheduling strategy due to data loss or lag.

[0022] Physical operation status data, i.e., multi-source data acquisition of the photovoltaic-storage-building complex, is achieved through photovoltaic sensing units, energy storage sensing units, load sensing units, environmental sensing units, and equipment sensing units. The photovoltaic sensing unit collects irradiance, module temperature, output power, and inverter power. The energy storage sensing unit collects battery SOC, power / voltage / current, cold / heat storage equipment flow rate, and inlet / outlet temperature. The load sensing unit collects electrical load, cold / heat load, sub-item meters, and occupancy rate. The environmental sensing unit collects room temperature, relative humidity, irradiance, and carbon dioxide concentration. The equipment sensing unit collects COP, speed, pressure, and start / stop status.

[0023] The data collection on the photovoltaic side, known as source-side sensing, involves deploying photovoltaic inverter data acquisition devices and environmental monitoring stations to collect data on photovoltaic panel temperature, solar irradiance, and real-time output power. Data collection on the energy storage side, known as storage-side sensing, integrates a battery management system (BMS) monitoring device to monitor the state of charge (SOC), state of health (SOH), charging and discharging current, and voltage of the energy storage batteries in real time. Data collection on the load side, known as load-side sensing, utilizes smart meters, building automation systems (BAS), and environmental sensors to collect information on individual electricity loads (air conditioning, lighting, power), indoor temperature and humidity, carbon dioxide concentration, and occupant status.

[0024] Network layer 2 is used to transmit multi-timescale supply and demand matching degree prediction information integrated from the upstream prediction system, as well as physical operation status data, to control layer 3; the multi-timescale supply and demand matching degree prediction information includes supply and demand matching degree prediction information at the first scale, the second scale, and the third scale.

[0025] Specifically, the multi-timescale supply and demand matching degree prediction information includes the first-scale (day-ahead, 24-hour) supply and demand matching degree prediction information, the second-scale (intra-day, 4-hour) supply and demand matching degree prediction information, and the third-scale (ultra-short-term, 15-minute) supply and demand matching degree prediction information. Network layer 2 constructs a highly reliable, low-latency data transmission channel device, supporting both wide-area and local-area communication, compatible with multiple IoT protocols, enabling seamless access and interoperability of heterogeneous devices, ensuring bidirectional and rapid transmission of cloud-edge-device data links, and ensuring that prediction information and physical state data are transmitted to control layer 3 without delay or loss, providing data support for hierarchical scheduling.

[0026] Data communication and connectivity are achieved through wide-area communication, local-area communication, and supply-demand matching prediction sequence access. Wide-area communication utilizes LoRaWAN long-distance transmission, fiber optics, and 5G / 6G to enable large-volume data uploads and optimization command distribution between edge nodes and cloud servers, ensuring low latency and high bandwidth. Local-area communication employs industrial Ethernet, WiFi, ZigBee, or LoRa technologies to achieve local networking and interconnection between edge nodes and terminal devices (sensors, actuators). Supply-demand matching prediction sequence access utilizes multi-timescale rolling prediction result sequences and supply-demand matching indicators, supporting multiple IoT protocols such as MQTT, CoAP, and ModbusTCP to ensure seamless access and interoperability of heterogeneous devices.

[0027] Control layer 3 is used to generate a global scheduling plan based on the supply and demand matching degree prediction information of the first scale and the time-of-use electricity price information of the power grid, with the optimization goal of maximizing the comprehensive supply and demand matching degree, including time matching degree index, spatial matching degree index and comprehensive matching degree index; based on the supply and demand matching degree prediction information of the second scale and the physical operation status data, the global scheduling plan is rolled over and revised, and revision instructions are generated; based on the supply and demand matching degree prediction information of the third scale and the physical operation status data, the revision instructions are judged locally autonomously, and a local response is triggered when supply and demand fluctuations are detected.

[0028] Specifically, the control layer 3 adopts a cloud-edge-device hierarchical collaborative architecture consisting of a cloud decision center, edge collaborative nodes, and terminal execution units. It is the brain of the system, responsible for executing multi-timescale hierarchical optimization tasks and ultimately generating precise device control commands.

[0029] The cloud-based decision center performs day-ahead global optimization: as the system's global command center, it is responsible for long-term (24-hour) macro-level scheduling decisions. It receives day-ahead supply-demand matching degree prediction sequences generated by upstream systems, and based on global matching degree indicators, seeks the globally optimal solution in a multi-dimensional objective space that balances supply-demand matching degree, operating costs, carbon emissions, and user comfort. It then formulates the next day's scheduling plan and flexible load response baseline, establishing the macro-level trajectory of system operation.

[0030] Edge collaborative nodes perform intraday rolling corrections: Edge servers deployed on the building cluster side are responsible for intraday (4-hour) rolling corrections. They receive intraday supply and demand matching prediction information updated every 15 minutes from the upstream system, compare the deviation between the real-time operating status and the previous day's plan, and use model predictive control (MPC) technology to predict and eliminate the risk of matching degradation caused by sudden weather changes or load fluctuations within the rolling window. They also perform smooth corrections to energy storage output and air conditioning setpoints to ensure that the system's operating trajectory does not deviate from the global optimal domain.

[0031] The terminal execution unit provides real-time rapid response: an embedded unit within the controller is responsible for rapid execution and autonomous response from ultra-short-term (15 minutes) to real-time (ms / s). This device receives ultra-short-term supply-demand matching prediction information (15-minute scale) and commands issued at the edge. Based on this, it performs local smoothing and autonomous optimization. When it predicts a sharp decline in supply-demand matching due to cloud cover within the next 15 minutes, it immediately triggers an autonomous smoothing mode, proactively utilizing local energy storage or power compensation based on flexible loads. Combined with droop control, it maintains microgrid voltage / frequency stability, ensuring system robustness during command gaps.

[0032] Application layer 4 is used to display the data collected by perception layer 1 and / or the output data of control layer 3.

[0033] Specifically, Application Layer 4 provides users and maintenance personnel with a visual interactive interface terminal device and decision support services. Its main features include: Panoramic monitoring: displaying real-time supply and demand matching indicators, photovoltaic consumption ratio, and system energy flow diagrams in visual chart form; Strategy interaction: allowing users to set comfort preferences and cost budgets, with the system dynamically adjusting the weights of optimization parameters based on user intent; Benefit assessment: generating daily / monthly reports, quantitatively displaying the electricity savings, carbon emission reductions, and cost savings achieved through optimized scheduling, intuitively presenting the system's economic and social value.

[0034] The photovoltaic-storage building complex multi-objective collaborative optimization scheduling system provided in this embodiment is built on a four-layer CPS architecture, namely, perception layer 1, network layer 2, control layer 3, and application layer 4. It adopts a closed-loop design of physical perception, information optimization, and entity control, with each layer working collaboratively to transform the supply and demand matching information from the upstream forecasting system into precise control commands. Among them, the core CPS control layer 3 adopts a cloud-edge-device layered collaborative architecture, deeply integrating multi-timescale supply and demand matching information to adapt to the dynamic scheduling needs of the photovoltaic-storage-building complex integrated energy system.

[0035] Furthermore, based on the above embodiments, the supply and demand matching degree prediction information at the first, second, and third scales in this embodiment all include multi-time-scale rolling prediction sequences, a set of supply and demand matching degree indicators, and flexible load adjustment potential assessment data; the multi-time-scale rolling prediction sequences include photovoltaic power output prediction sequences and building cluster load prediction sequences; the flexible load adjustment potential assessment data includes flexible load index, adjustable power range, and maximum continuous adjustment time.

