Energy storage power station energy storage coordination control method and system

Abstract

This invention discloses a coordinated control method and system for energy storage power stations, belonging to the field of automation control technology. The specific steps of the control method are as follows: Ⅰ. During the operation of the energy storage power station, the status data of each energy storage unit in the energy storage power station is collected and preprocessed in real time through edge nodes and uploaded to the central coordinator; Ⅱ. The central coordinator dynamically constructs a virtual energy storage cluster based on the current task requirements and the status of the energy storage units, and performs hybrid control of the energy storage power station; Ⅲ. Based on the real-time preprocessed status data, the dominant operating scenario is automatically determined, and the weights of each energy storage unit in the energy storage power station are dynamically updated. This invention can accurately assess the adaptability of each energy storage unit to the current grid task, ensuring timely task response, sufficient power, and adequate electricity. It can effectively cope with complex grid demands and resource fluctuations, improve prediction accuracy and foresight, achieve seamless connection from macro strategy to micro control, and improve response accuracy and control granularity.

Description

Technical Field

[0001] This invention relates to the field of automation control technology, and in particular to a method and system for coordinated control of energy storage in an energy storage power station. Background Technology

[0002] With the large-scale integration of renewable energy sources such as wind and solar power into the power system, the volatility and uncertainty of the power grid are becoming increasingly prominent, particularly in terms of widening peak-to-valley differences, increased load regulation pressure, and declining power quality. To improve the flexibility and stability of the power grid, energy storage systems, as an important means of regulating power, balancing loads, and improving the utilization rate of new energy sources, are receiving widespread attention. However, single energy storage systems often suffer from slow response times, inflexible dispatching, and low resource utilization efficiency during application. Especially in large-scale energy storage power plants, how to coordinate the operating status of multiple energy storage devices and achieve a unified, efficient, and dynamically responsive control strategy has become a pressing technical challenge.

[0003] A search revealed that Chinese patent CN117060597A discloses a method and system for coordinated control of energy storage in an energy storage power station. While this invention achieves the technical effect of improving the reliability and accuracy of coordinated control of energy storage, it cannot accurately assess the adaptability of each energy storage unit to the current power grid task, cannot effectively cope with complex power grid demand and resource fluctuations, reduces prediction accuracy and foresight, and has low response precision and control granularity. Therefore, we propose a method and system for coordinated control of energy storage in an energy storage power station. Summary of the Invention

[0004] The purpose of this invention is to address the deficiencies in the existing technology by proposing a coordinated control method and system for energy storage power stations.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A coordinated control method for energy storage in an energy storage power station, the specific steps of which are as follows:

[0007] Ⅰ. During the operation of the energy storage power station, the status data of each energy storage unit of the energy storage power station is collected and preprocessed in real time through edge nodes and uploaded to the central coordinator;

[0008] II. The central coordinator dynamically constructs a virtual energy storage cluster based on the current task requirements and the status of the energy storage units, and performs hybrid control of the energy storage power station;

[0009] Ⅲ. Based on the real-time preprocessed status data, automatically determine the current dominant operating scenario and dynamically update the weights of each energy storage unit in the energy storage power station;

[0010] IV. To make short-term predictions on fluctuations in new energy output, load changes, and the SOC evolution and health trends of energy storage units;

[0011] V. Based on the forecast results, continuously optimize and control the energy storage power station, assess the health status of each energy storage unit in real time, and dynamically adjust the charging and discharging strategies.

[0012] As a further aspect of the present invention, the specific steps of step I, which involve real-time acquisition and preprocessing of the status data of each energy storage unit in the energy storage power station via edge nodes, are as follows:

[0013] S1.1: Extract the status data of each energy storage unit within the preset sliding time window in real time, align the status data of different sampling frequencies and sources in time, sort the status data of each energy storage unit in descending order of time, and select the status data at the 25% and 75% positions of each sorted status data as the first quartile and third quartile of the corresponding energy storage unit status data.

[0014] S1.2: The distance between the first quartile and the third quartile is taken as the interquartile range. Then, based on the first quartile, the third quartile and the interquartile range, the normal threshold range of the state data of each energy storage unit is determined. Then, the state data within the normal threshold range is overrun, and the state data exceeding the upper and lower limits of the normal threshold range is removed.

[0015] S1.3: Linear interpolation is used to fill in missing values ​​in the state data of each energy storage unit extracted at different times. After filling, the noise information of the state data of each energy storage unit is smoothed by the exponential weighted moving average method. Then, min-max normalization is performed on each smoothed energy storage unit.

[0016] As a further aspect of the present invention, the specific steps for dynamically constructing the virtual energy storage cluster in step II are as follows:

[0017] S2.1: The central coordinator collects the current power grid and market task requirements, including the total required power capacity, the required duration, the maximum allowable power fluctuation rate, and the required response time, and integrates the collected demand information into the total task requirements;

[0018] S2.2: The edge nodes extract the maximum available power, available energy, efficiency coefficient, response time, and health score of each energy storage unit from the preprocessed status data of each energy storage unit, and establish real-time status indicators for each energy storage unit based on the extracted parameters.

