An operation optimization control method and system for industrial and commercial energy storage devices

Abstract

The application provides an operation optimization control method and system for industrial and commercial energy storage equipment, relates to the technical field of optimization control, obtains the physical connection relationship and real-time electrical quantity measurement data of each energy storage equipment, photovoltaic unit, transformer and key load node in an industrial and commercial park, constructs a heterogeneous graph neural network model, and embeds the real-time electrical quantity measurement data as node initial features into the heterogeneous graph neural network model; the change rate of the total incoming line power of the low-voltage side of the transformer is monitored in real time, a load jump event is determined, the upper and lower limits of the available charging and discharging power of the energy storage system are determined according to the real-time electrical quantity measurement data of each energy storage equipment, and a robust optimization model containing an opportunity constraint is constructed; and the charging and discharging strategy is obtained by potential game of each energy storage converter in the industrial and commercial park according to the robust optimization model. The application can realize load jump event perception and potential game distributed solution driven by a heterogeneous graph neural network, and improve the power supply reliability and operation economy of industrial and commercial energy storage.

Description

Technical Field

[0001] This application relates to the field of optimization control technology, and more specifically, to a method and system for optimizing the operation of industrial and commercial energy storage equipment. Background Technology

[0002] Currently, various optimization control methods have been disclosed in existing industrial and commercial energy storage equipment operation and control technologies. For example, a centralized optimization scheduling method based on ultra-short-term load forecasting and genetic algorithms predicts the load curve of the industrial park in future periods, aiming at peak-valley arbitrage and demand control, and centrally solves the charging and discharging power commands of each energy storage converter. Another example is the rolling optimization method based on model predictive control, which corrects prediction errors through real-time feedback and dynamically adjusts the energy storage charging and discharging strategy to track the planned curve. Furthermore, a distributed control method based on multi-agent reinforcement learning has also been proposed, where each energy storage converter acts as an independent agent, gradually optimizing its own charging and discharging strategy through interaction with the environment. In terms of safety control, existing technologies also include anti-reverse current protection methods based on real-time monitoring of transformer low-voltage side power, which immediately disconnect the energy storage charging circuit when power backflow is detected. Regarding topology modeling, existing research uses graph neural networks to extract features from grid nodes for load forecasting or state estimation. In terms of uncertainty handling, robust optimization and chance-constrained programming methods have been applied to power system dispatching decisions, addressing fluctuations in renewable energy and load by setting uncertainty sets or probabilistic constraints.

[0003] In existing operational control technologies for commercial and industrial energy storage devices, centralized optimization methods based on ultra-short-term load forecasting struggle to accurately detect power surges caused by impulsive loads, leading to demand charge control failures or malfunctions in anti-reverse current protection. Graph neural network modeling based on a single electrical topology ignores the dynamic impact of communication latency and data packet loss rate on the effectiveness of control command execution. Distributed control methods based on consensus algorithms are prone to charging and discharging strategy conflicts when global coupling constraints are lacking, making it difficult to achieve a balance between individual economic efficiency and global security. Therefore, improving the power supply reliability and operational economy of commercial and industrial energy storage through load surge event perception and distributed problem-solving driven by heterogeneous graph neural networks is a challenge facing the industry. Summary of the Invention

[0004] This application provides a method and system for optimizing the operation control of industrial and commercial energy storage equipment. It can improve the power supply reliability and operational economy of industrial and commercial energy storage through load jump event perception and distributed solution of potential game driven by heterogeneous graph neural network.

[0005] In a first aspect, this application provides an operation optimization control method for industrial and commercial energy storage equipment, the optimization control method comprising the following steps: The physical connection relationships and real-time electrical measurement data of various energy storage devices, photovoltaic units, transformers and key load nodes in the industrial and commercial park are obtained, and then a heterogeneous graph neural network model is constructed. The real-time electrical measurement data is embedded into the heterogeneous graph neural network model as the initial feature of the node. The rate of change of the total incoming power on the low-voltage side of the transformer is monitored in real time. When the absolute value of the rate of change exceeds the preset sudden change threshold, it is determined as a load jump event. The jump amplitude and duration are recorded. The upper and lower limits of the available charging and discharging power of the energy storage system are determined based on the real-time electrical measurement data of each energy storage device. In this way, a robust optimization model with chance constraints is constructed. Based on the robust optimization model, a potential game is performed on each energy storage converter in the industrial and commercial park to obtain the charging and discharging strategies when each energy storage converter reaches Nash equilibrium.

[0006] In this embodiment, the real-time electrical measurement data includes: Terminal voltage, charging and discharging current, state of charge, health status, and battery temperature of each energy storage device; Output power, DC-side voltage, and irradiance of each photovoltaic unit; The active power, reactive power, power factor, and winding temperature of the transformer; Active power, reactive power, and current harmonic distortion rate at each critical load node.

[0007] In this embodiment, constructing a heterogeneous graph neural network model and embedding the real-time electrical measurement data as initial node features into the heterogeneous graph neural network model specifically includes: Using the physical connections as edges and each energy storage device, photovoltaic unit, transformer, and key load node as vertices, a basic graph structure for a heterogeneous graph neural network model is constructed. The initial feature vector of each vertex is determined based on the real-time electrical measurement data; The heterogeneous graph neural network model is configured with power flow edges and control signal edges. The weights of the power flow edges are determined by the line impedance parameters, and the weights of the control signal edges are determined by the current communication delay and data packet loss rate. A graph attention mechanism is adopted to perform message passing with different aggregation functions on the power flow edge and the control signal edge respectively, iteratively update the feature representation of each node, and output a node embedding vector that integrates the coupling features of physical process and information process.

[0008] In this embodiment, the rate of change of the total incoming power on the low-voltage side of the transformer is monitored in real time. When the absolute value of the rate of change exceeds a preset sudden change threshold, it is determined to be a load jump event, and the jump amplitude and duration are recorded. Specifically, this includes: The total incoming power on the low-voltage side of the transformer is collected at a fixed sampling period, and the rate of change at the current moment is determined. A mutation threshold is set. When the absolute value of the rate of change exceeds the mutation threshold for N consecutive samples, it is determined that a load jump event is triggered, where N is a preset threshold for the number of consecutive exceedances. Record the magnitude and duration of the load jump event.

[0009] In this embodiment, constructing a robust optimization model with chance constraints specifically includes: Based on the real-time electrical measurement data of each energy storage device, the upper and lower limits of the available charging and discharging power of the energy storage system are determined; An objective function is constructed with the charging power and discharging power of the energy storage system as decision variables and minimizing the daily operating cost as the objective. Opportunity constraints are introduced to describe uncertain load fluctuations caused by load jump events; The opportunity constraint is transformed into a deterministic equivalent form, and the uncertainty distribution is empirically estimated using historical load jump event data, thus transforming the probability constraint into a deterministic inequality constraint under a finite number of sampling scenarios. By integrating the objective function with the transformed opportunity constraints, a robust optimization model is constructed, which outputs the feasible region boundary and cost coefficients.

[0010] In this embodiment, the robust optimization model refers to an optimization model that uses the charging and discharging power of the energy storage system as the decision variable, minimizes the daily operating cost as the objective function, and includes opportunity constraints.