[0036] The Time Matching Index (TMI) quantifies the degree of matching between supply and demand in the time dimension; the Spatial Matching Index (SMI) quantifies the absorption efficiency of supply and demand in the spatial dimension; and the Comprehensive Matching Index (CMI) assesses the overall self-balancing capability of the system. These three indices quantify the source-load supply-demand matching status from different dimensions, providing a unified and quantifiable basis for the hierarchical scheduling of control layer 3. Furthermore, the values ​​of all three indices fall within the [0, 1] range, facilitating comparison of indices and formulation of scheduling strategies at different time scales. The specific definitions, calculation logic, mathematical expressions, and effects of each index are as follows: Time Matching Index (TMI): This index quantifies the consistency between supply and demand in the time dimension, directly reflecting the source-load self-balancing level without relying on energy storage, and is used to guide the charging and discharging timing scheduling of energy storage systems.

[0037] (1) In the formula, TThis represents the total number of moments within the calculation time window, i.e., the total time step of the optimization period; P load ( t )and P pv ( t ) are the outputs of the upstream system at the same time scale. t The baseline load and predicted photovoltaic output at time t, calculated over a period of length t. T The calculation window; the closer the TMI value is to 1, the more synchronized the supply-side output and demand-side load are in time within this window, and the easier it is for the system to achieve local consumption and peak shaving through means such as energy storage charging and discharging, load shifting and reduction; conversely, it indicates that there is a significant time mismatch, requiring more flexible resources to participate or cross-time adjustment.

[0038] Spatial Matching Index (SMI): This index combines geographical distance and network impedance to quantify the spatial accessibility of supply and demand and the cost of transmission loss, and is used to guide the local consumption and neighborhood scheduling strategies of distributed power sources.

[0039] (2) In the formula, Indicates the photovoltaic source point output by the upstream prediction system. i The output power, kW; This represents the building cluster load point output by the upstream forecasting system. j The demand, kW; Indicates photovoltaic source point i to load point j The equivalent fusion distance; The spatial scale parameter is determined by the distribution network topology and transmission loss. The smaller the distance, the more significant the decrease in weight.

[0040] Comprehensive Matching Index (CMI): As a macroscopic state quantity for judging the current self-balancing capability of the system, when the CMI is low, the scheduling system automatically triggers the balancing mode; when the CMI is high, the low-cost, low-disturbance fine-tuning mode is selected first.

[0041] CMI = a - TMI + β - SMI (3) In the formula, α and β These represent the weights of the TMI and SMI indicators, respectively, assigning their importance in the overall calculation. α + β =1, ensuring that CMI ultimately falls within the [0, 1] interval. Its value is determined comprehensively based on historical data and optimization objectives under different scenarios, so that CMI can accurately reflect the actual matching ability under different scenarios and avoid the distortion of one-size-fits-all calculation.

[0042] Furthermore, based on the above embodiments, in this embodiment, the control layer 3 generates a global scheduling plan with the optimization objective of maximizing the comprehensive supply and demand matching degree, which includes time matching degree index, spatial matching degree index and comprehensive matching degree index. This includes: constructing a multi-objective optimization model with maximizing the optimized comprehensive supply and demand matching degree, minimizing comprehensive operating cost, minimizing carbon emissions and maximizing user comfort as four-dimensional optimization objectives; and solving the multi-objective optimization model using the improved NSGA-III algorithm to obtain the global scheduling plan.

[0043] Specifically, control layer 3 generates a global scheduling plan with the optimization objective of maximizing the overall supply and demand matching degree. This is achieved through two core steps: constructing a four-dimensional multi-objective optimization model and solving the problem using an improved NSGA-III algorithm. This process is executed by the cloud-based decision center in control layer 3, utilizing the high-performance computing capabilities of the cloud to complete long-term global optimization. Unlike traditional scheduling models that only focus on minimizing operating costs, the optimization model constructed in this embodiment achieves coordinated optimization of physical matching, economic operation, and low-carbon environmental protection. The improved NSGA-III algorithm enhances the solution efficiency and the engineering feasibility of the solution, as follows: The cloud-based decision-making center in control layer 3 uses four-dimensional optimization objectives: maximizing the optimized overall supply-demand matching degree, minimizing overall operating costs, minimizing carbon emissions, and maximizing user comfort. It constructs a model by combining physical state data collected from perception layer 1, multi-timescale supply-demand matching degree prediction information integrated from the upstream forecasting system, and flexible load adjustment potential assessment data. Before model construction, the cloud-based decision-making center first calls the data preprocessing module, utilizing the optimal time lag provided by the upstream forecasting system. The photovoltaic output prediction curve is preprocessed by time-series shifting to eliminate the inherent systematic misalignment between the photovoltaic peak and the load peak, providing an aligned physical benchmark for optimization. This four-dimensional multi-objective optimization model overcomes the limitations of traditional single-objective or dual-objective optimization, taking into account the interests of investors, the grid side, and the demand side, and resolving the nonlinear conflict problem of strong coupling between objectives. The mathematical expressions of each objective function are as follows: Objective function 1: Maximize the overall supply and demand matching degree This invention prioritizes improving the source-load matching degree as the primary optimization objective, aiming to fundamentally solve the problems of photovoltaic absorption and grid impact. The objective function reshapes the flexible load curve of the building complex through proactive intervention of decision variables, enabling it to actively track the photovoltaic output curve in both spatiotemporal dimensions. The calculation logic follows the aforementioned evaluation system, but replaces the evaluation object with the flexible total load after scheduling, which is superimposed with adjustment measures. By quantifying the supply-demand matching degree in both time and space, the effect of the scheduling strategy on improving the system's self-balancing capability is evaluated. Its optimization objective function is shown in equation (4).

[0044] (4) In the formula, This represents the optimized overall supply and demand matching degree, with a value range of [0, 1]. The closer it is to 1, the higher the source-load matching degree. This indicates the optimized time matching degree; This indicates the optimized spatial matching degree.

[0045] The optimized time matching degree formula is shown in equation (5). When the value is 1, the total time step T is fully matched with the source load, and all photovoltaic power is consumed locally with no power deviation. When the value is 0, the photovoltaic output and load are completely out of sync within the total time step T, and the photovoltaic power cannot be consumed locally.

[0046] (5) (6) In the formula, express t Real-time scheduling of various flexible loads and the total system load after energy storage, kW; express t Flexible load regulation that is constantly involved in scheduling, in kW; , They represent energy storage systems. t The charging and discharging power at any given time, in kW, is where the heat storage / discharging of the cold / heat storage device is converted into electrical power based on the overall COP of the air conditioning system.

[0047] The optimized spatial matching degree formula is shown in Equation (7). By adjusting the flexible load distribution of each building node, the total load nodes of the system actively move towards the more matched photovoltaic nodes, thereby improving the absorption efficiency.

[0048] (7) In the formula, This represents the building cluster load point output by the prediction system. j exist t It can constantly schedule various flexible loads and the total system load after energy storage, in kW.

[0049] Objective function 2: Minimize overall operating cost The objective function aims to reduce the system's total lifecycle operating expenses from an economic perspective through refined scheduling. Unlike traditional models that only focus on electricity costs, this invention constructs a three-dimensional cost model that covers grid interaction, equipment losses, and user compensation, achieving cost accounting for the entire lifecycle and all participating parties, which is more in line with the needs of full-process economic calculation in actual engineering projects. Its optimization objective function is shown in equation (8).

[0050] (8) In the formula, This represents the cost of grid interaction, i.e., the economic expenditure or revenue from purchasing or selling electricity, expressed in yuan. This represents the cost of energy storage device charging and discharging losses, expressed in yuan. This represents the incentive compensation paid to users participating in grid demand response, specifically the incentive fees paid by building operators to users in order to acquire demand-side regulation capabilities, expressed in yuan.

