Power dispatching method of energy storage system and energy storage system

Abstract

The invention relates to the technical field of energy storage systems, in particular to a power dispatching method of an energy storage system and the energy storage system.The power dispatching method comprises the steps that by deploying edge computing nodes and building a cloud global optimization engine, data support is provided for follow-up precise regulation and control, and timeliness and integrity of data acquisition are guaranteed; energy storage equipment capacity and power grid requirements are analyzed by means of an intelligent algorithm, a reasonable response priority is set, tasks are dynamically allocated, and efficient utilization of resources and accurate execution of the tasks are ensured; an MPC algorithm is adopted to take power grid frequency deviation as an optimization target, output instructions are optimized in a rolling mode, prediction errors are corrected, power grid frequency fluctuation adjustment of the energy storage system is achieved, and power grid stability is enhanced; and the long-time energy storage charging and discharging plan is optimized by taking the full life cycle cost minimization as the target, so that the stability of the power grid and the operation economy of the energy storage equipment are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage systems, and in particular to a power dispatching method and an energy storage system. Background Art

[0002] In the current energy storage system field, power dispatching still faces multiple challenges: Traditional energy storage system data acquisition and control architectures are mostly centralized, resulting in high cloud processing latency and weak edge-side data processing capabilities, leading to delayed device status monitoring; existing fixed-priority or threshold-based dispatching strategies cannot adapt to grid demand and dynamic changes in device status, leading to mismatched charging and discharging of energy storage devices or overuse; single-time-scale optimization leads to response conflicts, and the lack of dynamic optimization mechanisms for multi-type energy storage collaboration results in low resource utilization. To this end, the present invention proposes a power dispatching method and energy storage system for energy storage systems. Summary of the Invention

[0003] The purpose of the present invention is to solve the problems in the background technology and to propose a power dispatching method and an energy storage system for an energy storage system.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions: The power dispatching method of the energy storage system includes: S1. Deploy edge computing nodes, integrate data collection and power execution functions, and complete real-time collection of energy storage device status data. At the same time, build a cloud-based global optimization engine, integrate the energy storage device capability feature library and power price collection unit, and complete energy storage device data analysis and feature extraction. S2. Based on the collected energy storage device status data, an intelligent algorithm is used to analyze the energy storage device capabilities and grid requirements, set the response priority of each energy storage device, and dynamically allocate tasks; S3. Based on the set task allocation and priority, we coordinate the second-level control at the edge layer and the minute-level optimization at the cloud to build a multi-timescale optimization scheduling strategy. Specifically: At the edge computing layer, the MPC algorithm is used to optimize the grid frequency deviation and continuously optimize the output instructions for the next Np seconds. The energy storage system's response to grid frequency fluctuations within the next Np seconds is predicted, and the Kalman filter is used to correct the prediction error and adjust the dynamic response of the energy storage system to grid frequency fluctuations. In the cloud-based global optimization engine, ultra-short-term power forecasts and time-of-use electricity price signals are integrated to optimize long-term energy storage charging and discharging plans with the goal of minimizing lifecycle costs. The second-level MPC control results are combined with the minute-level economic scheduling results to form an optimized scheduling strategy with multiple time scales.

[0005] Furthermore, the process of real-time collection and feature extraction of energy storage device status data includes: Deploy edge computing nodes at the energy storage device end, integrating real-time data acquisition units and power execution units. High-precision sensors are configured in the real-time data acquisition units to obtain comprehensive energy storage device status data. Deploy a bidirectional DC / AC converter in the power execution unit to achieve bidirectional conversion between AC grid power and battery DC power, and collect grid frequency data in real time. The collected frequency data is stored in the local memory of the edge device in the form of a time series. Sensor data is transmitted to edge computing nodes via a low-latency bus. The edge computing nodes pre-process the collected raw data, encapsulate and encrypt the pre-processed data, and upload the encrypted data packets to the cloud-based global optimization engine in real time via 5G slicing networks or industrial optical fibers. Build a cloud-based global optimization engine that integrates the energy storage device capability feature library and the power and electricity price collection unit. The cloud-based global optimization engine receives encrypted data packets uploaded by the edge layer and decrypts them. It extracts key device features from the decrypted data and marks them as energy storage device capability features to build an energy storage device capability feature library. The energy storage device capability feature library is used to store the key features of various energy storage devices. The power and electricity price collection unit is used to collect ultra-short-term power forecast data and time-of-use electricity price signals. A communication protocol based on MQTT over TLS is defined, and the edge computing node and the cloud-based global optimization engine implement data interaction through subscription or publishing mode.

[0006] Furthermore, based on the collected energy storage device status data, an intelligent algorithm is used to analyze the energy storage device capabilities and grid requirements, set the response priority of each energy storage device, and dynamically allocate tasks. The process includes: Acquire real-time demand data and energy storage device status data from the grid dispatch center, and classify grid demand into high-frequency response demand, long-term regulation demand, and economic regulation demand; convert demand types into quantitative feature vectors; Build a multi-dimensional capability feature library for energy storage equipment groups; map the feature vectors of each type of energy storage equipment into a high-dimensional space to form a device capability profile; Based on the established energy storage device capability profile and the grid demand quantified feature vector, the energy storage device capability characteristics are correlated and matched with the grid demand characteristics. The matching degree between the energy storage device capability feature vector and the grid demand feature vector is calculated using cosine similarity. Based on the matching degree, initial priority weights are assigned to different types of energy storage devices. Based on the matching degree and initial priority weights, association rules between energy storage devices and grid demand are constructed. These association rules are used to generate an association matrix between energy storage devices and grid demand. Based on the generated correlation matrix between energy storage devices and grid demand, a reinforcement learning algorithm is used to adapt the dynamic priority allocation of energy storage devices and obtain the dynamic priority weights of each type of energy storage device; A multi-dimensional evaluation system is established by combining a multi-dimensional capability feature library. Based on the dynamic priority weight of the energy storage device, the entropy weight method is used to weight the sum of the scores of each dimension to calculate the dynamic priority score of the energy storage device. The energy storage devices are arranged in descending order according to the priority score to form a candidate response sequence. Through graph neural networks, we conduct in-depth analysis of the current operating status of the power grid and identify the type of power grid fluctuations. The objective function is constructed using the energy storage response power as the optimization variable. Equality constraints are also established. Complementary constraints are set for different grid fluctuation types. The Lagrange multipliers are dynamically adjusted based on the grid fluctuation type identification results and the candidate response sequence of the energy storage equipment. The dynamically adjusted Lagrange multipliers are combined with the complementary constraints to form the final set of constraint equations. After completing the construction of the constraint equation group, the constructed dynamic Lagrange multiplier method model is solved using the interior point method to obtain the output instructions of each energy storage device; the energy storage devices are assigned task types according to the complementary constraints, which are specifically divided into high-frequency response tasks, long-term regulation tasks, and economic regulation tasks; the real-time output instructions of multiple types of energy storage devices after task type assignment are sent to the energy storage device control terminal.