[0011] In this embodiment, the charging and discharging strategies for each energy storage converter in the industrial and commercial park to reach Nash equilibrium are obtained by performing a potential game based on the robust optimization model, specifically including: Acquire local decision variables and local observation information of each energy storage converter in the industrial and commercial park. The local decision variables include charging power and discharging power, and the local observation information includes state of charge and health status. Opportunity constraints are extracted from the robust optimization model as global coupling constraints; Under the premise of satisfying the global coupling constraints, the charging and discharging strategies of each energy storage converter are iteratively updated with the goal of minimizing the battery cycle life loss of each energy storage converter, and then the charging and discharging strategies are output.

[0012] In this embodiment, under the premise of satisfying the global coupling constraint, the charging and discharging strategies of each energy storage converter are iteratively updated with the goal of minimizing the battery cycle life loss of each energy storage converter, and the resulting charging and discharging strategies specifically include: The charging and discharging strategies of each energy storage converter are solved iteratively. In each iteration, only the marginal cost information of each energy storage converter is exchanged. The marginal cost information is the additional cost increment generated by each energy storage converter by increasing its current charging and discharging power by a unit power. Based on the marginal cost information exchanged by each energy storage converter, the charging and discharging strategies of each energy storage converter are updated synchronously. The update rule is: each energy storage converter independently updates its local charging and discharging power based on minimizing the sum of its own battery cycle life loss sub-objective and penalty term. Then, the dual variables of each energy storage converter are updated based on the average value of the decision variables of all energy storage converters. The difference between the potential function values ​​of two adjacent iterations is determined. When the difference is less than a preset convergence threshold, the iteration is stopped, and the charging and discharging power command of each energy storage converter when it reaches Nash equilibrium is output.

[0013] In this embodiment, the charging and discharging strategy includes: charging power commands and discharging power commands for each energy storage converter in each control cycle within the future control time domain.

[0014] Secondly, this application provides an operation optimization control system for commercial and industrial energy storage equipment, used to execute an operation optimization control method for commercial and industrial energy storage equipment, the optimization control system comprising: The data acquisition module is used to acquire the physical connection relationships and real-time electrical measurement data of various energy storage devices, photovoltaic units, transformers and key load nodes in the industrial and commercial park, and then construct a heterogeneous graph neural network model, embedding the real-time electrical measurement data as the initial features of the nodes into the heterogeneous graph neural network model; The robust optimization module is used to monitor the rate of change of the total incoming power on the low-voltage side of the transformer in real time. When the absolute value of the rate of change exceeds the preset sudden change threshold, it is determined as a load jump event, and the jump amplitude and duration are recorded. Based on the real-time electrical measurement data of each energy storage device, the upper and lower limits of the available charging and discharging power of the energy storage system are determined, and then a robust optimization model containing chance constraints is constructed. The strategy output module is used to perform potential game theory on each energy storage converter in the industrial and commercial park based on the robust optimization model to obtain the charging and discharging strategy when each energy storage converter reaches Nash equilibrium.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The system acquires the physical connections and real-time electrical measurement data of various energy storage devices, photovoltaic units, transformers, and key load nodes within the industrial and commercial park. A heterogeneous graph neural network model is then constructed, embedding the real-time electrical measurement data as initial node features into this model. The system monitors the rate of change of the total incoming power on the low-voltage side of the transformer in real time. When the absolute value of this rate of change exceeds a preset abrupt change threshold, it is identified as a load jump event, and the jump amplitude and duration are recorded. Based on the real-time electrical measurement data of each energy storage device, the upper and lower limits of the available charging and discharging power of the energy storage system are determined, and a robust optimization model with chance constraints is constructed. Based on the robust optimization model, a potential game is performed on each energy storage converter within the industrial and commercial park to obtain the charging and discharging strategies for each energy storage converter when it reaches Nash equilibrium.

[0016] Therefore, this application firstly constructs a heterogeneous graph neural network model with physical connections as edges and energy storage devices, photovoltaic units, transformers, and key load nodes as vertices, embedding real-time electrical measurement data as initial node features. This achieves a unified graph representation of all elements in the industrial and commercial park, simultaneously capturing the physical coupling relationship of power flow edges and the information coupling relationship of control signal edges. This provides a collaborative perception foundation integrating physical and information processes for accurate perception of subsequent load jump events, scope of impact estimation, and distributed game decision-making. Secondly, by real-time monitoring of the rate of change of the total incoming power on the low-voltage side of the transformer and employing a continuous over-limit judgment mechanism to trigger load jump event recording, the false trigger rate and computational overhead are reduced. This constructs a model containing opportunity factors... The robust optimization model of the bundle transforms the uncertain fluctuations caused by load jump events into a probabilistic feasible region. It achieves adaptive setting of safety margin under the flexibility of allowing a certain probability of constraint violation, thus resolving the contradiction in impact load scenarios where either excessive conservatism leads to poor economic efficiency or underestimation leads to decreased power supply reliability. Finally, by treating the chance constraints in the robust optimization model as global coupling constraints, and using the minimization of the battery cycle life loss of each energy storage converter as a sub-objective for potential game iterative solution, the charging and discharging strategies of each energy storage converter converge to Nash equilibrium under the premise of exchanging only marginal cost information. This achieves the unity of individual economic efficiency and global security, and reduces the communication burden and privacy leakage risk in the distributed collaborative optimization of multiple energy storage devices.

[0017] In summary, the technical solution adopted in this application can improve the power supply reliability and operational economy of industrial and commercial energy storage through load jump event perception and distributed solution of potential game driven by heterogeneous graph neural network. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application 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 only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is an exemplary flowchart of an operation optimization control method for industrial and commercial energy storage equipment provided in this application; Figure 2 This is a message passing diagram of a heterogeneous graph neural network provided in this application; Figure 3 This is a flowchart illustrating the iterative solution of the potential game provided in this application; Figure 4 This is a modular structure diagram of an operation optimization control system for industrial and commercial energy storage equipment provided in this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] This application provides a method and system for optimizing the operation of energy storage equipment in industrial and commercial parks. The core of this method is to acquire the physical connections and real-time electrical measurement data of various energy storage devices, photovoltaic units, transformers, and key load nodes within the industrial and commercial park, and then construct a heterogeneous graph neural network model. The real-time electrical measurement data is embedded into the heterogeneous graph neural network model as initial features of the nodes. The method monitors the rate of change of the total incoming power on the low-voltage side of the transformer in real time. When the absolute value of the rate of change exceeds a preset abrupt change threshold, it is determined as a load jump event, and the jump amplitude and duration are recorded. Based on the real-time electrical measurement data of each energy storage device, the upper and lower limits of the available charging and discharging power of the energy storage system are determined, and a robust optimization model containing chance constraints is constructed. Based on the robust optimization model, a potential game is performed on each energy storage converter within the industrial and commercial park to obtain the charging and discharging strategy when each energy storage converter reaches Nash equilibrium.