[0051] (9) In the formula, express t The power exchange between the photovoltaic system and the power grid is positive for purchasing electricity and negative for selling electricity. When there is a surplus of photovoltaic power, it sells electricity to the grid and draws electricity from the grid when there is a load shortage. express t Time-of-use electricity price on the power grid, in yuan / kWh; express t The on-grid price of photovoltaic power is RMB / kWh.

[0052] (10) In the formula, This represents the initial investment cost of the energy storage system, expressed in yuan. express t The depth of discharge (0~1) of the energy storage device, i.e., the discharge amount / battery rated capacity, is a core indicator affecting battery life (the larger the DOD, the fewer the battery cycle count and the faster the life decay). This represents the maximum theoretical number of cycles associated with DOD, and it has a non-linear inverse relationship with DOD. Indicates energy storage system t The charging and discharging power at any given moment, in kW; Δt Indicates the time step.

[0053] (11) In the formula, The unit power compensation price represents the non-essential load that can be reduced by the user, which causes inconvenience to the user and is compensated at a higher price, in yuan / kWh. express t The regulating power, kW, can be reduced at any time to decrease the load. This indicates the unit power compensation price for loads that can be shifted. This means that when a user shifts the load from peak hours to off-peak hours, only the electricity consumption time changes, there is no usage loss, and the compensation price is relatively low, in yuan / kWh. express t Adjustable power (kW) for load shifting at any time.

[0054] Objective function 3: Minimize carbon emissions The core of this optimization objective function is to reduce the indirect carbon footprint of building complexes from the perspective of optimizing the electricity consumption structure. This differs from the traditional approach of simply reducing the total amount of electricity consumed. By using a scheduling strategy, the system purchases more electricity during low-carbon periods and less during high-carbon periods, enabling users to track and absorb clean power generation from the grid, ultimately contributing to emission reduction across society. This optimization objective only accounts for the indirect carbon emissions from electricity purchased from the grid; the carbon emissions from local photovoltaic self-consumption are zero. The optimization objective function is shown in equation (12).

[0055] (12) In the formula, express t The building complex's power purchase capacity from the power grid at any given time, in kW; express t The system measures the dynamic carbon emission factor of the power grid in real time, expressed as kgCO2 / kWh. This factor is no longer a traditional annual average but a dynamic curve that varies over time and is strongly correlated with the real-time power generation mix of the upstream grid. For example, the factor is lower during peak wind power generation late at night and higher during peak coal-fired power generation in the evening. The system acquires or predicts this factor in real time through the CPS network layer 2, guiding the energy storage system to charge during low-carbon periods and discharge during high-carbon periods, thereby contributing to overall emission reduction on the user side.

[0056] Objective function 4: Maximize user comfort This optimization objective function is geared towards the user experience of energy scheduling in building complexes. It breaks away from the traditional model's design approach of treating comfort as a rigid constraint, integrating thermal comfort and light comfort into a maximizable benefit objective. Through an elastic quantification model, it achieves a dynamic balance between energy efficiency and user experience under extreme conditions, which is a key objective to ensure the long-term operation of the system. User experience is the core of the acceptability of the scheduling strategy. This optimization objective integrates two core building indoor environmental indicators: thermal comfort and light comfort, covering the most intuitive user experience. Its optimization objective function is shown in equation (13).

[0057] (13) In the formula, This represents the weighting factor for thermal comfort. This represents the weighting factor for lighting environment comfort, and satisfies... This ensures the normalization of evaluations; express t The predicted thermal sensation index at any given time has indoor temperature and environmental parameters as its core independent variables, with values ​​ranging from [-3, +3]. express tThe thermal comfort tolerance limit at any given time; express t Indoor illuminance requirement at any given time; express t The actual indoor illuminance value at any given time.

[0058] The cloud-based decision center uses physical state data collected by perception layer 1, prediction sequences output by the upstream prediction system, matching degree indices, optimal time lag, and flexible boundaries as prior information. It then applies the improved NSGA-III algorithm to solve a four-dimensional multi-objective optimization model, searching for the Pareto optimal solution set within the feasible region of the constraints. The improved NSGA-III algorithm is specifically optimized for the characteristics of four-dimensional objectives, solving the problems of low optimization efficiency and poor engineering feasibility of solutions in high-dimensional solution spaces found by traditional algorithms. Ultimately, it outputs a global scheduling plan that considers all objectives, providing a macroscopic reference trajectory for intraday rolling corrections and real-time autonomous response.

[0059] This embodiment constructs a four-dimensional multi-objective optimization model and uses an improved NSGA-III algorithm to solve it, so that the global scheduling plan is no longer biased towards a single objective, but finds the best balance point among multiple conflicting objectives, ensuring the scientific and global nature of the global scheduling plan, and laying the foundation for subsequent hierarchical scheduling.

[0060] Furthermore, based on the above embodiments, this embodiment employs an improved NSGA-III algorithm to solve the multi-objective optimization model and obtain a global scheduling plan, including: generating reference points in the four-dimensional target space using the Das-Dennis method and normalizing the target using the ideal point-extreme point method; initializing the population based on the optimal time lag in the supply-demand matching degree prediction information of the first scale to obtain an initial population; performing genetic operations on the initial population based on the comprehensive matching degree index and adaptively adjusting the crossover and mutation parameters to generate a progeny population; performing projection repair or penalty function punishment on individuals that violate the constraints to ensure that the progeny individuals are located in the constrained feasible region; introducing a matching degree-first screening and scoring formula to perform environmental selection on the progeny population, retaining individuals whose supply-demand matching degree and user comfort safety margin meet the preset conditions to form a new generation population; repeating the iteration until convergence, and then selecting the optimal solution from the Pareto optimal solution set as the day-ahead global scheduling plan using the ideal point similarity ranking method.

[0061] Specifically, (1) Reference point generation and target normalization The improvement of this step compared to the standard NSGA-III algorithm is to perform normalization adaptation for the characteristics of four-dimensional targets. The cloud algorithm module uses the Das-Dennis method to uniformly generate reference points on the unit hyperplane of the four-dimensional target space. The target dimension is 4, the number of segments is p, and thus the total number of reference points is determined. To ensure that the dimensions of different targets are consistent, the ideal point-extreme point method is used for normalization, as shown in equation (14). F 1 (supply and demand matching degree) F 2 (Operating Costs) F 3 (carbon emissions) and F 4. (Comfort) is mapped to a comparable scale, that is, the four different targets with different dimensions, namely matching degree (0~1), cost (yuan), carbon emission (kg) and comfort (0~1), are mapped to the same scale to solve the problem of unbalanced allocation of reference points caused by the inconsistency of different target dimensions and ensure the stability of the algorithm's search direction.

[0062] (14) In the formula, Represents the normalized i-th i One target value; Indicates the first i One target value, , They are the first i The minimum and maximum values ​​of each target within the search space.

[0063] (2) Population initialization and decision variable encoding A hybrid initialization method combining randomness and heuristics is adopted to ensure the diversity of the solution set, improve the matching degree of the initial population, and reduce the number of iterations of the NSGA-III algorithm. Decision variables include energy storage charging and discharging power, flexible load adjustment, and, when necessary, the charging and discharging heat of cold and hot energy storage. The decision variable encoding is shown in Equation (15). A portion of the initial individuals utilize the obtained optimal lag time. Perform heuristic construction, for example, before the photovoltaic peak. Pre-charge / pre-cool in advance to improve performance. TMI opt The initial level, allowing the initial population to F The supply-demand matching degree is significantly higher than that of pure random initialization, which accelerates the convergence of the algorithm; another part is randomly generated within the constraint boundary to avoid the homogenization of the population caused by heuristic initialization and ensure that the algorithm can search for the global optimal solution.