[0007] Furthermore, the process of mapping the feature vectors of each type of energy storage device into a high-dimensional space to form a device capability profile includes: The kernel function is used to transform the inner product operation in the original low-dimensional feature space into the inner product operation in the high-dimensional feature space, thereby realizing the high-dimensional mapping of the energy storage device feature vector. In the energy storage device capability feature mapping, the kernel parameters are adjusted according to different energy storage device types and device operation scenarios. The K-means clustering algorithm is used to perform cluster analysis on the energy storage device capability feature vectors mapped to the high-dimensional space. The energy storage device capability clusters obtained by clustering are associated and matched with the grid demand types.

[0008] Furthermore, the process of constructing association rules between energy storage devices and grid demands and generating an association matrix between energy storage devices and grid demands includes: Set a minimum support threshold and a minimum confidence threshold; the minimum support is used to measure the frequency of the energy storage device-grid demand combination in the entire data set; the minimum confidence reflects the strength of the association between the energy storage device-grid demand combinations; use the Apriori algorithm to mine frequently occurring energy storage device-grid demand combinations; during the mining process, filter out frequent item sets that meet the minimum confidence condition based on the set minimum support threshold; based on the frequent item sets, generate the first association rule based on the minimum confidence threshold; Using the decision tree algorithm, the energy storage device capability characteristics and grid demand characteristics are used as input variables, and the priority weight of the energy storage device is used as the output variable to construct a decision tree model; the second association rule is extracted from the decision tree model: starting from the root node, along the branch path of the decision tree, the splitting characteristics and splitting threshold of each node are combined into the condition part of the rule, and the priority weight adjustment direction and amplitude ΔW of the leaf node relative to the root node or other reference nodes are used as the conclusion part of the rule; based on the constructed second association rule, an association matrix between energy storage devices and grid demand is generated, where the rows of the association matrix represent energy storage devices, the columns represent grid demand types, and the dimension of the matrix is determined according to the type of energy storage devices and the type of grid demand; the first association rule is used as the basis for the analysis of the relationship between energy storage devices and grid demand. Then it is integrated with the second association rule; based on the integrated association rule, the priority weight adjustment value between the energy storage device and the grid demand is calculated and used as an element in the association matrix; for each association rule, if the rule condition is met, the priority weight of the energy storage device is adjusted according to the amplitude ΔW determined in the rule conclusion; the calculated priority weight adjustment value is filled in the corresponding position of the association matrix, and the association matrix is updated; weights are assigned to the initial priority weight and the priority weight adjustment value in the association matrix, and the degree of influence of the two on the final association matrix is comprehensively calculated; the initial priority weight and the priority weight adjustment value in the association matrix are comprehensively calculated using the weighted summation method to obtain the final energy storage device and grid demand association matrix.

[0009] Furthermore, the MPC algorithm is used to optimize the grid frequency deviation and to continuously optimize the output command for the next Np seconds. The energy storage system's response to grid frequency fluctuations within the next Np seconds is predicted, and the prediction error is corrected using Kalman filtering. The process of adjusting the energy storage system's dynamic response to grid frequency fluctuations includes the following: Correlate and analyze the collected real-time grid frequency data, historical grid frequency data stored locally at the edge computing layer, and energy storage system operation data; Establish a dynamic response model of the energy storage system; Determine the prediction time domain Np of the MPC algorithm based on the grid frequency change and the response speed of the energy storage system; set the control time domain Nc of the MPC algorithm, which is less than or equal to the prediction time domain Np; Using the established dynamic response model of the energy storage system and the current grid frequency and energy storage system operating data, we predict the grid frequency changes within Np seconds and the response of the energy storage system to grid frequency fluctuations. Using the predicted grid frequency deviation as the optimization target, the output command of the energy storage system for the next Np seconds is optimized in a rolling manner within the control time domain Nc. The optimization process minimizes the preset measure of the grid frequency deviation within the predicted time domain by searching for possible energy storage system charge and discharge power combinations. The optimized first-moment energy storage system output command is sent to the energy storage system's control terminal, which adjusts the charging and discharging power according to the command, enabling the energy storage system to quickly respond to grid frequency fluctuations. In each control cycle, after the actual grid frequency data and energy storage system operating data are updated, the error between the actual frequency and the frequency predicted by the MPC algorithm, as well as the error between the actual charging and discharging power of the energy storage system and the predicted power, are calculated. Using the Kalman filter algorithm, the state estimate and covariance matrix of the Kalman filter are updated according to the current prediction error and the prior estimate of the Kalman filter; The dynamic response model of the energy storage system is modified based on the updated state estimation of the Kalman filter.

[0010] Furthermore, by integrating ultra-short-term power forecasts with time-of-use electricity price signals and minimizing lifecycle costs, the process of optimizing long-term energy storage charging and discharging plans includes: Aggregate all energy storage device status data uploaded by the edge computing layer to form a global status data set of the energy storage system; Obtain ultra-short-term power forecast datasets and time-of-use electricity price signal datasets; Integrate the acquired ultra-short-term power forecast dataset, time-of-use electricity price signal dataset, and energy storage system global status dataset to form a complete dataset for minute-level optimization. Based on an integrated, minute-level optimized complete data set, a charging and discharging plan for the energy storage system is developed. For charging period selection, the ultra-short-term power forecast data set is combined to identify peak periods for renewable energy generation. For discharging period selection, the time-of-use electricity price signal data set is used to target peak electricity consumption periods. Furthermore, the full lifecycle cost items associated with the energy storage system's charging and discharging plan are identified, including the initial investment cost of the energy storage equipment, operation and maintenance costs, charging and discharging loss costs, battery aging costs, and opportunity costs. Based on the identified cost items, a weighted summation method is used to construct the life cycle cost function; Taking the minimization of the whole life cycle cost as the optimization goal, a genetic algorithm is used to find the optimal energy storage system charging and discharging plan to minimize the whole life cycle cost; The energy storage system charging and discharging plan obtained through minute-level optimization is passed as reference information to the MPC controller at the edge computing layer. When rolling-optimizing the output instructions of the energy storage system, the MPC controller integrates the minute-level economic dispatch results, the real-time status of the current power grid, and the real-time operating data of the energy storage system to form an optimized dispatch strategy at multiple time scales.