[0022] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1As shown in the figure, this is an exemplary flowchart of an operation optimization control method for commercial and industrial energy storage equipment according to this embodiment of the present application. The optimization control method includes the following steps: In step S1, the physical connection relationships and real-time electrical measurement data of each energy storage device, photovoltaic unit, transformer and key load node in the industrial and commercial park are obtained, and then a heterogeneous graph neural network model is constructed. The real-time electrical measurement data is embedded into the heterogeneous graph neural network model as the initial feature of the node.

[0023] In practical implementation, the physical connection relationships and real-time electrical measurement data of various energy storage devices, photovoltaic units, transformers, and key load nodes within the industrial and commercial park can be obtained. Specifically, the primary electrical wiring diagram and communication topology diagram of the industrial and commercial park are first acquired. The primary electrical wiring diagram records the physical connection relationships between various energy storage devices, photovoltaic units, transformers, and key load nodes within the park, including feeder connection paths between each device, line impedance parameters, the configuration locations of circuit breakers and disconnect switches, and the electrical distance between each node. The communication topology diagram records the communication link relationships between the corresponding smart terminals of each device, including communication protocol type, communication medium, data transmission direction, and communication delay characteristics. Real-time electrical measurement data of each node is collected by smart acquisition terminals deployed on-site within the park. For energy storage devices, the terminal voltage, charging and discharging current, state of charge (SOC), state of health, and battery temperature of each battery cluster connected to each energy storage inverter are collected through the battery management system. The SOC is estimated using a fusion of the ampere-hour integral method and Kalman filtering, while the state of health is comprehensively evaluated based on the battery cycle count and internal resistance detection results. For photovoltaic units, the data is collected by the photovoltaic inverter controller. The system monitors the output power, DC-side voltage, and ambient irradiance of a photovoltaic (PV) string. Irradiance is obtained from an irradiance meter installed near the PV array and used to predict the trend of PV output changes. For the transformer, the system collects active power, reactive power, power factor, and winding temperature data using smart meters installed at the transformer's low-voltage main input line. Active and reactive power are used to calculate the rate of change of total input power, the power factor is used to assess the reactive power compensation needs of the industrial park, and the winding temperature is used to monitor whether the transformer is overloaded. For critical load nodes, data is collected from various feeder branch boxes. Smart meters collect active power, reactive power, and current harmonic distortion rate from each key load node. The harmonic distortion rate is used to assess the power quality characteristics of the load. Then, timestamp alignment and data preprocessing are performed on all the above data. A linear interpolation method is used to resample all the data to the same control cycle frequency. For missing or abnormal data, the previous valid sample value is used to fill the missing data or the data is repaired by moving average filtering based on historical data. This allows us to obtain the physical connection relationship and real-time electrical measurement data of each energy storage device, photovoltaic unit, transformer, and key load node in the industrial and commercial park.

[0024] In this embodiment, constructing a heterogeneous graph neural network model and embedding the real-time electrical measurement data as the initial feature of the nodes into the heterogeneous graph neural network model can be achieved through the following steps: Using the physical connections as edges and each energy storage device, photovoltaic unit, transformer, and key load node as vertices, a basic graph structure for a heterogeneous graph neural network model is constructed. The initial feature vector of each vertex is determined based on the real-time electrical measurement data; The heterogeneous graph neural network model is configured with power flow edges and control signal edges. The weights of the power flow edges are determined by the line impedance parameters, and the weights of the control signal edges are determined by the current communication delay and data packet loss rate. A graph attention mechanism is adopted to perform message passing with different aggregation functions on the power flow edge and the control signal edge respectively, iteratively update the feature representation of each node, and output a node embedding vector that integrates the coupling features of physical process and information process.

[0025] In practical implementation, firstly, using physical connections as edges and each energy storage device, photovoltaic unit, transformer, and key load node as vertices, a basic graph structure for the heterogeneous graph neural network model is constructed. Specifically, each energy storage converter and its connected battery cluster within the park is defined as an energy storage device vertex, each photovoltaic string and its inverter as a photovoltaic unit vertex, each distribution transformer as a transformer vertex, and the key electrical equipment at the end of each feeder as a load node vertex. The physical connections between vertices, i.e., the feeder connection paths, are defined as the edges of the graph structure. The resulting basic graph structure is an undirected graph with the total number of vertices equal to the total number of vertices within the park. The sum of the number of the four types of equipment is equal to the number of feeder connections between each piece of equipment. Secondly, the initial feature vector of each vertex can be determined based on real-time electrical measurement data. That is, for the vertex of the energy storage equipment, the initial feature vector includes: terminal voltage, charging and discharging current, state of charge, health state and battery temperature; for the vertex of the photovoltaic unit, the initial feature vector includes: output power, DC side voltage and irradiance; for the vertex of the transformer, the initial feature vector includes: active power, reactive power, power factor and winding temperature; for the vertex of the load node, the initial feature vector includes: active power, reactive power and current harmonic distortion rate, thus obtaining the initial feature vector of each vertex.Then, power flow edges and control signal edges can be configured for the heterogeneous graph neural network model. The weights of the power flow edges are determined by the line impedance parameters, while the weights of the control signal edges are determined by the current communication delay and data packet loss rate. That is, for the power flow edge, its weight is determined by the line impedance parameters of the feeder connecting the two vertices. The line impedance parameters include line resistance and line reactance, which can be obtained from the primary wiring diagram of the park's electrical system. For the feeder connecting vertex i and vertex j, the unit length resistance value, unit length reactance value, and line length of the feeder are obtained. The unit length resistance value depends on the material and cross-sectional area of ​​the feeder conductor, and the unit length reactance value depends on the feeder's laying method and phase-to-phase distance. Multiplying the unit length resistance value by the line length yields the total line resistance, and multiplying the unit length reactance value by the line length yields the total line reactance. Adding the total line resistance and total line reactance yields the total line impedance value. The reciprocal of the total line impedance, i.e., the admittance value, is then used as the power flow edge. The weight of a flow edge is determined by its admittance value. A larger admittance value results in a larger weight, indicating stronger electrical coupling between the two vertices. Conversely, a smaller admittance value results in a smaller weight, indicating weaker electrical coupling. For two vertices within the same park without a direct feeder connection, there is no power flow edge between them, and the weight is zero. For control signal edges, the weight is determined by the current communication delay and data loss rate. The current communication delay and data loss rate are obtained, and the probability of successful data packet transmission is calculated (1 minus the data loss rate). Then, the attenuation factor of the communication delay is calculated. This is done by setting a reference time constant (typically 50 milliseconds) based on the performance requirements of the park's communication network, calculating the ratio of the communication delay to the reference time constant, and then using the natural constant e as the base and the opposite of this ratio as the exponent to calculate the exponential function value. The result is used as the attenuation factor of the communication delay. The weight of the control signal edge is obtained by multiplying the probability of successful data packet transmission by the attenuation factor of the communication delay.