[0064] (15) in, Indicates energy storage system t Charge / discharge power at any given time, in kW. ; express t Flexible load regulation that participates in scheduling at all times, in kW. ; Indicates cold / heat storage device t Constant charging and discharging power (heating and cooling), kW .

[0065] (3) Constraint handling and feasibility repair Multi-layer constraint checks are performed on newly generated individuals to ensure the constraint closure of the NSGA-III algorithm, ensuring that all individuals in the evolutionary process satisfy hard constraints and avoiding solutions that are optimal in the algorithm but infeasible in engineering. The constraint checks mainly include power balance, SOC boundary, cold and hot energy storage capacity boundary, building temperature and lighting comfort boundary, and flexible adjustment energy constraints. Individuals that slightly violate the boundaries are repaired by projection, and severely infeasible solutions are reduced by a penalty function strategy to ensure that the evolutionary process is carried out within the executable domain. The specific formula is shown in Equation (16). Through the multi-layer strategy of checking-repairing-penalizing, the algorithm evolution process is always ensured to be within the constraint feasible domain, and the output solutions can be converted into actual scheduling instructions.

[0066] (16) In the formula, This represents the amount of violation of the k-th constraint (such as the deviation of the power balance constraint or the amount of SOC exceeding the boundary). This represents a penalty factor, which punishes individuals who violate constraints and reduces their fitness.

[0067] (4) Genetic operations and iteration The cloud-based system simulates binary crossover (SBX) and polynomial mutation to generate offspring populations and performs multiple generations of evolutionary iterations. The core improvement is the adaptive adjustment of mutation parameters based on the CMI (Comprehensive Matching Index), allowing the NSGA-III algorithm's search strategy to align with the system's operating state and avoiding the inefficiency caused by a one-size-fits-all approach to crossover and mutation parameters. The specific formula is shown below. Crossover and mutation parameters are adaptively adjusted according to the scenario: when the CMI is low, the mutation rate is appropriately increased to enhance global search capabilities; when the CMI is high, the mutation rate is decreased to improve the stability and executability of the solution.

[0068] (17) In the formula, The distribution index representing the polynomial variation. The smaller the value, the greater the variation. , These represent the minimum and maximum values ​​of the variation index, respectively.

[0069] (5) Environment selection based on matching degree In the environmental selection phase of each generation of evolution, the NSGA-III algorithm determines which scheduling schemes can enter the next generation through an improved elite retention strategy. Traditional NSGA-III algorithms typically use random selection when faced with multiple solutions with similar performance, located in the same non-dominated level, and with similar crowding levels. This invention innovatively introduces a matching-first evaluation criterion. That is, when some individuals must be eliminated, the system calculates the comprehensive score of each scheme and prioritizes retaining schemes with higher supply-demand matching and greater comfort and safety margins. This mechanism ensures that the algorithm's evolutionary direction always closely follows the two core requirements of improving absorption and ensuring comfort, avoiding ineffective searches in low-value regions during the evolutionary process, thus obtaining higher-quality scheduling strategies that are more in line with engineering realities within a limited computation time. The specific selection and scoring formula is as follows: (18) In the formula, Indicates the comfort margin. The larger the size, the greater the comfort level. , These represent the weights of the optimization objectives.

[0070] (6) Convergence Criterion and Decision Output The iteration terminates when the maximum number of iterations is reached or when there is no significant improvement in the Pareto front for several consecutive generations. After convergence, the algorithm outputs a set of Pareto optimal solutions and uses the ideal solution similarity ranking method (TOPSIS method) to calculate the comprehensive priority of each Pareto solution. Based on the current system's operating preferences, a compromise optimal solution is automatically selected as the basic scheduling plan for the day-ahead. The decision variables of this optimal solution are encapsulated into control instruction packets, which are distributed to the edge collaborative nodes through CPS network layer 2 and serve as the benchmark for intraday rolling corrections.

[0071] This embodiment improves the NSGA-III algorithm through multiple improvements, making it more suitable for solving four-dimensional multi-objective optimization models. The output global scheduling plan takes into account the matching degree of supply and demand and the interests of all parties, and has good engineering feasibility, providing a reliable benchmark for intraday rolling correction.

[0072] Furthermore, based on the above embodiments, the control layer 3 in this embodiment is also used to: integrate flexible load adjustment potential assessment data from the upstream forecasting system; map the flexible load adjustment potential assessment data into decision variable constraints in the optimization model, and control the scheduling instructions not to exceed the physical limits of the building complex and the user comfort boundary.

[0073] Specifically, control layer 3 integrates flexible load adjustment potential assessment data from the upstream forecasting system through network layer 2. This data is calculated by the upstream forecasting system based on user comfort constraints and includes the Flexible Load Index (FLI), adjustable power range, and maximum continuous adjustment time. This comprehensively quantifies the adjustment capability of flexible loads in the building complex and provides a clear adjustment boundary reference for the scheduling strategy formulation of control layer 3.

[0074] Control layer 3 directly maps the integrated flexible load adjustment potential assessment data into decision variable constraints in the optimization model, transforms the flexible load index (FLI) and adjustable power range into instantaneous adjustment power constraints, and transforms the maximum continuous adjustment time into total adjustment power constraints for the time period. This ensures that the decision variables of the optimization model are always within the adjustment capacity of the flexible load, and mathematically guarantees that the dispatching instructions do not exceed the physical limits of the building complex and the boundaries of user comfort.

[0075] This embodiment deeply binds the flexible load adjustment potential assessment data with the constraints of the optimization model, making the allocation of flexible resources more quantitative and avoiding the blind adjustment of flexible loads in traditional scheduling. It fully taps the flexible adjustment potential of the building complex, prevents excessive adjustment from interfering with users' normal energy use, and improves the user acceptability of the scheduling strategy.

[0076] Furthermore, the multi-timescale hierarchical scheduling strategy is a core component of system operation, strictly adhering to the cloud-edge-device three-layer architecture of CPS control layer 3, and implementing hierarchical control. Through the coordinated cooperation of devices at different levels, the system ensures the precise implementation of scheduling commands from macro to micro levels.

[0077] Specifically, the macro-level scheduling layer (cloud-based decision center—day-ahead global optimization) is the brain of the scheduling system. It performs long-term, coarse-grained global optimal planning, providing lower layers with insurmountable optimization boundaries and reference trajectories to prevent the middle / lower layers from getting trapped in local optima. The key parameters and settings for this layer are as follows: Time scale: 24 hours in advance, time step 1 hour.

[0078] Input data: Upstream system provides day-ahead comprehensive matching degree forecast information (CMI, TMI, SMI, etc.) for the next 24 hours, flexible load index (FLI), photovoltaic / load forecast values, and grid time-of-use electricity price information.

[0079] Execution logic: The cloud server performs global optimization based on the constructed multi-objective optimization model and the improved NSGA-III algorithm. Under the premise of meeting all-day energy balance and comfort constraints, the system plans the reference power for energy storage charging and discharging for each hour of the following day, guided by maximizing CMI and minimizing operating costs. and flexible load response benchmark .

[0080] This level primarily addresses the issues of overall economic efficiency and macro-level compatibility. For example, it plans to charge the battery when electricity prices are low the following day and photovoltaic forecasts are insufficient, and to enhance the absorption capacity through energy storage discharge or pre-cooling / preheating buildings during peak photovoltaic periods. It also increases load regulation during periods with high flexible load indices.