[0011] Energy storage system, including: Layered architecture deployment and data acquisition module: Deploy edge computing nodes, integrate data acquisition and power execution functions, and complete real-time collection of energy storage device status data. At the same time, build a cloud-based global optimization engine, integrate the energy storage device capability feature library and power price collection unit, and complete energy storage device data analysis and feature extraction. Dynamic task adaptive allocation module: Based on the collected energy storage device status data, it uses intelligent algorithms to analyze the energy storage device capabilities and grid requirements, sets the response priority of each energy storage device, and dynamically allocates tasks; Multi-timescale optimization scheduling module: Based on the set task allocation and priority, it coordinates the second-level control at the edge layer and the minute-level optimization in the cloud to build a multi-timescale optimization scheduling strategy.

[0012] Compared with existing technologies, the advantages of the power dispatching method and energy storage system provided by the present invention are: The present invention sets the response priority of energy storage equipment and dynamically allocates tasks through intelligent algorithms to ensure the rationality of task allocation; by constructing a multi-time-scale optimization scheduling strategy, the edge layer uses the MPC algorithm for second-level control, quickly responds to frequency fluctuations with the grid frequency deviation as the target, and combines Kalman filtering to correct prediction errors, enhance dynamic response capabilities, and ensure grid frequency stability; through the cloud-based global optimization engine, ultra-short-term power predictions and time-of-use electricity price signals are integrated to optimize long-term energy storage charging and discharging plans at the minute level, reducing the cost of the entire life cycle; by combining second-level and minute-level results, a multi-time-scale optimization scheduling strategy that takes into account both real-time and economic efficiency is formed, thereby improving the operating efficiency of the energy storage system and the power supply quality of the grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of the power dispatching method for the energy storage system proposed in the present invention.

[0014] Figure 2 This is a module diagram of the energy storage system proposed in this invention. DETAILED DESCRIPTION

[0015] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the implementation regulations described are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0016] See also Figure 1 The present invention provides a power dispatching method for an energy storage system, comprising: S1. Deploy edge computing nodes, integrate data collection and power execution functions, and complete real-time collection of energy storage device status data. At the same time, build a cloud-based global optimization engine, integrate the energy storage device capability feature library and power price collection unit, and complete energy storage device data analysis and feature extraction. S2. Based on the collected energy storage device status data, an intelligent algorithm is used to analyze the energy storage device capabilities and grid requirements, set the response priority of each energy storage device, and dynamically allocate tasks; S3. Based on the set task allocation and priority, we coordinate the second-level control at the edge layer and the minute-level optimization at the cloud to build a multi-timescale optimization scheduling strategy. Specifically: At the edge computing layer, the MPC algorithm is used to optimize the grid frequency deviation and continuously optimize the output instructions for the next Np seconds. The energy storage system's response to grid frequency fluctuations within the next Np seconds is predicted, and the Kalman filter is used to correct the prediction error and adjust the dynamic response of the energy storage system to grid frequency fluctuations. In the cloud-based global optimization engine, ultra-short-term power forecasts and time-of-use electricity price signals are integrated to optimize long-term energy storage charging and discharging plans with the goal of minimizing lifecycle costs. The second-level MPC control results are combined with the minute-level economic scheduling results to form an optimized scheduling strategy with multiple time scales.

[0017] It should be further explained that in the specific implementation process, edge computing nodes are deployed, data collection and power execution functions are integrated, and real-time data collection of energy storage device status data is completed. At the same time, a cloud-based global optimization engine is built, integrating the energy storage device capability feature library and power price collection unit to complete the energy storage device data analysis and feature extraction process. Deploy an edge computing node at the energy storage device end, integrating a real-time data acquisition unit and a power execution unit. The real-time data acquisition unit is equipped with high-precision sensors (temperature sensor, SOC / SOH monitoring module, etc.) to monitor key parameters such as the battery's state of charge (SOC), state of health (SOH), real-time power, and temperature in real time, obtaining comprehensive energy storage device status data. A bidirectional DC / AC converter is deployed in the power execution unit to achieve bidirectional conversion (charging or discharging) between AC grid power and battery DC power. A frequency sensor is also integrated to collect grid frequency data in real time. The collected frequency data is stored in the local memory of the edge device in the form of a time series. Specifically, the power execution unit is used to convert AC grid power into DC power for storage in the battery, and then invert the battery DC power into AC power to feed back to the grid or load. Sensor data is transmitted to edge computing nodes via low-latency buses (such as CAN buses). Edge computing nodes perform preprocessing operations on the collected raw data, including filtering, normalization, and outlier detection. The preprocessed data is encapsulated into a standard format and encrypted using a symmetric encryption algorithm. The encrypted data packets are uploaded to the cloud-based global optimization engine in real time via 5G slicing networks or industrial optical fibers. Build a cloud-based global optimization engine, integrating the energy storage equipment capability feature library and the power price acquisition unit; the cloud-based global optimization engine receives the encrypted data packets uploaded by the edge layer, decrypts them with the corresponding key, and restores them to the original data format; performs secondary cleaning on the decrypted data (removes redundant data, fills missing values), extracts key equipment features (such as SOC, SOH, etc.) and marks them as energy storage equipment capability features, and builds an energy storage equipment capability feature library; the energy storage equipment capability feature library is used to store key features of various energy storage devices, such as power-capacity curves and other key data; the power price acquisition unit is used to collect ultra-short-term power forecast data and time-of-use electricity price signals; it is understandable that the ultra-short-term power forecast data is collected in the following ways: The industry's meteorological department or power trading center obtains renewable energy power forecast data, load power forecast data, and overall power demand forecast data for the future period (such as 15 minutes to 1 hour). This data is based on weather forecasts, historical load data, and other information, and is predicted using machine learning methods to provide a basis for formulating reasonable charging and discharging plans for the energy storage system. The time-of-use electricity price signal is collected by connecting with the power market trading platform to obtain the time-of-use electricity price curve, including peak and valley electricity prices and ancillary service compensation prices. The time-of-use electricity price signal reflects the supply and demand relationship and cost changes in the power market. The energy storage system can optimize the charging and discharging plan based on the price signal to achieve economic operation. A communication protocol based on MQTT over TLS is defined, and the edge computing node and the cloud-based global optimization engine realize data interaction through subscription or publishing mode.