[0026] In addition, in specific implementation, a graph attention mechanism can be used to perform message passing with different aggregation functions on the power flow edge and the control signal edge respectively, iteratively updating the feature representation of each node, and outputting a node embedding vector that integrates the coupling features of the physical process and the information process. That is, for each node, for the power flow edge, all its neighboring nodes connected through the power flow edge are determined. These neighboring nodes are devices that have direct feeder connections with the node. The attention coefficient between the node and each power flow neighboring node is calculated, that is, the feature vector of the node is concatenated with the feature vectors of the neighboring nodes, and then passed through a single-layer feedforward neural network. The unnormalized attention coefficient is mapped to a scalar value and then multiplied by the power flow edge weight. A larger power flow edge weight results in a larger unnormalized attention coefficient, indicating that neighboring nodes with stronger electrical coupling receive more attention. The unnormalized attention coefficient is then normalized by summing the unnormalized attention coefficients of all power flow neighbors of the given node. The normalized attention coefficient is then divided by the sum of the unnormalized attention coefficients of each neighbor node, resulting in a normalized attention coefficient. The normalized attention coefficient is between 0 and 1, and the sum of the attention coefficients of all power flow neighbors is 1. Finally, each power flow... The eigenvectors of neighboring nodes are multiplied by their corresponding normalized attention coefficients. Then, the weighted eigenvectors of all neighboring nodes are summed to obtain aggregated neighbor information. This aggregated neighbor information is then superimposed on the node's own eigenvector and transformed using a nonlinear activation function to obtain the physical process characteristic components of the node on the power flow edge. For the control signal edge, for each node, all its neighboring nodes connected through the control signal edge are identified. These neighboring nodes are intelligent terminals with communication links to the node. The attention coefficient between the node and each control signal neighboring node is calculated, similar to the calculation for the power flow edge. The feature vector of a point is concatenated with the feature vectors of its neighboring nodes, mapped to a scalar value through a single-layer feedforward neural network, and then multiplied by the control signal edge weights to obtain an unnormalized attention coefficient. A gating mechanism is then introduced for selective aggregation. The core of the gating mechanism is to determine whether the information of each control signal neighboring node is worth adopting. The specific criterion is whether the control signal edge weights are greater than a preset gating threshold. The gating threshold is set according to the performance requirements of the campus communication network. If the control signal edge weights are greater than the gating threshold, it indicates that the communication latency is short and the data packet loss rate is low, the communication link quality is good, the gating is opened, and the information of that neighboring node is allowed to enter the aggregation process.If the control signal edge weight is less than the gating threshold, it indicates a long communication delay or a high data packet loss rate, resulting in poor communication link quality. The gating is closed, and the information of this neighbor node is masked and does not participate in the aggregation process. For neighbor nodes with gating enabled, their unnormalized attention coefficients are normalized. The unnormalized attention coefficients of all gating-enabled control signal neighbor nodes are summed, and then the unnormalized attention coefficient of each gating-enabled neighbor node is divided by this sum to obtain the normalized attention coefficient. The feature vector of each gating-enabled control signal neighbor node is then multiplied by its corresponding normalized attention coefficient. The weighted feature vectors of all gating-enabled neighbor nodes are then summed to obtain the aggregated neighbor information. This aggregated neighbor information is superimposed on the node's own feature vector and then transformed using a nonlinear activation function to obtain the information process feature components of the node on the control signal edge. Finally, the physical process feature components are combined with the information process... Feature components are concatenated to form a fused feature vector. Then, a fully connected layer compresses the dimension of the fused feature vector to a preset hidden layer dimension. Finally, after layer normalization, the updated node feature representation is output. This process is completed in one layer of a graph attention network. The output node feature representation is then used as input to the next layer of the graph attention network, repeating the steps of power flow edge aggregation, control signal edge aggregation, and feature fusion. Through multiple iterations (three layers in this embodiment), the node's feature representation gradually incorporates information from higher-order neighbor nodes. The first layer of the graph attention network enables nodes to perceive information from their direct neighbors; the second layer enables nodes to perceive information from neighbors within a two-hop distance; and the third layer enables nodes to perceive information from neighbors within a three-hop distance. After iterative updates through multiple layers of the graph attention network, the node embedding vector for each node is finally output. This node embedding vector is a high-dimensional vector representation that integrates the coupled features of the physical and information processes.

[0027] like Figure 2As shown in the schematic diagram of message passing in the heterogeneous graph neural network provided in this application, the heterogeneous graph neural network model uses each energy storage device, photovoltaic unit, transformer, and key load node as vertices, and uses the power flow edge determined by the physical connection relationship and the control signal edge determined by the communication connection relationship as edges to construct the heterogeneous graph basic structure. Among them, the weight of the power flow edge is determined by the line impedance parameter, and the weight of the control signal edge is determined by the current communication delay and data packet loss rate. The message passing process is divided into three layers: In the first layer, each node receives the electrical measurement features of the adjacent nodes along the power flow edge, and aggregates them using an aggregation function based on impedance weights to output the physical process feature vector; In the second layer, each node receives the control command features of the adjacent nodes along the control signal edge, and aggregates them using an aggregation function based on attention coefficients to output the information process feature vector; In the third layer, the physical process feature vector and the information process feature vector of the same node are concatenated, and after mapping by a fully connected layer, the node embedding vector that integrates the coupling features of the physical process and the information process is output. In the figure, the nodes with bold borders represent the target nodes of the current message passing.

[0028] In step S2, the rate of change of the total incoming power on the low-voltage side of the transformer is monitored in real time. When the absolute value of the rate of change exceeds the preset sudden change threshold, it is determined as a load jump event, and the jump amplitude and duration are recorded. The upper and lower limits of the available charging and discharging power of the energy storage system are determined based on the real-time electrical measurement data of each energy storage device, and then a robust optimization model containing chance constraints is constructed.

[0029] In this embodiment, the rate of change of the total incoming power on the low-voltage side of the transformer is monitored in real time. When the absolute value of the rate of change exceeds a preset sudden change threshold, it is determined to be a load jump event, and the jump amplitude and duration are recorded. This can be achieved through the following steps: The total incoming power on the low-voltage side of the transformer is collected at a fixed sampling period, and the rate of change at the current moment is determined. A mutation threshold is set. When the absolute value of the rate of change exceeds the mutation threshold for N consecutive samples, it is determined that a load jump event is triggered, where N is a preset threshold for the number of consecutive exceedances. Record the magnitude and duration of the load jump event.