[0081] Output: Generate a global scheduling baseline plan for the next 24 hours and distribute it to the edge server via 5G / private network as a reference baseline for the next level.

[0082] Furthermore, based on the above embodiments, in this embodiment, the control layer 3 performs rolling corrections on the global scheduling plan, including: using model predictive control to track the global scheduling plan as the objective, and using the suppression of power fluctuations of energy storage devices and the guarantee of indoor temperature comfort as penalties, and solving the optimization problem within the rolling window to generate correction instructions; model predictive control performs smooth corrections on energy storage output and air conditioning set temperature based on the second-scale supply and demand matching degree prediction information and physical operating status data in each rolling cycle.

[0083] Specifically, the regional coordination layer (edge ​​collaboration nodes—intraday rolling correction) is the central hub of the scheduling system. It performs medium-cycle, fine-grained rolling optimization corrections to eliminate errors in daily forecasts, ensuring the scheduling trajectory aligns with actual operating conditions. It also connects the global baseline in the cloud with the real-time execution at the terminal. The key parameters and settings for this layer are as follows: Time scale: 4 hours in advance, time step 15 minutes.

[0084] Input data includes: the day-ahead baseline plan issued by the cloud, the supply and demand matching forecast information for the next 4 hours updated every 15 minutes by the upstream system, and the real-time building operation status uploaded by the perception layer 1, such as battery SOC, indoor temperature, and actual photovoltaic output.

[0085] Execution Logic: Edge servers deployed across the building complex employ Model Predictive Control (MPC) to periodically initiate rolling optimization (every 15 minutes or 1 hour). The MPC controller tracks the day-ahead baseline plan but allows for fine-tuning within a certain range to accommodate intraday forecast errors. It solves the optimization problem within a rolling window (the next 4 hours) but only issues instructions for the first time step (the next 15 minutes).

[0086] Mathematical Model: A quadratic programming (QP) model is constructed, which has a fast solution speed, is adaptable to the computing power of edge servers, and can achieve rapid optimization within 15 minutes. This ensures that the corrected trajectory tracks the daily plan as closely as possible, while suppressing drastic fluctuations and satisfying real-time constraints. The objective function is as follows: (19) In the formula, express k The deviation between the actual energy storage power and the cloud-based reference power is constantly monitored to ensure that the lower-level corrections do not deviate from the globally optimal trajectory. express k The rate of change of actual energy storage power at all times can suppress frequent start-ups and shutdowns or sudden power changes in equipment and extend equipment life. express k The deviation between the actual indoor temperature and the set temperature is monitored at all times, prioritizing user comfort and reflecting the scheduling principle of prioritizing user experience. , These represent the penalty coefficients for the two optimization objectives, respectively. This indicates the prediction time domain, set to 4 hours.

[0087] Output: Generates power correction instructions and air conditioning set temperature correction amounts for the next 15 minutes, and sends the final instruction, which combines the base value and the correction value, to the terminal execution unit in real time.

[0088] Furthermore, based on the above embodiments, in this embodiment, the control layer 3 performs local autonomous judgment on the correction command according to the supply and demand matching degree prediction information of the third scale and the physical operation status data. This includes: based on the supply and demand matching degree prediction information of the third scale, when it is detected that the predicted value of the supply and demand matching degree within the future preset time window is lower than the preset threshold, triggering the autonomous smoothing mode, controlling the energy storage device to switch to the discharge preparation state in advance or reserving power support capacity by reducing non-critical loads; and during the power scheduling process of executing the correction command, monitoring the local bus voltage and frequency in real time, and when a voltage sag or frequency deviation is detected, controlling the photovoltaic inverter or bidirectional converter to automatically adjust the output according to the preset droop characteristic curve to provide millisecond-level transient support.

[0089] Specifically, the real-time control layer 3 (terminal execution unit—real-time autonomous response) acts as the limbs of the scheduling system, responsible for short-cycle, millisecond-level physical execution and autonomous response. It translates digital instructions from the upper layers into physical actions of the devices, while also handling ultra-short-term sudden disturbances such as cloud cover and voltage dips to ensure system physical safety. It is the final link in implementing scheduling instructions. The key parameters and settings for this layer are as follows: Time scale: 15 minutes in advance, time step in seconds / milliseconds.

[0090] Input data: final scheduling instructions issued from the edge side, local ultra-short-term supply and demand matching degree prediction information (15-minute scale, targeting extremely short-term features such as cloud cover), and local port voltage / frequency sampling values.

[0091] Execution logic: It is divided into two modes: normal execution and autonomous response. In normal state, it accurately executes the upper-level instructions. In case of sudden disturbances, it does not need to wait for the upper-level instructions and responds autonomously locally, realizing a combination of passive execution and active protection.

[0092] 1) Conventional mode: The terminal unit embedded in the photovoltaic inverter, bidirectional converter and smart home appliance controller is responsible for converting the digital instructions from the upper layer into physical actions.

[0093] 2) Autonomous Smoothing Mode: When the terminal receives an ultra-short-term forecast indicating a significant risk of cloud cover in the next 15 minutes (i.e., a sharp drop in TMI and an expected decrease in photovoltaic output within seconds), the terminal can automatically trigger the smoothing mode using its local edge computing capabilities without waiting for instructions from the upper layer. For example, the energy storage system can immediately switch to a discharge preparation state, or non-critical loads can be disconnected via smart sockets to reserve power support capacity.

[0094] 3) Droop Control Stabilization Mode: During power dispatching, the terminal device monitors the local bus voltage and frequency in real time (millisecond level). If a photovoltaic instantaneous drop or a sudden increase in load causes a voltage sag or frequency shift, the photovoltaic inverter, bidirectional converter, etc., immediately adjust their output automatically according to the preset droop characteristic curve to achieve millisecond-level support. Before a new equilibrium point is established, millisecond-level transient stability is achieved, avoiding large voltage or frequency shifts that could lead to equipment tripping or system collapse. This is the last line of defense for the physical safety of the power system.

[0095] Furthermore, based on the above embodiments, the control layer 3 in this embodiment is subject to the following constraints when generating a global scheduling plan or correction instruction: real-time power balance constraints of the system, used to ensure that the total power on the supply side is equal to the total power on the demand side at any given time; energy storage system operation constraints, including upper and lower limits of state of charge constraints, charging and discharging power constraints, and energy capacity constraints and charging and discharging power constraints of cold or heat storage devices; building passive energy storage constraints, which quantify the dynamic changes in indoor temperature based on building thermal inertia and a first-order thermal network model, combined with the upper and lower limits of temperature comfort set by the user; flexible load scheduling boundary constraints, including upper and lower limits of instantaneous adjustment power and total adjustment power constraints within the scheduling window, the upper and lower limits of instantaneous adjustment power and the total adjustment power limits are determined based on the flexible load index and potential assessment results.

[0096] Specifically, to ensure that the optimization results are physically feasible, operationally safe, and acceptable to users, the cloud-based decision center constructs the following multi-dimensional set of rigid constraints based on the device parameters, real-time status data, and upstream prediction results uploaded by the CPS perception layer 1: Constraint 1: Real-time power balance constraint of the system (physical hard constraint) This constraint is a global, mandatory physical constraint with no room for flexible adjustment. Its core purpose is to ensure that the total power on the supply side equals the total power on the demand side at any given time, maintaining the stability of the grid's voltage and frequency. The constraint conditions are as follows: (20) In the formula, the left side represents the total power on the supply side, including the predicted output of photovoltaic power, the amount of electricity exchanged with the grid, and the discharge of energy storage; the right side represents the total power on the demand side, including the baseline load, the flexible load adjustment, and the charging of energy storage. This constraint clearly stipulates that the system must meet the energy conservation requirement in real time to avoid load loss due to power shortage or voltage overruns and equipment protection activation due to power excess.