[0018] It should be further explained that in the specific implementation process, based on the collected energy storage device status data, an intelligent algorithm is used to analyze the energy storage device capabilities and grid demand, set the response priority of each energy storage device, and dynamically allocate tasks as follows: Acquire real-time demand data (such as frequency fluctuations and power shortage forecasts) and energy storage device status data (SOC, SOH, etc.) from the grid dispatch center, and categorize grid demand into high-frequency response demand (i.e., frequency disturbance scenarios requiring millisecond-level response), long-term regulation demand (i.e., power shortage scenarios requiring continuous support of minutes or more), and economic regulation demand (i.e., peak-valley arbitrage, ancillary service market scenarios, etc.). Convert demand types into quantitative feature vectors, including frequency fluctuation amplitude, power shortage forecast value, and electricity price difference. A multi-dimensional capability feature library is constructed for energy storage device groups. The feature vectors of each type of energy storage device (such as flywheels, lithium batteries, and flow batteries) are mapped to a high-dimensional space to form a device capability profile. The process of mapping the feature vectors of each type of energy storage device to a high-dimensional space to form a device capability profile includes: The inner product operation in the original low-dimensional feature space is converted into the inner product operation in the high-dimensional feature space through the kernel function, thereby realizing the high-dimensional mapping of the energy storage device feature vector; in the energy storage device capability feature mapping, the kernel parameters are adjusted according to different energy storage device types and device operation scenarios; the K-means clustering algorithm is used to perform cluster analysis on the energy storage device capability feature vector mapped to the high-dimensional space; the clustered energy storage device capability cluster is associated with the grid demand type; for example, the high-frequency response demand of the grid usually requires the energy storage device to have a fast power regulation capability, and the high-power energy storage device cluster is associated with the high-frequency response demand; the long-term regulation demand focuses on the energy storage capacity of the energy storage device, and the high-energy energy storage device cluster is associated with the long-term regulation demand; the economic regulation demand focuses on the operating cost and benefits of the energy storage device, and the economic energy storage device cluster is associated with the economic regulation demand; Based on the established energy storage device capability profile and grid demand quantified feature vector, the energy storage device capability characteristics are correlated and matched with the grid demand characteristics. The matching degree between the energy storage device capability feature vector and the grid demand feature vector is calculated using cosine similarity. Initial priority weights are assigned to different types of energy storage devices based on the matching degree. Based on the matching degree and initial priority weights, association rules between energy storage devices and grid demand are constructed. These association rules are used to generate an association matrix between energy storage devices and grid demand, which serves as input for subsequent dynamic priority calculations. The process of constructing association rules between energy storage devices and grid demand and generating the association matrix between energy storage devices and grid demand includes the following steps: A minimum support threshold and a minimum confidence threshold are set. The minimum support is used to measure the frequency of the energy storage device-grid demand combination in the entire data set. The minimum confidence reflects the strength of the association between the energy storage device and grid demand combinations, that is, the probability that a specific energy storage device will be called when a grid demand occurs. The Apriori algorithm is used to mine frequently occurring energy storage device-grid demand combinations. During the mining process, frequent item sets that meet the minimum confidence condition are screened based on the set minimum support threshold. Based on the frequent item sets, the first association rule is generated according to the minimum confidence threshold. Using the decision tree algorithm, the energy storage device capability characteristics and grid demand characteristics are used as input variables, and the priority weight of the energy storage device is used as the output variable to construct a decision tree model. The second association rule is extracted from the decision tree model: starting from the root node, along the branch path of the decision tree, the split characteristics and split threshold of each node are combined into the condition part of the rule, and the priority weight adjustment direction (increase or decrease) and amplitude ΔW of the leaf node relative to the root node or other reference nodes are used as the conclusion part of the rule. Based on the constructed second association rule, an association matrix between energy storage devices and grid demand is generated, where the rows of the association matrix represent energy storage devices, the columns represent the type of grid demand, and the dimension of the matrix is based on the type of energy storage device and the type of grid demand. Type determination; when constructing the association matrix, if no association rules (including the first association rule and the second association rule) are mined between the energy storage device and the grid demand type, the value of the corresponding position in the initial association matrix is 0 or a very small value, indicating that there is no association or a very weak association between the two; the first association rule and the second association rule are integrated: for the same energy storage device-grid demand combination involved in the two rules, the final association rule is determined; based on the integrated association rule, the priority weight adjustment value between the energy storage device and the grid demand is calculated and used as an element in the association matrix: if the association rule indicates that energy storage device i has a high degree of association with grid demand j, the value of the corresponding position (i, j) in the association matrix is large; For each association rule, if the rule condition is met, the priority weight of the energy storage device is adjusted according to the amplitude ΔW determined in the rule conclusion; specifically, the rule condition includes the satisfaction condition for adjusting the priority weight based on the first association rule, the satisfaction condition for adjusting the priority weight based on the second association rule, and the satisfaction condition for adjusting the priority weight based on the combination of the two rules; wherein, the condition part of the first association rule is composed of the specific attributes or states of the energy storage device-grid demand combination. For example, if the grid demand type is load peak demand and the energy storage device type is lithium battery, the priority weight of the energy storage device is increased by ΔW1. Only when the actual grid monitoring data shows that the current grid demand type is load peak demand and the actual The rule condition is met only when the type of energy storage device called is a lithium battery. According to the rule, the priority weight of the lithium battery energy storage device is increased by ΔW1. The second association rule is extracted from the decision tree model. The rule condition corresponds to the split characteristics and split thresholds of each node on the path from the root node to the leaf node in the decision tree. For example, a branch rule of the decision tree is: if the grid demand type is a high-frequency response demand and the power response time of the energy storage device is less than t milliseconds, then the priority weight of the energy storage device is increased by ΔW2. Only when the actual grid demand type is a high-frequency response demand and the power response time of the energy storage device is less than t milliseconds, the rule condition is met. According to the rule, the priority weight of the energy storage device is increased by ΔW2.If the energy storage device-grid demand combination satisfies