[0030] In practical implementation, firstly, the total incoming power on the low-voltage side of the transformer is collected at a fixed sampling period, and the rate of change at the current moment is determined. Specifically, a high-precision smart meter is installed at the total incoming line on the low-voltage side of the transformer. This smart meter can measure the total active power flowing through the low-voltage side of the transformer in real time. The smart meter collects the total incoming power data at a fixed sampling period, typically 1 second, meaning the total incoming power value at the current moment is recorded once per second. The sampling period can be adjusted according to the load characteristics and control accuracy requirements of the industrial park. Each time new sampling data arrives, the total incoming power value collected at the current moment is subtracted from the total incoming power value at the previous sampling moment. The power value is used to obtain the change in power. This change is then divided by the sampling period to obtain the power change rate per unit time. A sudden change threshold can then be set. When the absolute value of the change rate exceeds the threshold for N consecutive samplings, a load jump event is triggered, where N is a preset threshold for the number of consecutive over-limits. The sudden change threshold is preset based on the transformer's rated capacity and is equal to a proportional coefficient multiplied by the transformer's rated capacity. For example, for a transformer with a rated capacity of 1000 kVA, if the proportional coefficient is 0.1, the sudden change threshold is 100 kW / s. That is, when the absolute value of the change rate of the total incoming power exceeds 1... When the power change rate reaches 00 kW / s, it is considered abnormal, and a load jump event may exist. A load jump event is triggered when the absolute value of the change rate exceeds the sudden change threshold for N consecutive sampling periods. Here, N is a preset threshold for the number of consecutive exceedances; in this embodiment, it is set to 3 times. That is, a valid load jump event is only confirmed when the absolute value of the change rate exceeds the sudden change threshold for three consecutive sampling periods. Finally, the jump amplitude and duration of the load jump event can be recorded. That is, after a load jump event is determined, the event trigger time is determined. The event trigger time is the sampling period when the absolute value of the change rate exceeds the sudden change threshold for the Nth consecutive time. The sampling time is determined by first identifying the stable operating time before the event occurs. The stable operating time is the time point before the event triggering time when the absolute value of the power change rate has not yet exceeded the mutation threshold. Usually, it is the (N+1)th sampling time before the event triggering. The total incoming power value at the event triggering time is subtracted from the total incoming power value at the stable operating time, and the absolute value is taken to obtain the jump amplitude. The start time of the duration is the sampling time when the absolute value of the change rate first exceeds the mutation threshold, and the end time of the duration is the sampling time when the absolute value of the change rate is lower than half of the mutation threshold for the first two consecutive sampling periods after exceeding the mutation threshold. The length of time that has elapsed is the duration.

[0031] Preferably, in this embodiment, the upper and lower limits of the available charging and discharging power of the energy storage system are determined based on the real-time electrical measurement data of each energy storage device, and then a robust optimization model including chance constraints is constructed. This can be achieved through the following steps: Based on the real-time electrical measurement data of each energy storage device, the upper and lower limits of the available charging and discharging power of the energy storage system are determined; An objective function is constructed with the charging power and discharging power of the energy storage system as decision variables and minimizing the daily operating cost as the objective. Opportunity constraints are introduced to describe uncertain load fluctuations caused by load jump events; The opportunity constraint is transformed into a deterministic equivalent form, and the uncertainty distribution is empirically estimated using historical load jump event data, thus transforming the probability constraint into a deterministic inequality constraint under a finite number of sampling scenarios. By integrating the objective function with the transformed opportunity constraints, a robust optimization model is constructed, which outputs the feasible region boundary and cost coefficients.

[0032] In practical implementation, firstly, the upper and lower limits of the available charging and discharging power of the energy storage system can be determined based on the real-time electrical measurement data of each energy storage device. That is, the feature information of the vertices of each energy storage device can be extracted from the node embedding vectors output by the heterogeneous graph neural network model. For each energy storage converter, its charging power limit is determined by the following factors: the rated power of the energy storage converter itself, the difference between the current state of charge and the maximum allowable state of charge, the rated capacity of the battery, and the charging efficiency. The charging power limit cannot exceed the rated power, nor can it exceed the power required to charge the battery from the current state of charge to the maximum allowable state of charge in a single control cycle. If the current state of charge has reached or exceeded the maximum allowable state of charge, the charging power limit is forcibly set to zero, that is, charging is not allowed to continue. The upper limit of the discharge power of an energy storage converter is determined by the following factors: the rated power of the energy storage converter itself, the difference between the current state of charge and the minimum allowable state of charge, the rated capacity of the battery, and the discharge efficiency. The upper limit of the discharge power cannot exceed the rated power, nor can it exceed the power that can be released by discharging the battery from the current state of charge to the minimum allowable state of charge in a single control cycle. If the current state of charge has reached or fallen below the minimum allowable state of charge, the upper limit of the discharge power is forcibly set to zero, that is, further discharge is not allowed. The upper limit of the charging power of all energy storage converters is added together to obtain the upper limit of the total charging power of the entire energy storage system. The upper limit of the discharge power of all energy storage converters is added together to obtain the upper limit of the total discharge power of the entire energy storage system. The lower limit of the available charging and discharging power of the entire energy storage system is 0.Secondly, the charging and discharging power of the energy storage system can be used as decision variables, and an objective function can be constructed with the goal of minimizing daily operating costs. Daily operating costs consist of three components: demand electricity cost, negative peak-valley arbitrage revenue, and energy storage depreciation cost. Demand electricity cost refers to the electricity charge based on the maximum demand of the industrial and commercial park during the billing cycle. This cost is calculated using a sliding window maximum demand prediction algorithm. The embedding vectors of each node output from the heterogeneous graph neural network model are used as input, connected to a fully connected regression layer, and the output is the load forecast value for future periods. These future load forecast values ​​form a basic load forecast sequence. Based on this sequence, the average peak power within the future sliding time window is simulated. This peak power is multiplied by the demand electricity price to obtain an estimate of the demand electricity cost. Negative peak-valley arbitrage revenue refers to the revenue gained by the energy storage system through charging during off-peak hours and discharging during peak hours. The negative of the profit is calculated by multiplying the net discharge power (total discharge power minus total charging power) of the energy storage system within each control cycle by the current electricity price, and then by the control cycle duration. This yields the arbitrage profit for that cycle. Summing the arbitrage profits from all control cycles and taking the negative value yields a negative peak-valley arbitrage profit. A smaller value indicates a larger arbitrage profit. Energy storage depreciation cost refers to the lifespan depreciation cost incurred by the energy storage system due to battery cycle aging during charging and discharging. The depreciation cost of each energy storage converter is related to its charging and discharging power and current health status. Higher charging and discharging power leads to more severe battery cycle aging and higher depreciation costs. Lower health status and closer to the end of battery life result in higher depreciation costs per charge and discharge cycle. Summing the depreciation costs of all energy storage converters yields the total energy storage depreciation cost. The objective function is constructed by adding these three components together and minimizing this sum.