[0097] Constraint 2: Energy storage system operation constraints (equipment safety constraints) For electrochemical energy storage and cold / heat storage devices, establish refined state models and boundary constraint models to prevent overcharging and over-discharging from shortening lifespan and to avoid comfort or equipment safety risks caused by excessive regulation of cold and heat storage.

[0098] The state model of the energy storage device is as follows: (twenty one) In the formula, express t The state of charge of the energy storage device at any given time reflects the remaining amount of electricity in the energy storage device; , These represent the charging efficiency and discharging efficiency of the energy storage device, respectively, with values ​​[0, 1].

[0099] The safety capacity boundary conditions of the energy storage device are shown in Equation (22). By limiting the upper and lower limits of SOC, the deep overcharging and over-discharging of the battery are avoided, thereby fundamentally ensuring battery safety and extending cycle life. The charging and discharging power limiting conditions are shown in Equations (23) and (24). The charging and discharging rates are constrained to not exceed the rated capacity of the converter, and the battery overheating and accelerated aging caused by short-term high-power switching are avoided.

[0100] (twenty two) (twenty three) (twenty four) In the formula, , These represent the minimum and maximum values ​​of the state of charge of the energy storage device, respectively. , These represent the maximum charging power and maximum discharging power of the energy storage device, respectively.

[0101] The state model of the cold / heat storage device is as follows: (25) In the formula, express t The remaining energy of the cold / heat storage device at any given time, in kWh; , They represent t The power of the cold / heat storage device for charging and releasing cold energy at any given time, in kW; , These represent the charging and discharging efficiencies of the cold / heat storage device, respectively, with values ​​[0, 1].

[0102] The energy capacity boundary condition of the cold / heat storage device is shown in Equation (26), and the charge / discharge power boundary is shown in Equations (27) and (28). By limiting the energy storage capacity and instantaneous charge / discharge capacity of cold and heat storage, the device's lifespan is reduced due to frequent overload operation, and it is matched with the cold and heat load demand of the building complex, providing achievable physical support for the comfort index in the objective function.

[0103] (26) (27) (28) In the formula, , These represent the minimum remaining capacity and maximum rated capacity of the cold / heat storage device, respectively, in kWh; , These represent the maximum charging thermal power and the maximum releasing thermal power of the cold / heat storage device, respectively.

[0104] Constraint 3: Building passive energy storage constraints (environmental comfort constraints) By using the thermal capacity characteristics of the building envelope and interior furnishings as passive energy storage, the building's thermal inertia is transformed from an uncontrollable environmental factor into a dispatchable and flexible resource. By quantifying the dynamic changes in indoor temperature through a first-order RC thermal network model and combining it with the temperature comfort zone boundary, the building's own thermal / cold storage capacity can be used to participate in energy dispatch while meeting user comfort requirements.

[0105] The dynamic quantification model for indoor temperature is as follows: (29) In the formula, express t The indoor temperature at any given time, in °C; express t The outdoor temperature at any given time, in °C; R and C represent the building's equivalent thermal resistance and heat capacity, respectively, in °C / kW and kWh / °C. express t The building's internal heat gain at any given time, in W / m².

[0106] The boundary conditions for indoor temperature comfort and light environment comfort are shown in Equation (30) and Equation (31) respectively, ensuring that the indoor temperature and light environment are always maintained within the comfort range set by the user during the scheduling process. When the external working conditions change suddenly, the system prioritizes ensuring that the temperature boundary and light environment boundary are not broken, and then seeks the optimal energy consumption and matching degree within the feasible domain.

[0107] (30) (31) In the formula, , Indicates the upper and lower limits of indoor temperature comfort, in °C; , lx represents the upper and lower limits of indoor lighting comfort.

[0108] Constraint 4: Boundary Constraints for Flexible Load Scheduling (User Willingness Constraints) By transforming users' subjective acceptance and the actual availability of equipment into quantifiable mathematical boundaries, flexible load scheduling is upgraded from purely technical power regulation to optimized scheduling that takes into account both technical feasibility and user acceptance. This constraint is directly linked to the upstream flexible load index and potential assessment results, defining scheduling constraints from two dimensions: instantaneous power and total power consumption per period.

[0109] The instantaneous power regulation constraint and the total power regulation constraint for a time period are shown in Equation (32) and Equation (33), respectively. The former limits the maximum power that can be reduced or shifted at each moment to avoid affecting the normal operation of key equipment; the latter limits the total regulation energy within a scheduling window to prevent excessive and frequent use of flexible resources from interfering with the normal usage rhythm of users.

[0110] (32) (33) In the formula, , express t The upper and lower limits of the adjustable power of the flexible load at any given time are determined by the flexible load index and the potential assessment results, in kW; This represents the total adjustable power limit within the flexible scheduling window, in kWh.

[0111] Furthermore, based on the above embodiments, in this embodiment, the application layer 4 is also used to: feed back the execution results of the control layer 3 to the control layer 3; the control layer 3 dynamically calibrates the parameters of the system model according to the deviation between the execution results and the actual response values, the parameters including the equivalent thermal resistance and thermal capacity parameters of the building, the capacity attenuation coefficient and charging and discharging efficiency of the energy storage device, and the user comfort preference weight; the network layer 2 is also used to trigger a hierarchical takeover strategy when communication is interrupted, including when the control layer 3 is interrupted in communication with the cloud, the edge side switches to an independent optimization mode to maintain operation based on the historical typical daily data and the most recent effective daily plan stored locally, and when the terminal is interrupted in communication with the edge side, the terminal switches to a local autonomous mode to maintain local voltage and frequency stability and equipment safety as constraints for autonomous control.

[0112] Specifically, in addition to displaying the data collected by the perception layer 1 and the output data of the control layer 3, the application layer 4 is also used to feed back the execution results of the control layer 3 to the control layer 3. The execution results include the actual execution status of the scheduling instructions, the actual response value of the device, etc., providing a basis for the dynamic calibration of the model parameters of the control layer 3.

[0113] Building upon hierarchical scheduling, a closed-loop feedback system spanning the cloud, edge, and device was further constructed. This mechanism not only handles real-time command interaction but, more importantly, enables dynamic calibration between the physical system and the optimization model, ensuring that the optimization strategy is always based on the most realistic system state.

[0114] (1) Two-way information flow interaction Downlink control flow (constraint injection and instruction refinement): Cloud → Edge: The baseline plan issued before the deadline serves as a coarse-grained guide, while defining the optimization boundary for the MPC control of the edge layer to ensure that the edge layer can only be adjusted within the optimization boundary, avoiding sacrificing global economy due to local optimization.

[0115] Edge → Terminal: Issue intraday correction instructions as fine-grained control, transforming mathematical optimization results into physical action instructions, and attaching priority labels to ensure that the terminal prioritizes critical business when resource conflicts occur.

[0116] Uplink feedback flow (state awareness and deviation reporting): Terminal → Edge (High Frequency): Real-time upload of physical status at the second / minute level, including actual power, real-time room temperature, and temporary user intervention behavior.

[0117] Edge → Cloud (Low Frequency): The edge server analyzes the cause of the deviation by comparing the issued command value with the actual measured response value of the terminal. If the deviation is caused by equipment aging, the edge layer corrects the equipment parameter model of the terminal. If the deviation is caused by changes in the building's thermal characteristics, the building's RC model parameters are corrected, and then the updated parameters are uploaded to the cloud.