both the first association rule and the second association rule, the rule priority is pre-set. For example, assuming that the priority of the first association rule is higher than that of the second association rule, when both rules are satisfied at the same time, the priority weight is adjusted according to the first association rule with a higher priority; if the two rules have the same priority, further judgment is made based on factors such as the confidence level of the rule; the calculated priority weight adjustment value is filled into the corresponding position of the association matrix, and the association matrix is updated: for the case where there are multiple association rules, the calculation results of multiple association rules are synthesized using methods such as weighted average or maximum value to obtain the final association matrix element value; weights are assigned to the initial priority weight and the priority weight adjustment value in the association matrix, and the degree of influence of the two on the final association matrix is comprehensively calculated; the initial priority weight and the priority weight adjustment value in the association matrix are synthesized using the weighted summation method to obtain the final energy storage device and grid demand association matrix; Based on the generated association matrix between energy storage devices and grid demand, a reinforcement learning algorithm is used to adapt the dynamic priority allocation of energy storage devices: the energy storage device capability characteristics and grid demand characteristics are used as state space inputs; the priority weights of various types of energy storage devices are defined as the action space; a reward function is constructed: the response speed, regulation accuracy and economic efficiency of the energy storage device are used to reward or punish each action; among which the response speed is the frequency recovery time and the power gap filling time, the regulation accuracy is the frequency deviation suppression rate and the power gap filling error, and the economic efficiency includes the charging and discharging costs and the ancillary service income; the current energy storage device status and grid demand are input into the trained reinforcement learning model, and the dynamic priority weights of various types of energy storage devices are output; A multi-dimensional evaluation system is established by combining a multi-dimensional capability feature library, including response speed (milliseconds / seconds), energy density (kWh), cycle life (number of charge and discharge cycles), and SOC state (charge percentage). Based on the dynamic priority weight of the energy storage device, the entropy weight method is used to weight the sum of the scores of each dimension to calculate the dynamic priority score of the energy storage device. Specifically, the dynamic priority weight reflects the importance of different types of energy storage equipment under the current grid demand, and the entropy weight method determines the objective weight of each dimension in the comprehensive score based on the degree of dispersion of the scores of each dimension. The objective weight is then adjusted in combination with the dynamic priority weight to obtain the final comprehensive score. The energy storage devices are arranged in descending order according to the priority score to form a candidate response sequence. Graph neural networks are used to deeply analyze the current operating status of the power grid and identify grid fluctuation types. The grid is abstracted into a dynamic graph structure with power generation nodes as vertices and transmission lines as edges, constructing a grid topology relationship network. Real-time grid operation data (such as node frequency deviation) is embedded in the graph vertices, and edge attributes are associated with physical parameters such as line impedance. Graph convolution is used to aggregate features of the graph structure, automatically extracting topological dependency features (such as local frequency propagation paths and power shortage diffusion patterns). The resulting grid fluctuation type classification results are output (i.e., distinguishing between frequency disturbances and power shortages. For frequency disturbances, node clusters with abnormal frequency change rates are identified and marked as high-frequency response requirements; for power shortages, areas with insufficient power injection are identified and marked as long-term regulation requirements). Furthermore, grid fluctuation type labels are generated for subsequent constraint generation. Using energy storage response power as the optimization variable, an objective function (such as minimizing frequency deviation / power shortfall) is constructed. This objective function comprehensively considers multiple factors, including grid stability, energy storage equipment lifespan, and operating costs, to maximize overall benefits. At the same time, equality constraints are established (including global grid constraints: power generation + energy storage output = load demand, and energy storage constraints: SOC charge and discharge upper and lower limits, and maximum charge and discharge power) to ensure that the energy storage equipment meets basic grid power balance requirements during operation. Complementary constraints are set for different types of grid fluctuations: For frequency disturbances, based on the characteristics of grid frequency disturbances, a complementary constraint is set that the energy storage response is negatively correlated with the frequency change rate: when the grid frequency drops, the energy storage device is required to discharge and inject power into the grid to support the frequency recovery; when the grid frequency rises, the energy storage device is charged, absorbing excess power from the grid and suppressing further frequency increases, thereby achieving rapid response and regulation to frequency disturbances; for power shortages, a complementary constraint is set that the energy storage response is positively correlated with the power shortage amount, that is, the greater the grid power shortage, the higher the power output of the energy storage device, and by timely filling the grid power gap, the stable operation of the grid is guaranteed; Based on the grid fluctuation type identification results and the candidate response sequence of energy storage devices, the Lagrange multiplier is dynamically adjusted (in the case of high-frequency disturbances, the multiplier value of flywheel energy storage is increased (prioritizing frequency response); in the case of power shortages, the multiplier value of battery energy storage is increased (prioritizing power replenishment)), strengthening the leading role of high-priority devices in the complementary constraints. The dynamically adjusted Lagrange multiplier is combined with the above complementary constraints to form the final set of constraint equations, ensuring that the energy storage response is complementary to the grid fluctuation characteristics. After completing the construction of the constraint equation group, the constructed dynamic Lagrange multiplier method model is solved using the interior point method to obtain the output instructions of each energy storage device. The energy storage devices are assigned task types according to the complementary constraints, specifically divided into high-frequency response tasks, long-term regulation tasks, and economic regulation tasks (for example, energy storage devices with high priority weights (such as flywheels) are assigned high-frequency response tasks, and the output instructions are adjusted in real time according to frequency fluctuations (for example, updated once per second); energy storage devices with low priority weights (such as lithium batteries) are assigned long-term regulation tasks, and the output instructions are smoothly adjusted according to power shortage predictions (for example, updated once per minute); for economic regulation tasks, they are assigned to low-cost energy storage according to the electricity price difference). The real-time output instructions of the multi-type energy storage devices after task type assignment are sent to the energy storage device control terminal; It is understandable that the interior point method is an optimization algorithm that iteratively approximates the optimal solution within the feasible domain. The constraint conditions are integrated into the objective function by introducing a barrier function. This embodiment is used to solve the nonlinear programming problem of the dynamic Lagrange multiplier model of the energy storage system.