[0033] In addition, in practical implementation, opportunity constraints can be introduced to describe the uncertain load fluctuations caused by load jump events. That is, the opportunity constraint requires that the probability of an event in which the grid power supply plus the total energy storage discharge power minus the total energy storage charging power is greater than the actual load plus the safety margin is not less than a preset risk level threshold, which is a threshold between 0 and 0.For values ​​between 1 and 1, when the risk level threshold is 0, the opportunity constraint degenerates into a deterministic constraint; when the risk level threshold is larger, the system has a higher tolerance for uncertainty. Then, the opportunity constraint can be transformed into a deterministic equivalent form. Historical load shift event data can be used to empirically estimate the uncertainty distribution, transforming the probabilistic constraint into a deterministic inequality constraint under a finite number of sampling scenarios. That is, historical load shift event data of the park is collected, including the magnitude and duration of each shift, as well as the load waveform before and after the event. This historical data is used as samples of the uncertainty distribution, and these samples are used to empirically estimate the uncertainty distribution, assuming possible future... If load jump events follow the same distribution as historical data, various possible future scenarios can be simulated by sampling historical data. A certain number of scenarios are randomly selected from the historical data, each corresponding to a possible load jump pattern. For each sampled scenario, the random variables in the chance constraints are replaced with specific values ​​specific to that scenario, transforming the probabilistic constraints into deterministic inequality constraints for that scenario. Then, the deterministic inequality constraints from all sampled scenarios are merged to obtain a set of deterministic inequality constraints. When the number of sampled scenarios is large enough, this set of deterministic constraints can approximate the original chance constraints with arbitrary precision. Finally, the objective function and the transformed... To address opportunity constraints, a robust optimization model is constructed, outputting the feasible region boundary and cost coefficients. Specifically, the robust optimization model includes: decision variables are the charging and discharging power of each energy storage converter in each control cycle; the objective function is to minimize the daily operating cost, which is the sum of demand electricity cost, negative peak-valley arbitrage revenue, and energy storage depreciation cost; constraints include: upper and lower power limits (the charging and discharging power of each energy storage converter cannot exceed the available charging and discharging power limits); state of charge (SOC) constraints (the SOC of each energy storage converter must always be maintained between the minimum and maximum allowable SOC during operation); and deterministic inequality constraints transformed from opportunity constraints. In each sampling scenario, the sum of the grid power supply and the total energy storage discharge power minus the total energy storage charging power must be greater than or equal to the actual load plus the safety margin in that scenario. The robust optimization model output includes: the feasible region boundary and cost coefficients. The feasible region boundary refers to the range of values ​​defined by the constraints that all decision variables must satisfy, i.e., the feasible interval for the charging and discharging power of each energy storage converter. The cost coefficients are the weight parameters of each component in the objective function, including demand electricity price, time-of-use electricity price, and energy storage loss coefficient. The robust optimization model is an optimization model that uses the charging and discharging power of the energy storage system as the decision variable, minimizes the daily operating cost as the objective function, and includes opportunity constraints.

[0034] It should be noted that by monitoring the rate of change of the total incoming power on the low-voltage side of the transformer in real time and using a continuous over-limit judgment mechanism to trigger load jump event recording, the false trigger rate and computational overhead are reduced. A robust optimization model containing chance constraints is constructed, which transforms the uncertain fluctuations caused by load jump events into a probabilistic feasible region. Under the flexibility of allowing a certain probability of constraint violation, the adaptive setting of the safety margin is achieved, which solves the contradiction in the case of impact load scenarios where either excessive conservatism leads to poor economy or insufficient estimation leads to a decrease in power supply reliability.

[0035] In step S3, the charging and discharging strategies of each energy storage converter in the industrial and commercial park are obtained by performing a potential game on each energy storage converter according to the robust optimization model to achieve Nash equilibrium.

[0036] In this embodiment, the charging and discharging strategies for each energy storage converter in the industrial and commercial park to reach Nash equilibrium are obtained by performing a potential game based on the robust optimization model, which can be implemented using the following steps: Acquire local decision variables and local observation information of each energy storage converter in the industrial and commercial park. The local decision variables include charging power and discharging power, and the local observation information includes state of charge and health status. Opportunity constraints are extracted from the robust optimization model as global coupling constraints; Under the premise of satisfying the global coupling constraints, the charging and discharging strategies of each energy storage converter are iteratively updated with the goal of minimizing the battery cycle life loss of each energy storage converter, and then the charging and discharging strategies are output.

[0037] In practical implementation, local decision variables and local observation information of each energy storage converter within the industrial and commercial park can be obtained. The local decision variables include charging power and discharging power, and the local observation information includes state of charge and health status. Specifically, local decision variables refer to the control quantities that each energy storage converter can autonomously adjust, including charging power and discharging power; local observation information refers to the state quantities that each energy storage converter can directly measure through local sensors or obtain through local calculations, including state of charge and health status. Secondly, opportunity constraints can be extracted from the robust optimization model as global coupling constraints. That is, the global coupling constraints require that the grid power supply plus all energy storage converters... The total discharge power of the converter minus the total charging power of all energy storage converters must be greater than or equal to the actual load of the park plus the safety margin. This inequality holds with a probability not lower than the preset risk level threshold. This constraint couples the charging and discharging decisions of each energy storage converter together through the power balance relationship. Any change in the decision of any energy storage converter will affect the ability of other energy storage converters to meet this constraint. Then, a potential function can be constructed, which is set as the negative of the sum of the total operating cost of the park and the total demand electricity cost. The design of the potential function follows the principle that when the value of the potential function increases, at least one energy storage converter's individual sub-objective is improved, and the individual sub-objectives of other energy storage converters do not deteriorate.

[0038] Finally, in this embodiment, under the premise of satisfying the global coupling constraint, the charging and discharging strategies of each energy storage converter are iteratively updated with the goal of minimizing the battery cycle life loss of each energy storage converter, and the charging and discharging strategies are output. The specific steps for implementing the charging and discharging strategies are as follows: The charging and discharging strategies of each energy storage converter are solved iteratively. In each iteration, only the marginal cost information of each energy storage converter is exchanged. The marginal cost information is the additional cost increment generated by each energy storage converter by increasing its current charging and discharging power by a unit power. Based on the marginal cost information exchanged by each energy storage converter, the charging and discharging strategies of each energy storage converter are updated synchronously. The update rule is: each energy storage converter independently updates its local charging and discharging power based on minimizing the sum of its own battery cycle life loss sub-objective and penalty term. Then, the dual variables of each energy storage converter are updated based on the average value of the decision variables of all energy storage converters. The difference between the potential function values ​​of two adjacent iterations is determined. When the difference is less than a preset convergence threshold, the iteration is stopped, and the charging and discharging strategies of each energy storage converter when Nash equilibrium is reached are output.