[0118] (2) Adaptive correction of model parameters Based on the deviation between the execution results fed back from the application layer 4 and the actual response values, the control layer 3 dynamically calibrates, or adaptively corrects, the parameters of the system model. The cloud-based decision center has a model dynamic calibration engine that uses the deviation data fed back from the uplink to periodically identify and correct the parameters of the system's core optimization model, thereby achieving self-iteration of the strategy. Physical parameter correction: If it is found that the building's temperature rises faster than the model prediction under the same weather conditions, the system will automatically correct the building's thermal resistance and thermal capacity parameters, thereby improving the accuracy of virtual energy storage assessment.

[0119] Equipment aging adaptation: By monitoring the relationship between battery charge / discharge curves and SOC changes over a long period of time, the battery capacity decay coefficient and charge / discharge efficiency parameters are dynamically updated to prevent scheduling plans from becoming unexecutable due to equipment aging.

[0120] User behavior update: The system uses machine learning algorithms to analyze users' records of manual intervention in air conditioning temperature settings and dynamically adjusts the weight coefficients in the comfort objective function to make the scheduling strategy more in line with users' actual usage habits.

[0121] (3) Abnormal takeover and fault tolerance mechanism To address extreme situations such as cloud-edge or edge-device communication interruptions or equipment failures, a tiered takeover strategy should be established, prioritizing the protection of system physical security. Cloud-Edge Disconnection: When the edge server loses its connection to the cloud, it automatically switches to independent optimization mode. It uses historical typical daily data stored locally and the most recent effective daily plan as a basis, relying solely on intraday rolling optimization on the edge side to maintain system operation and ensure basic supply and demand matching.

[0122] Edge-to-end link failure: When the terminal equipment cannot receive commands from the edge side, it immediately reverts to the local autonomous mode. With the goal of maintaining local voltage / frequency stability and equipment safety constraints, the photovoltaic inverter switches to power-limited operation, and the energy storage system stops unnecessary charging and discharging, providing only basic support to smooth out fluctuations and ensure system safety.

[0123] like Figure 1 As shown, the functions of control layer 3 are explained in general, mainly implementing five steps: S1, S2, S3, S4, and S5.

[0124] S1: Multi-source heterogeneous data access and scenario analysis. This mainly includes: physical state data acquisition, divided into source measurement, storage side, and load side data acquisition; upstream forecast data integration, divided into multi-timescale rolling forecast sequences, supply and demand matching index sets, and flexible adjustment potential assessment data.

[0125] S2: Construct a multi-objective optimization model based on supply and demand matching degree. This includes constructing the objective function and setting constraints. The objective function aims to maximize the overall supply and demand matching degree, minimize the overall operating cost, minimize carbon emissions, and maximize user comfort. Constraints include: real-time power balance constraints, energy storage system operation constraints, building passive energy storage constraints, and flexible load dispatch boundary constraints.

[0126] S3: Solving using an improved NSGA-III algorithm guided by matching degree. Reference point generation and target normalization are performed based on the Das-Dennis method, matching degree, cost, carbon emissions, and comfort. Population initialization and decision variable encoding are performed based on "random + heuristic" initialization, energy charge / discharge power and heat, and flexible adjustment. Constraint processing and feasibility repair are performed based on penalty function strategy, power balance, SOC, and comfort boundary. Genetic operations and iterations are performed based on binary crossover, polynomial mutation, and adaptive adjustment of mutation parameters according to CMI. Matching degree-based environmental selection is achieved based on a matching degree screening strategy, prioritizing solutions with higher supply-demand matching degree and greater comfort. Convergence criteria and decision output are achieved by ranking ideal solutions by similarity and calculating the comprehensive priority of each Pareto solution.

[0127] S4: Multi-timescale hierarchical scheduling strategy. The macro-scheduling layer implements day-ahead global optimization and global optimization based on the improved NSGA-III algorithm. The regional scheduling layer implements intraday rolling correction and uses the MPC method to initiate rolling optimization. Real-time control layer 3 implements real-time autonomous response, autonomous smoothing, and droop control voltage stabilization mode.

[0128] S5: Multi-level closed-loop feedback and adaptive correction mechanism. It achieves bidirectional information flow interaction through downlink control flow and uplink feedback flow. It achieves adaptive parameter correction through physical parameter calibration, equipment aging adaptation, and user behavior updates. It implements anomaly takeover and fault tolerance mechanisms through "cloud-edge" disconnection (independent optimization mode) and "edge-device" disconnection (local autonomous mode).

[0129] This invention, through the construction of a multi-objective collaborative optimization scheduling system for photovoltaic-storage building clusters based on supply and demand matching, has made comprehensive innovations in physical architecture, control logic, and optimization system, and has the following significant beneficial effects: 1. It pioneered an optimization paradigm driven by "supply and demand matching degree" to improve photovoltaic consumption and source-load synergy.

[0130] Unlike traditional dispatching methods that only aim to passively absorb photovoltaic power or reduce electricity costs, this invention innovatively upgrades "supply and demand matching degree (CMI, TMI, SMI)" from an evaluation indicator to a core optimization objective. By actively reshaping the flexible load pattern to "track" the photovoltaic output curve, it fundamentally solves the problem of source-load spatiotemporal misalignment, significantly reduces the impact of photovoltaic fluctuations on the power grid, and greatly improves energy utilization efficiency.

[0131] 2. Construct a multi-time-scale hierarchical scheduling system that balances global optimization with real-time stability control.

[0132] By constructing a hierarchical architecture of "daytime macro planning - intraday rolling correction - real-time autonomous response," the system can handle global planning, intraday deviations, and instantaneous disturbances at different time scales: the cloud layer ensures macro-optimal performance and resource coordination; the edge layer utilizes MPC technology to effectively respond to sudden intraday weather changes; and the terminal layer solves the problem of second-level fluctuations such as cloud cover through an "autonomous smoothing mode." This hierarchical mechanism solves the problem of "achieving both global optimization and second-level stability," making scheduling commands engineering-executable and ensuring stable system operation under extreme conditions.

[0133] 3. Achieve synergistic optimization of economic efficiency and low carbon emissions, significantly reducing system operating costs and carbon emissions.

[0134] This invention considers economic factors such as time-of-use pricing, battery lifespan degradation, and demand response compensation in its optimization model, and introduces a dynamic carbon emission factor of the power grid as a low-carbon target, achieving a synergistic trade-off between cost, carbon emissions, and matching degree. Compared with strategies that rely solely on peak-valley electricity price arbitrage, this invention explicitly incorporates battery lifespan degradation, avoiding lifespan losses caused by high-frequency charging and discharging, and thus facilitating optimal economic performance throughout the entire lifecycle.

[0135] 4. The synergy between flexible resource quantification and comfort constraints improves scheduling feasibility and user acceptability.

[0136] This invention directly maps the "Flexible Load Index (FLI)" and "Adjustable Boundary" into scheduling decision variables, and incorporates thermal comfort and illuminance deviation into a multi-objective optimization system. By hierarchically classifying and allocating various flexible loads, low-cost adjustment resources are prioritized while ensuring that comfort does not exceed limits. This "quantitative constraint + flexible coordination" mechanism avoids excessive intervention in users' normal energy consumption, reduces reliance on electrochemical energy storage and its lifespan, and significantly improves the long-term sustainability of the demand response strategy.

[0137] 5. Closed-loop feedback and dynamic model calibration ensure long-term operational accuracy and system robustness.

[0138] By establishing a two-way feedback mechanism, the system can periodically identify and dynamically calibrate model parameters using "command-response" deviations and key state characteristics, reducing the risk of model mismatch caused by equipment aging and changes in operating conditions. The edge side can fine-tune parameters for short-term deviations, while the cloud side performs centralized calibration and model updates for long-term deviations, ensuring the scheduling strategy remains executable and consistent throughout its lifecycle. Combined with independent optimization and local autonomy mechanisms in network outage scenarios, the system can maintain its safety boundaries under extreme conditions such as communication anomalies and equipment failures, enhancing robustness and reliability in engineering environments.