[0019] It should be further explained that in the specific implementation process, based on the set task allocation and priority, the process of coordinating the second-level control at the edge layer and the minute-level optimization at the cloud side to build a multi-timescale optimization scheduling strategy is as follows: The collected real-time grid frequency data, historical grid frequency data stored locally at the edge computing layer, and energy storage system operation data are correlated and analyzed to uncover potential patterns between grid frequency changes and the charging and discharging of the energy storage system. Energy storage system operation data is collected in real time through edge computing nodes, including process parameters (such as charge and discharge power) and system logs when the energy storage device executes instructions. This data is then integrated with long-term performance indicators (such as cycle life degradation) stored in a cloud-based time-series historical database and archived dispatch instruction data. Establish a dynamic response model for the energy storage system. This model reflects the dynamic relationship between the energy storage system's charge and discharge processes, power output, and grid frequency. It uses state-space equations to describe the model, taking into account characteristics such as the energy storage device's charge and discharge efficiency and self-discharge rate. Determine the prediction time domain N of the MPC algorithm based on the grid frequency change and the response speed of the energy storage system; set the control time domain of the MPC algorithm, which is less than or equal to the prediction time domain; Using the established energy storage system dynamic response model and the current grid frequency and energy storage system operating data, we predict the grid frequency changes within the next N seconds and the energy storage system's response to grid frequency fluctuations. Using the predicted grid frequency deviation as the optimization target, the output command of the energy storage system is optimized over the next N seconds within the control time domain. The optimization process searches for possible energy storage system charge and discharge power combinations to minimize a preset measure (such as the sum of squares) of the grid frequency deviation within the predicted time domain. The optimized first-moment energy storage system output command is sent to the energy storage system's control terminal, which adjusts the charging and discharging power according to the command, enabling the energy storage system to quickly respond to grid frequency fluctuations. In each control cycle, after the actual grid frequency data and energy storage system operating data are updated, the error between the actual frequency and the frequency predicted by the MPC algorithm, as well as the error between the actual charging and discharging power of the energy storage system and the predicted power, are calculated. Using the Kalman filter algorithm, the state estimate and covariance matrix of the Kalman filter are updated based on the current prediction error and the Kalman filter's prior estimate. The Kalman filter algorithm continuously uses new measurement data to correct its estimate of the energy storage system state, thereby improving the accuracy of predictions of grid frequency changes and the energy storage system's operating status. Based on the updated state estimate of the Kalman filter, the dynamic response model of the energy storage system is corrected. The correction method is to adjust the parameters in the dynamic response model of the energy storage system, such as the charging and discharging efficiency parameters and response time parameters of the energy storage device. Aggregate all energy storage device status data uploaded by the edge computing layer to form a global status data set of the energy storage system; Obtain ultra-short-term power forecast datasets and time-of-use electricity price signal datasets; Integrate the acquired ultra-short-term power forecast dataset, time-of-use electricity price signal dataset, and energy storage system global status dataset to form a complete dataset for minute-level optimization. Based on an integrated, minute-level optimized complete data set, a charging and discharging plan for the energy storage system is developed. For charging period selection, the ultra-short-term power forecast data set is combined to identify peak periods for renewable energy generation (such as wind and photovoltaic power). For discharging period selection, the time-of-use electricity price signal data set is used to target peak electricity consumption periods (such as evening residential peak electricity consumption and periods of concentrated industrial production). Furthermore, the full lifecycle cost items associated with the energy storage system's charging and discharging plan are identified, including the initial investment cost of the energy storage equipment, operation and maintenance costs, charging and discharging loss costs, battery aging costs, and opportunity costs. Based on the identified cost items, a weighted summation approach is used to construct a full lifecycle cost function, where the weights are adjusted according to the importance of different cost items and actual operating conditions. For example, at the beginning of the energy storage system lifecycle, the initial investment cost has a higher weight. As the equipment's operating time increases, the weights of operation and maintenance costs and battery aging costs gradually increase. With the minimization of the whole life cycle cost as the optimization goal, the genetic algorithm is used to find the optimal energy storage system charging and discharging plan to minimize the whole life cycle cost; the process of finding the optimal energy storage system charging and discharging plan using the genetic algorithm is as follows: J1. The charging and discharging power of the energy storage system in each optimization period (such as minute level) is used as the core decision variable: Assuming that the optimization cycle is 24 hours, each hour is divided into 60 periods, and each energy storage device has three states in each period: charging, discharging or inaction, which are discretized into different power gears (such as charging power -100kW, -50kW, 0kW, discharging power 50kW). , 100kW, etc.); J2. Determine the time periods for charging or discharging of the energy storage device, represented by binary coding: real number coding is used for charging and discharging power (continuous variables, such as -100kW to 100kW with an accuracy of 0.1kW); integer coding is used for charging and discharging time period allocation (discrete variables, such as 0=no action, 1=charging, 2=discharging), and a vector with the same number of optimized time periods is constructed to describe the charging and discharging time period allocation scheme of the energy storage device; J3. Within the range of charging and discharging power, randomly generate the initial population according to a uniform distribution. For example, for real number coding, if the charging and discharging power range is -100kW To 100kW, randomly generate a real number between -100 and 100 as the charge and discharge power for each optimization period. Repeat this process to generate a sufficient number of individuals (e.g., 100) to form the initial population. Based on constraints such as the energy storage system's charge and discharge power limit and battery capacity limit, randomly generate the initial population within the range that satisfies the constraints. For example, based on the energy storage device's maximum charge and discharge power and its current state of charge, determine the optional range of charge and discharge power for each period, and then randomly generate initial individuals within this range. J4. Calculate the various costs of each individual based on the identified lifecycle cost items, including the energy storage device's initial investment cost, operation and maintenance costs, charge and discharge loss costs, battery aging costs, and opportunity costs. For example, the initial investment cost is calculated directly based on the energy storage device's capacity and purchase price. The operation and maintenance cost is calculated based on the charge and discharge power and power variation (the greater the power and the more frequent the variation, the higher the operation and maintenance cost). The charge and discharge loss cost is calculated based on the charge and discharge power and energy conversion efficiency. The battery aging cost is calculated based on the number of battery cycles and the depth of charge and discharge (e.g., one deep discharge counts as one cycle, and a shallow charge and discharge counts as 0.5 times counts as 1 cycle), combined with the battery replacement cost amortization; opportunity cost: compare the actual discharge income with the theoretical income difference during the peak electricity price period; J5. Weighted sum of various costs according to the preset weights to obtain the full life cycle cost of each individual; the inverse of the full life cycle cost is used as the fitness value; linear transformation is performed on the full life cycle cost to map it to a suitable fitness value range; J6. The probability of being selected is calculated based on the individual's fitness value; for real number coding, a crossover point is randomly selected, and the genes of the two parent individuals after the crossover point are exchanged to generate two offspring individuals. For example, for two parent individuals represented by 24-dimensional real number vectors, a time period (such as the 12th time period) is randomly selected as the crossover point, and the charge and discharge power values of the two parent individuals after the 12th time period are exchanged to obtain two offspring individuals; for integer coding, a single-point crossover method is used, but it is necessary to pay attention to whether the individuals after the crossover meet the constraints; for real number coding, each gene is mutated with a certain mutation probability Pm Within the gene's value range, a new value is randomly generated using a uniform distribution to replace the original gene value. For example, if the original value of a gene is 50kW, after mutation, a new value between -100kW and 100kW may be randomly generated. For integer encoding, uniform mutation is used to randomly select a new gear from the discretized power levels to replace the original one. J7. For individuals that do not meet the constraints, a penalty term is added to their fitness value. For example, if the charge and discharge power of the energy storage device exceeds its maximum charge and discharge power limit, or the battery power level does not reach the expected range at the end of the optimization cycle, a corresponding penalty term is designed based on the degree of constraint violation. J8. A maximum number of iterations is set. When the algorithm reaches the preset maximum number of iterations, the algorithm terminates. For example, the maximum number of iterations can be set to 500. When the algorithm terminates, the optimal individual in the current population is output, that is, the optimal charge and discharge plan for the energy storage system. The charge and discharge plan includes the charge and discharge power values for each optimization period and the allocation of charge and discharge periods. The energy storage system charging and discharging plan obtained through minute-level optimization is passed as reference information to the MPC controller at the edge computing layer. When rolling-optimizing the output instructions of the energy storage system, the MPC controller integrates the minute-level economic dispatch results, the real-time status of the current power grid, and the real-time operating data of the energy storage system to form an optimized dispatch strategy at multiple time scales.