[0039] In practical implementation, firstly, the charging and discharging strategies of each energy storage converter can be iteratively solved. In each iteration, only the marginal cost information of each energy storage converter is exchanged. That is, before the iterative solution process begins, the charging and discharging strategies of each energy storage converter are initialized. The initial strategy can be set to none of the energy storage converters charging or discharging, or the total charging and discharging power calculated by the robust optimization model can be initially allocated according to the rated power ratio of each energy storage converter. After entering the iteration loop, in each iteration, the energy storage converters only exchange marginal cost information, without exchanging complete historical data on the state of charge or health status information. Marginal cost information refers to the marginal cost information of each energy storage converter in its current state of charge. The marginal cost is the additional cost per unit of power added to the charging and discharging power. For example, for an energy storage converter, if it is currently discharging, the marginal cost represents the increased battery life loss cost for discharging one more kilowatt-hour; if it is currently charging, the marginal cost represents the increased battery life loss cost for charging one more kilowatt-hour. The magnitude of the marginal cost is closely related to the current state of charge and health of the energy storage converter: the closer the state of charge is to the boundary and the lower the health status, the higher the marginal cost. Then, based on the marginal cost information exchanged between each energy storage converter, the charging and discharging strategies of each energy storage converter can be updated synchronously. The update rule is: each energy storage converter uses its own battery cycle life loss sub-target and... The criterion is to minimize the sum of penalty terms. Local charging and discharging power is updated independently. Then, the dual variables of each energy storage converter are updated based on the average of the decision variables of all energy storage converters. That is, each energy storage converter needs to solve a local sub-optimization problem when updating its local decision. The inputs to this sub-optimization problem include: the current state of charge and health of the local converter, the local charging and discharging strategy from the previous iteration, the local dual variables from the previous iteration, and the average marginal cost information received from other energy storage converters. The objective of the sub-optimization problem is to minimize the sum of local battery cycle life loss and penalty terms. Constraints include local power upper and lower limits and state of charge constraints. In the variable update phase, the average value of the decision variables of all energy storage converters is first calculated. That is, each energy storage converter broadcasts the updated net discharge power to other energy storage converters. After receiving the net discharge power of all other energy storage converters, each energy storage converter calculates the arithmetic mean. Each energy storage converter updates its own dual variable based on this average value. The update rule for the dual variable is: the new dual variable is equal to the old dual variable plus the difference between the local net discharge power and the global average net discharge power. The dual variable reflects the cumulative deviation between the local decision and the global average decision and is used to adjust the strength of the penalty term in the next iteration. The above two phases are executed in sequence to complete one iteration.Finally, the difference between the potential function values ​​of two adjacent iterations can be determined. The iteration stops when the difference is less than a preset convergence threshold, and the charging and discharging strategies for each energy storage converter when reaching Nash equilibrium are output. Specifically, after each iteration, the potential function value after the current iteration is calculated and compared with the potential function value of the previous iteration to determine the difference between the potential function values ​​of two adjacent iterations. The potential function is set as the negative of the sum of the total operating cost and the total electricity demand of the park. When the difference between the potential function values ​​of two adjacent iterations is less than the preset convergence threshold, the iteration process is considered to have converged, and the iteration stops, indicating that the change in the potential function value is very small, and the iteration continues. This will not bring significant economic improvements. The convergence threshold can be preset by expert advice. When the iteration stops, the charging and discharging strategies currently held by each energy storage converter are the charging and discharging strategies at the point of Nash equilibrium. In the Nash equilibrium state, no single energy storage converter can further reduce its battery cycle life loss by changing its charging and discharging strategy. That is, each energy storage converter has reached its optimal response while the strategies of other energy storage converters remain unchanged. The charging and discharging strategies include: the charging power command and discharging power command of each energy storage converter for each control cycle in the future control time domain.

[0040] like Figure 3 The diagram shown is a flowchart illustrating the iterative solution of the potential game provided in this embodiment. The iterative solution process of the potential game includes the following steps: Step S301, initialize the charging and discharging power, state of charge, health state, dual variables, and convergence threshold of each energy storage converter; Step S302, determine whether the current iteration number is less than the maximum iteration number and convergence has not been reached. If so, proceed to step S303; otherwise, jump to step S307; Step S303, each energy storage converter calculates its local marginal cost information and broadcasts it to other energy storage converters through a publish / subscribe protocol; Step S304, each energy storage converter calculates its own battery cycle life loss item. Using the criterion of minimizing the sum of the target and penalty terms, the local charging and discharging power is updated independently; in step S305, the dual variables of each energy storage converter are updated synchronously based on the average value of the updated decision variables of all energy storage converters; in step S306, the potential function value of the current iteration is calculated, and it is determined whether the difference between the current potential function value and the potential function value of the previous iteration is less than the preset convergence threshold. If so, it is determined that Nash equilibrium has been reached and the process proceeds to step S307; otherwise, the iteration count is incremented by 1 and the process returns to step S302; in step S307, the charging and discharging strategies of each energy storage converter when Nash equilibrium is reached are output, including the charging power command and discharging power command for each control cycle.

[0041] It should be noted that by treating the opportunity constraints in the robust optimization model as global coupling constraints, and taking the minimization of the battery cycle life loss of each energy storage converter as the sub-objective for iterative solution of potential game, the charging and discharging strategies of each energy storage converter converge to Nash equilibrium under the premise of only exchanging marginal cost information, the unity of individual economy and global security is achieved, and the communication burden and privacy leakage risk in the distributed collaborative optimization of multiple energy storage devices are reduced.

[0042] Therefore, this application firstly constructs a heterogeneous graph neural network model with physical connections as edges and energy storage devices, photovoltaic units, transformers, and key load nodes as vertices, embedding real-time electrical measurement data as initial node features. This achieves a unified graph representation of all elements in the industrial and commercial park, simultaneously capturing the physical coupling relationship of power flow edges and the information coupling relationship of control signal edges. This provides a collaborative perception foundation integrating physical and information processes for accurate perception of subsequent load jump events, scope of impact estimation, and distributed game decision-making. Secondly, by real-time monitoring of the rate of change of the total incoming power on the low-voltage side of the transformer and employing a continuous over-limit judgment mechanism to trigger load jump event recording, the false trigger rate and computational overhead are reduced. This constructs a model containing opportunity factors... The robust optimization model of the bundle transforms the uncertain fluctuations caused by load jump events into a probabilistic feasible region. It achieves adaptive setting of safety margin under the flexibility of allowing a certain probability of constraint violation, thus resolving the contradiction in impact load scenarios where either excessive conservatism leads to poor economic efficiency or underestimation leads to decreased power supply reliability. Finally, by treating the chance constraints in the robust optimization model as global coupling constraints, and using the minimization of the battery cycle life loss of each energy storage converter as a sub-objective for potential game iterative solution, the charging and discharging strategies of each energy storage converter converge to Nash equilibrium under the premise of exchanging only marginal cost information. This achieves the unity of individual economic efficiency and global security, and reduces the communication burden and privacy leakage risk in the distributed collaborative optimization of multiple energy storage devices.

[0043] In summary, the technical solution adopted in this application can improve the power supply reliability and operational economy of industrial and commercial energy storage through load jump event perception and distributed solution of potential game driven by heterogeneous graph neural network.

[0044] Example 2: This application provides an operation optimization control system for industrial and commercial energy storage equipment, referencing... Figure 4 As shown in the figure, this is a modular structure diagram of an operation optimization control system for industrial and commercial energy storage equipment according to this embodiment of the present application. The optimization control system includes: The data acquisition module 100 is used to acquire the physical connection relationship and real-time electrical measurement data of various energy storage devices, photovoltaic units, transformers and key load nodes in the industrial and commercial park, and then construct a heterogeneous graph neural network model, embedding the real-time electrical measurement data as the initial feature of the node into the heterogeneous graph neural network model; The robust optimization module 200 is used to monitor the rate of change of the total incoming power on the low-voltage side of the transformer in real time. When the absolute value of the rate of change exceeds the preset sudden change threshold, it is determined as a load jump event, and the jump amplitude and duration are recorded. Based on the real-time electrical measurement data of each energy storage device, the upper and lower limits of the available charging and discharging power of the energy storage system are determined, and then a robust optimization model containing chance constraints is constructed. The strategy output module 300 is used to perform potential game theory on each energy storage converter in the industrial and commercial park based on the robust optimization model to obtain the charging and discharging strategy when each energy storage converter reaches Nash equilibrium.