[0139] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0140] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes, characterized in that, include: The perception layer is used to collect real-time physical operation status data of the photovoltaic-storage building complex; The network layer is used to transmit multi-timescale supply and demand matching degree prediction information integrated from the upstream prediction system, as well as the physical operation status data, to the control layer; the multi-timescale supply and demand matching degree prediction information includes supply and demand matching degree prediction information at the first scale, supply and demand matching degree prediction information at the second scale, and supply and demand matching degree prediction information at the third scale. The control layer adopts a cloud-edge-device hierarchical collaborative architecture consisting of a cloud decision center, edge collaborative nodes, and terminal execution units. It is used to generate a global scheduling plan based on the supply and demand matching degree prediction information of the first scale and the grid time-of-use electricity price information, with the optimization goal of maximizing the comprehensive supply and demand matching degree, including time matching degree index, spatial matching degree index, and comprehensive matching degree index. Based on the supply and demand matching degree prediction information of the second scale and the physical operation status data, the global scheduling plan is rolled out and a correction instruction is generated; Based on the supply and demand matching degree prediction information of the third scale and the physical operation status data, the correction instruction is judged locally and autonomously, and a local response is triggered when supply and demand fluctuations are detected. The application layer is used to display the data collected by the perception layer and / or the output data of the control layer.

2. The multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes according to claim 1, characterized in that, The supply and demand matching degree prediction information at the first scale, the second scale, and the third scale all include multi-time-scale rolling prediction sequences, a set of supply and demand matching degree indicators, and flexible load adjustment potential assessment data. The multi-time-scale rolling prediction sequences include photovoltaic power output prediction sequences and building cluster load prediction sequences. The flexible load adjustment potential assessment data includes flexible load index, adjustable power range, and maximum continuous adjustment time. The time matching index is used to quantify the degree of matching between the supply side and the demand side in the time dimension; The spatial matching degree index is used to quantify the absorption efficiency of the supply side and the demand side in the spatial dimension. The comprehensive matching degree index is used to evaluate the overall self-balancing ability of the system.

3. The multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes according to claim 2, characterized in that, The control layer generates a global scheduling plan with the optimization objective of maximizing the comprehensive supply-demand matching degree, which includes time matching degree index, spatial matching degree index, and comprehensive matching degree index. This plan includes: A multi-objective optimization model is constructed with four-dimensional optimization objectives: maximizing the optimized overall supply and demand matching degree, minimizing overall operating costs, minimizing carbon emissions, and maximizing user comfort. The improved NSGA-III algorithm is used to solve the multi-objective optimization model to obtain the global scheduling plan.

4. The multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes according to claim 3, characterized in that, The improved NSGA-III algorithm is used to solve the multi-objective optimization model to obtain a global scheduling plan, including: Reference points are generated in the four-dimensional target space using the Das-Dennis method, and the target is normalized using the ideal point-extreme point method. Based on the optimal time lag in the supply and demand matching degree prediction information of the first scale, the population is initialized to obtain the initial population. Genetic operations are performed on the initial population based on the comprehensive matching index, and the crossover and mutation parameters are adaptively adjusted to generate the offspring population; projection repair or penalty function punishment is performed on individuals that violate the constraints to ensure that the offspring individuals are within the constraint feasible region; A matching-first screening and scoring formula is introduced to select the offspring population for environmental selection, retaining individuals whose supply-demand matching degree and user comfort and safety margin meet preset conditions to form a new generation population; After repeated iterations until convergence, the optimal solution is selected from the Pareto optimal solution set using the ideal point similarity ranking method, and used as the day-ahead global scheduling plan.

5. The multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes according to claim 1, characterized in that, The control layer is also used for: Integrate data on the assessment of flexible load regulation potential from upstream forecasting systems; The data on the assessment of flexible load adjustment potential is mapped to the decision variable constraints in the optimization model, and the control and scheduling instructions are kept within the physical limits of the building complex and the boundaries of user comfort.

6. The multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes according to claim 1, characterized in that, The control layer performs rolling revisions to the global scheduling plan, including: Model predictive control is adopted, with the goal of tracking the global scheduling plan, and with the penalty terms of suppressing power fluctuations of energy storage devices and ensuring indoor temperature comfort, to solve the optimization problem within a rolling window and generate correction instructions. The model predictive control, based on the second-scale supply-demand matching degree prediction information and physical operation status data, smoothly corrects the energy storage output and air conditioning set temperature in each rolling cycle.

7. The multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes according to claim 1, characterized in that, The control layer performs local autonomous judgment on the correction instruction based on the supply and demand matching degree prediction information of the third scale and the physical operation status data, including: Based on the supply and demand matching degree prediction information of the third scale, when the predicted value of supply and demand matching degree within the future preset time window is detected to be lower than the preset threshold, the autonomous smoothing mode is triggered, and the energy storage device is controlled to switch to the discharge preparation state in advance or reserve power support capacity by reducing non-critical loads. Furthermore, during the power scheduling process of executing the correction command, the local bus voltage and frequency are monitored in real time. When a voltage sag or frequency shift is detected, the photovoltaic inverter or bidirectional converter is controlled to automatically adjust its output according to the preset droop characteristic curve to provide millisecond-level transient support.

8. The multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes according to claim 1, characterized in that, The control layer is subject to the following constraints when generating a global scheduling plan or correction instructions: The system's real-time power balance constraint is used to ensure that the total power on the supply side equals the total power on the demand side at any given time. Energy storage system operation constraints include upper and lower limits of state of charge, charging and discharging power constraints, and energy capacity constraints and charging and discharging power constraints of cold or heat storage devices; Building passive energy storage constraints are based on building thermal inertia and a first-order thermal network model to quantify dynamic changes in indoor temperature, combined with user-defined upper and lower limits for temperature comfort. Flexible load dispatch boundary constraints include instantaneous adjustment power upper and lower limit constraints and total adjustment power constraints within the dispatch window. The instantaneous adjustment power upper and lower limit constraints and total adjustment power constraints are determined based on the flexible load index and potential assessment results.

9. The multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes according to claim 1, characterized in that, The physical operating status data includes: Photovoltaic panel temperature, irradiance, and output power on the photovoltaic side; The energy storage side includes the battery's state of charge, health status, charging and discharging power, voltage and temperature, as well as the heat exchange fluid flow rate and inlet and outlet temperatures of the cold or heat storage equipment. The load side includes indoor temperature, carbon dioxide concentration, occupancy rate, and sub-item electrical loads, which include lighting load, socket load, power load, and air conditioning load.

10. The multi-objective collaborative optimization scheduling system for photovoltaic-storage building complexes according to claim 1, characterized in that, The application layer is also used for: The execution result of the control layer is fed back to the control layer; The control layer dynamically calibrates the parameters of the system model based on the deviation between the execution result and the actual response value. The parameters include the building's equivalent thermal resistance and thermal capacity parameters, the capacity decay coefficient and charging and discharging efficiency of the energy storage device, and the user comfort preference weight. The network layer is also used to trigger a hierarchical takeover strategy when communication is interrupted, including switching the edge side to an independent optimization mode to maintain operation based on locally stored historical typical daily data and the most recent valid daily plan when communication between the control layer and the cloud is interrupted, and switching the terminal to a local autonomous mode to maintain local voltage and frequency stability and device safety as constraints for autonomous control when communication between the terminal and the edge side is interrupted.

Images