[0020] See also Figure 2 The present invention provides an energy storage system, comprising: Layered architecture deployment and data acquisition module: Deploy edge computing nodes, integrate data acquisition and power execution functions, and complete real-time collection of energy storage device status data. At the same time, build a cloud-based global optimization engine, integrate the energy storage device capability feature library and power price collection unit, and complete energy storage device data analysis and feature extraction. Dynamic task adaptive allocation module: Based on the collected energy storage device status data, it uses intelligent algorithms to analyze the energy storage device capabilities and grid requirements, sets the response priority of each energy storage device, and dynamically allocates tasks; Multi-timescale optimization scheduling module: Based on the set task allocation and priority, it coordinates the second-level control at the edge layer and the minute-level optimization in the cloud to build a multi-timescale optimization scheduling strategy.

[0021] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0022] It should be understood that determining B based on A does not mean determining B based solely on A. B can also be determined based on A and / or other information.

[0023] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0024] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A power dispatching method for an energy storage system, characterized by: S1. Deploy edge computing nodes, integrate data collection and power execution functions, and complete real-time collection of energy storage device status data. At the same time, build a cloud-based global optimization engine, integrate the energy storage device capability feature library and power price collection unit, and complete energy storage device data analysis and feature extraction. S2. Based on the collected energy storage device status data, an intelligent algorithm is used to analyze the energy storage device capabilities and grid requirements, set the response priority of each energy storage device, and dynamically allocate tasks; S3. Based on the set task allocation and priority, we coordinate the second-level control at the edge layer and the minute-level optimization at the cloud to build a multi-timescale optimization scheduling strategy. Specifically: At the edge computing layer, the MPC algorithm is used to optimize the grid frequency deviation and continuously optimize the output instructions for the next Np seconds. The energy storage system's response to grid frequency fluctuations within the next Np seconds is predicted, and the Kalman filter is used to correct the prediction error and adjust the dynamic response of the energy storage system to grid frequency fluctuations. In the cloud-based global optimization engine, ultra-short-term power forecasts and time-of-use electricity price signals are integrated to optimize long-term energy storage charging and discharging plans with the goal of minimizing lifecycle costs. The second-level MPC control results are combined with the minute-level economic scheduling results to form an optimized scheduling strategy with multiple time scales.

2. The power dispatching method of the energy storage system according to claim 1, characterized in that: The process of real-time collection and feature extraction of energy storage device status data includes: Deploy edge computing nodes at the energy storage device end, integrating real-time data acquisition units and power execution units. High-precision sensors are configured in the real-time data acquisition units to obtain comprehensive energy storage device status data. Deploy a bidirectional DC / AC converter in the power execution unit to achieve bidirectional conversion between AC grid power and battery DC power, and collect grid frequency data in real time. The collected frequency data is stored in the local memory of the edge device in the form of a time series. Sensor data is transmitted to edge computing nodes via a low-latency bus. The edge computing nodes pre-process the collected raw data, encapsulate and encrypt the pre-processed data, and upload the encrypted data packets to the cloud-based global optimization engine in real time via 5G slicing networks or industrial optical fibers. Build a cloud-based global optimization engine that integrates the energy storage device capability feature library and the power and electricity price collection unit. The cloud-based global optimization engine receives encrypted data packets uploaded by the edge layer and decrypts them. It extracts key device features from the decrypted data and marks them as energy storage device capability features to build an energy storage device capability feature library. The energy storage device capability feature library is used to store the key features of various energy storage devices. The power and electricity price collection unit is used to collect ultra-short-term power forecast data and time-of-use electricity price signals. A communication protocol based on MQTT over TLS is defined, and the edge computing node and the cloud-based global optimization engine implement data interaction through subscription or publishing mode.

3. The power dispatching method of the energy storage system according to claim 2, characterized in that: Based on the collected energy storage device status data, an intelligent algorithm is used to analyze the energy storage device capabilities and grid requirements, set the response priority of each energy storage device, and dynamically allocate tasks. The process includes: Acquire real-time demand data and energy storage device status data from the grid dispatch center, and classify grid demand into high-frequency response demand, long-term regulation demand, and economic regulation demand; convert demand types into quantitative feature vectors; Build a multi-dimensional capability feature library for energy storage equipment groups; map the feature vectors of each type of energy storage equipment into a high-dimensional space to form a device capability profile; Based on the established energy storage device capability profile and the grid demand quantified feature vector, the energy storage device capability characteristics are correlated and matched with the grid demand characteristics. The matching degree between the energy storage device capability feature vector and the grid demand feature vector is calculated using cosine similarity. Based on the matching degree, initial priority weights are assigned to different types of energy storage devices. Based on the matching degree and initial priority weights, association rules between energy storage devices and grid demand are constructed. These association rules are used to generate an association matrix between energy storage devices and grid demand. Based on the generated correlation matrix between energy storage devices and grid demand, a reinforcement learning algorithm is used to adapt the dynamic priority allocation of energy storage devices and obtain the dynamic priority weights of each type of energy storage device; A multi-dimensional evaluation system is established by combining a multi-dimensional capability feature library. Based on the dynamic priority weight of the energy storage device, the entropy weight method is used to weight the sum of the scores of each dimension to calculate the dynamic priority score of the energy storage device. The energy storage devices are arranged in descending order according to the priority score to form a candidate response sequence. Through graph neural networks, we conduct in-depth analysis of the current operating status of the power grid and identify the type of power grid fluctuations; The objective function is constructed using the energy storage response power as the optimization variable. Equality constraints are also established. Complementary constraints are set for different grid fluctuation types. The Lagrange multipliers are dynamically adjusted based on the grid fluctuation type identification results and the candidate response sequence of the energy storage equipment. The dynamically adjusted Lagrange multipliers are combined with the complementary constraints to form the final set of constraint equations. After completing the construction of the constraint equation group, the constructed dynamic Lagrange multiplier method model is solved using the interior point method to obtain the output instructions of each energy storage device; the energy storage devices are assigned task types according to the complementary constraints, which are specifically divided into high-frequency response tasks, long-term regulation tasks, and economic regulation tasks; the real-time output instructions of multiple types of energy storage devices after task type assignment are sent to the energy storage device control terminal.

4. The power dispatching method of the energy storage system according to claim 3, characterized in that: The process of mapping the feature vector of each type of energy storage device into a high-dimensional space to form a device capability profile includes: The kernel function is used to transform the inner product operation in the original low-dimensional feature space into the inner product operation in the high-dimensional feature space, thereby realizing the high-dimensional mapping of the energy storage device feature vector. In the energy storage device capability feature mapping, the kernel parameters are adjusted according to different energy storage device types and device operation scenarios. The K-means clustering algorithm is used to perform cluster analysis on the energy storage device capability feature vectors mapped to the high-dimensional space. The energy storage device capability clusters obtained by clustering are associated and matched with the grid demand types.