[0045] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0046] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0047] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for optimizing the operation and control of industrial and commercial energy storage equipment, characterized in that, The optimized control method includes the following steps: The physical connections and real-time electrical measurement data of various energy storage devices, photovoltaic units, transformers, and key load nodes within an industrial and commercial park are acquired. A heterogeneous graph neural network model is then constructed, embedding the real-time electrical measurement data as initial node features into the model. The model uses physical connections to determine power flow edges and communication connections to determine control signal edges. A graph attention mechanism is used to perform message passing with different aggregation functions on the two types of edges, outputting node embedding vectors that integrate the coupling features of physical and information processes. These node embedding vectors are then input into a sliding window maximum demand prediction algorithm to obtain a base load prediction sequence. Based on this sequence, an estimated demand electricity cost is determined. The system monitors the rate of change of the total incoming power on the low-voltage side of the transformer in real time. When the absolute value of the rate of change exceeds a preset abrupt change threshold, it is determined as a load jump event, and the jump amplitude and duration are recorded. Based on the real-time electrical measurement data of each energy storage device, the upper and lower limits of the available charging and discharging power of the energy storage system are determined. Using the charging and discharging power of the energy storage system as decision variables, an objective function is constructed with the goal of minimizing the daily operating cost, which includes an estimated value of the demand electricity cost. Opportunity constraints are introduced to describe the uncertain load fluctuations caused by load jump events. The opportunity constraints are transformed into a deterministic equivalent form, and the uncertainty distribution is empirically estimated using historical load jump event data. The probabilistic constraints are transformed into deterministic inequality constraints under a finite number of sampling scenarios. The objective function and the transformed opportunity constraints are integrated to construct a robust optimization model, which outputs the feasible region boundary and cost coefficient. The opportunity constraints in the robust optimization model serve as global coupling constraints to constrain the power balance relationship between the total charging and discharging power of each energy storage converter and the actual load of the park. The local decision variables and local observation information of each energy storage converter in the industrial and commercial park are obtained. The local decision variables include charging power and discharging power, and the local observation information includes state of charge and health status. The opportunity constraints are extracted from the robust optimization model as global coupling constraints. The global coupling constraint is used as the constraint condition that must be satisfied for power balance, and the feasible domain boundary is used as the value range constraint for the local decision variables of each energy storage converter. Under the premise of satisfying the global coupling constraint, the charging and discharging strategy of each energy storage converter is iteratively updated with the minimization of the battery cycle life loss of each energy storage converter as the sub-objective, and then the charging and discharging strategy is output.

2. The operation optimization control method for industrial and commercial energy storage equipment as described in claim 1, characterized in that, The real-time electrical measurement data includes: Terminal voltage, charging and discharging current, state of charge, health status, and battery temperature of each energy storage device; Output power, DC-side voltage, and irradiance of each photovoltaic unit; The active power, reactive power, power factor, and winding temperature of the transformer; Active power, reactive power, and current harmonic distortion rate at each key load node.

3. The operation optimization control method for industrial and commercial energy storage equipment as described in claim 1, characterized in that, Constructing a heterogeneous graph neural network model, specifically embedding the real-time electrical measurement data as initial node features into the heterogeneous graph neural network model, includes: Using the physical connections as edges and each energy storage device, photovoltaic unit, transformer, and key load node as vertices, a basic graph structure for a heterogeneous graph neural network model is constructed. The initial feature vector of each vertex is determined based on the real-time electrical measurement data; The heterogeneous graph neural network model is configured with power flow edges and control signal edges. The weights of the power flow edges are determined by the line impedance parameters, and the weights of the control signal edges are determined by the current communication delay and data packet loss rate. A graph attention mechanism is adopted to perform message passing with different aggregation functions on the power flow edge and the control signal edge respectively, iteratively update the feature representation of each node, and output a node embedding vector that integrates the coupling features of physical process and information process.

4. The operation optimization control method for industrial and commercial energy storage equipment as described in claim 1, characterized in that, The rate of change of total incoming power on the low-voltage side of the transformer is monitored in real time. When the absolute value of the rate of change exceeds a preset sudden change threshold, it is determined to be a load jump event, and the jump amplitude and duration are recorded. Specifically, this includes: The total incoming power on the low-voltage side of the transformer is collected at a fixed sampling period, and the rate of change at the current moment is determined. A mutation threshold is set. When the absolute value of the rate of change exceeds the mutation threshold for N consecutive samples, it is determined that a load jump event is triggered, where N is a preset threshold for the number of consecutive exceedances. Record the magnitude and duration of the load jump event.

5. The operation optimization control method for industrial and commercial energy storage equipment as described in claim 1, characterized in that, The robust optimization model refers to an optimization model that uses the charging and discharging power of the energy storage system as the decision variable, minimizes the daily operating cost as the objective function, and includes opportunity constraints.

6. The operation optimization control method for industrial and commercial energy storage equipment as described in claim 1, characterized in that, Under the premise of satisfying the global coupling constraints, with the minimization of the battery cycle life loss of each energy storage converter as the sub-objective, the charging and discharging strategies of each energy storage converter are iteratively updated, and the charging and discharging strategies are output, specifically including: The charging and discharging strategies of each energy storage converter are solved iteratively. In each iteration, only the marginal cost information of each energy storage converter is exchanged. The marginal cost information is the additional cost increment generated by each energy storage converter by increasing its current charging and discharging power by a unit power. Based on the marginal cost information exchanged by each energy storage converter, the charging and discharging strategies of each energy storage converter are updated synchronously. The update rule is: each energy storage converter independently updates its local charging and discharging power based on minimizing the sum of its own battery cycle life loss sub-objective and penalty term. Then, the dual variables of each energy storage converter are updated based on the average value of the decision variables of all energy storage converters. The difference between the potential function values ​​of two adjacent iterations is determined. When the difference is less than a preset convergence threshold, the iteration is stopped, and the charging and discharging strategies of each energy storage converter when Nash equilibrium is reached are output.

7. The operation optimization control method for industrial and commercial energy storage equipment as described in claim 1, characterized in that, The charging and discharging strategy includes: charging power commands and discharging power commands for each energy storage converter in each control cycle within the future control time domain.

8. An operation optimization control system for commercial and industrial energy storage equipment, used to execute the operation optimization control method for commercial and industrial energy storage equipment as described in any one of claims 1 to 7, characterized in that, The optimized control system includes: The data acquisition module is used to acquire the physical connection relationships and real-time electrical measurement data of various energy storage devices, photovoltaic units, transformers and key load nodes in the industrial and commercial park, and then construct a heterogeneous graph neural network model, embedding the real-time electrical measurement data as the initial features of the nodes into the heterogeneous graph neural network model; The robust optimization module is used to monitor the rate of change of the total incoming power on the low-voltage side of the transformer in real time. When the absolute value of the rate of change exceeds the preset sudden change threshold, it is determined as a load jump event, and the jump amplitude and duration are recorded. Based on the real-time electrical measurement data of each energy storage device, the upper and lower limits of the available charging and discharging power of the energy storage system are determined, and then a robust optimization model containing chance constraints is constructed. The strategy output module is used to perform potential game theory on each energy storage converter in the industrial and commercial park based on the robust optimization model to obtain the charging and discharging strategy when each energy storage converter reaches Nash equilibrium.

Images