5. The power dispatching method of the energy storage system according to claim 3, characterized in that: The process of establishing association rules between energy storage devices and grid demand and generating an association matrix between energy storage devices and grid demand includes: Set a minimum support threshold and a minimum confidence threshold; the minimum support is used to measure the frequency of the energy storage device-grid demand combination in the entire data set; the minimum confidence reflects the strength of the association between the energy storage device-grid demand combinations; use the Apriori algorithm to mine frequently occurring energy storage device-grid demand combinations; during the mining process, filter out frequent item sets that meet the minimum confidence condition based on the set minimum support threshold; based on the frequent item sets, generate the first association rule based on the minimum confidence threshold; Using the decision tree algorithm, the energy storage device capability characteristics and grid demand characteristics are used as input variables, and the priority weight of the energy storage device is used as the output variable to construct a decision tree model; the second association rule is extracted from the decision tree model: starting from the root node, along the branch path of the decision tree, the splitting characteristics and splitting threshold of each node are combined into the condition part of the rule, and the priority weight adjustment direction and amplitude ΔW of the leaf node relative to the root node or other reference nodes are used as the conclusion part of the rule; based on the constructed second association rule, an association matrix between energy storage devices and grid demand is generated, where the rows of the association matrix represent energy storage devices, the columns represent grid demand types, and the dimension of the matrix is determined according to the type of energy storage devices and the type of grid demand; the first association rule is used as the basis for the analysis of the relationship between energy storage devices and grid demand. Then it is integrated with the second association rule; based on the integrated association rule, the priority weight adjustment value between the energy storage device and the grid demand is calculated and used as an element in the association matrix; for each association rule, if the rule condition is met, the priority weight of the energy storage device is adjusted according to the amplitude ΔW determined in the rule conclusion; the calculated priority weight adjustment value is filled in the corresponding position of the association matrix, and the association matrix is updated; weights are assigned to the initial priority weight and the priority weight adjustment value in the association matrix, and the degree of influence of the two on the final association matrix is comprehensively calculated; the initial priority weight and the priority weight adjustment value in the association matrix are comprehensively calculated using the weighted summation method to obtain the final energy storage device and grid demand association matrix.

6. The power dispatching method of the energy storage system according to claim 1, characterized in that: The MPC algorithm is used to optimize the grid frequency deviation and to continuously optimize the output command for the next Np seconds. The energy storage system's response to grid frequency fluctuations within the next Np seconds is predicted, and the Kalman filter is used to correct the prediction error. The process of adjusting the energy storage system's dynamic response to grid frequency fluctuations includes the following: Correlate and analyze the collected real-time grid frequency data, historical grid frequency data stored locally at the edge computing layer, and energy storage system operation data; Establish a dynamic response model of the energy storage system; Determine the prediction time domain Np of the MPC algorithm based on the grid frequency change and the response speed of the energy storage system; set the control time domain Nc of the MPC algorithm, which is less than or equal to the prediction time domain Np; Using the established dynamic response model of the energy storage system and the current grid frequency and energy storage system operating data, we predict the grid frequency changes within Np seconds and the response of the energy storage system to grid frequency fluctuations. Using the predicted grid frequency deviation as the optimization target, the output command of the energy storage system for the next Np seconds is optimized in a rolling manner within the control time domain Nc. The optimization process minimizes the preset measure of the grid frequency deviation within the predicted time domain by searching for possible energy storage system charge and discharge power combinations. The optimized first-moment energy storage system output command is sent to the energy storage system's control terminal, which adjusts the charging and discharging power according to the command, enabling the energy storage system to quickly respond to grid frequency fluctuations. In each control cycle, after the actual grid frequency data and energy storage system operating data are updated, the error between the actual frequency and the frequency predicted by the MPC algorithm, as well as the error between the actual charging and discharging power of the energy storage system and the predicted power, are calculated. Using the Kalman filter algorithm, the state estimate and covariance matrix of the Kalman filter are updated according to the current prediction error and the prior estimate of the Kalman filter; The dynamic response model of the energy storage system is modified based on the updated state estimation of the Kalman filter.

7. The power dispatching method of the energy storage system according to claim 1, characterized in that: By integrating ultra-short-term power forecasts with time-of-use electricity price signals and minimizing lifecycle costs, the process of optimizing long-term energy storage charging and discharging plans includes: Aggregate all energy storage device status data uploaded by the edge computing layer to form a global status data set of the energy storage system; Obtain ultra-short-term power forecast datasets and time-of-use electricity price signal datasets; Integrate the acquired ultra-short-term power forecast dataset, time-of-use electricity price signal dataset, and energy storage system global status dataset to form a complete dataset for minute-level optimization. Based on an integrated, minute-level optimized complete data set, a charging and discharging plan for the energy storage system is developed. For charging period selection, the ultra-short-term power forecast data set is combined to identify peak periods for renewable energy generation. For discharging period selection, the time-of-use electricity price signal data set is used to target peak electricity consumption periods. Furthermore, the full lifecycle cost items associated with the energy storage system's charging and discharging plan are identified, including the initial investment cost of the energy storage equipment, operation and maintenance costs, charging and discharging loss costs, battery aging costs, and opportunity costs. Based on the identified cost items, a weighted summation method is used to construct the life cycle cost function; Taking the minimization of the whole life cycle cost as the optimization goal, a genetic algorithm is used to find the optimal energy storage system charging and discharging plan to minimize the whole life cycle cost; The energy storage system charging and discharging plan obtained through minute-level optimization is passed as reference information to the MPC controller at the edge computing layer. When rolling-optimizing the output instructions of the energy storage system, the MPC controller integrates minute-level economic dispatch results, the real-time status of the current power grid, and the real-time operating data of the energy storage system to form an optimized dispatch strategy at multiple time scales.

8. Energy storage system, characterized in that, A power dispatching method for an energy storage system according to any one of claims 1 to 7, the system comprising: Layered architecture deployment and data acquisition module: Deploy edge computing nodes, integrate data acquisition and power execution functions, and complete real-time collection of energy storage device status data. At the same time, build a cloud-based global optimization engine, integrate the energy storage device capability feature library and power price collection unit, and complete energy storage device data analysis and feature extraction. Dynamic task adaptive allocation module: Based on the collected energy storage device status data, it uses intelligent algorithms to analyze the energy storage device capabilities and grid requirements, sets the response priority of each energy storage device, and dynamically allocates tasks; Multi-timescale optimization scheduling module: Based on the set task allocation and priority, it coordinates the second-level control at the edge layer and the minute-level optimization in the cloud to build a multi-timescale optimization scheduling strategy.

Images