[0019] S2.3: Based on the total current task requirements, calculate the comprehensive matching score between each energy storage unit and the total current task requirements, sort them from high to low according to the comprehensive matching score, filter out each energy storage unit with a score lower than the preset matching score, and construct the corresponding candidate energy storage subset based on the remaining energy storage units.

[0020] S2.4: Verify whether the candidate energy storage subset can meet the total power and response indicators required by the task. If the sum of the current maximum available power of each energy storage unit in the candidate energy storage subset is lower than the required total power capacity, or the sum of the current available energy is lower than the sum of the total power capacity within the required duration, then it is determined that the candidate energy storage subset does not meet the total power and response indicators required by the task.

[0021] S2.5: If the candidate energy storage subset does not meet the total power and response indicators required by the task, the matching score is reset and the energy storage units are re-selected to construct a candidate energy storage subset until the candidate energy storage subset meets the total power and response indicators required by the task. Then, the energy storage units in the candidate energy storage subset are logically aggregated into a virtual cluster, and the unified attributes of the virtual cluster are recorded, including cluster members, the sum of the current maximum available power of each energy storage unit and the sum of the current available energy, average response time and average health factor.

[0022] S2.6: The central coordinator updates the total task demand in real time based on the current grid and market task requirements. Edge nodes upload the current status indicators of each energy storage unit in real time. When the total task demand changes or the candidate energy storage subset does not meet the total power and response indicators required by the task, the update mechanism is triggered, and the comprehensive matching score between each energy storage unit and the current total task demand is recalculated to update the candidate energy storage subset.

[0023] As a further aspect of the present invention, the specific calculation formula for the comprehensive matching score in S2.3 is as follows:

[0024]

[0025] In the formula, Representing the The overall matching score of each energy storage unit; Representing the The current maximum available power of each energy storage unit; This represents the total required power capacity; Representing the The current available energy of each energy storage unit; This represents the required duration; Representing the Response time of each energy storage unit; This represents the required response time; Representing the Health score of each energy storage unit; , , as well as These represent the weighting coefficients for the capability dimension.

[0026] As a further aspect of the present invention, the specific steps of the short-term prediction in step IV are as follows:

[0027] S3.1: Edge nodes preprocess the historical state data of each energy storage unit in the current virtual cluster, and extract the wind power output, photovoltaic irradiance, temperature, humidity, station load, cluster SOC weighted average, maximum SOC deviation and disturbance factor of each new energy and energy storage state evolution characteristics from the preprocessed historical state data, so as to construct the corresponding training sample set and validation sample set.

[0028] S3.2: Decompose the prediction task of energy storage power station into multiple sub-objectives, including new energy power prediction, short-term load change prediction, energy storage unit SOC evolution trajectory prediction, health index trend prediction, and local frequency disturbance sensitivity index, and select the corresponding learning model according to each sub-objective.

[0029] S3.3: Based on the training sample set, the learning model corresponding to each sub-target is trained independently and the performance index is recorded. Then, the error value of each learning model on the validation sample set is calculated. Then, the fusion weight is reversibly allocated according to the error of each learning model on the validation sample. When the error value of each learning model converges to the preset range during the training and validation process, the training is stopped.

[0030] S3.4: Input the state data of each energy storage unit in the energy storage power station into each learning model. Then, each learning model outputs the prediction results of the corresponding sub-objective. Based on the allocated fusion weights of each sub-objective, the prediction results of each sub-objective are weighted and fused to generate the final overall target prediction result. Collect the overall target prediction results for continuous time periods and draw the corresponding state trend map of the energy storage power station.

[0031] An energy storage power station energy storage coordination and control system includes an acquisition and processing module, a state perception module, a scene recognition module, a cluster construction module, a weight adjustment module, a state prediction module, a rolling optimization module, a health assessment module, a decomposition execution module, and a communication coordination module.

[0032] The data acquisition and processing module is used to collect various types of data from the operation of the energy storage power station, and to filter, normalize, detect anomalies, and extract features from the collected data.

[0033] The status perception module assesses the overall operating status of the energy storage power station in real time based on the pre-processed data.

[0034] The scene recognition module is used to identify the current operating scene based on the current operating status of the energy storage power station and historical data;

[0035] The cluster construction module dynamically combines energy storage units of different technology types into a virtual energy storage cluster based on control objectives and real-time status.

[0036] The weight adjustment module dynamically adjusts the weight parameters of each operating target of the energy storage power station based on the scene recognition results.

[0037] The state prediction module is used to make short-term predictions of the various operating states of the energy storage power station.

[0038] The rolling optimization module is used to construct a charging and discharging command plan for the energy storage power station based on forecast information, equipment status, grid constraints and market signals;

[0039] The health assessment module is used to continuously monitor the aging and operational health of each energy storage unit in the energy storage power station, and to assess the health factors of each energy storage unit.

[0040] The decomposition and execution module is used to decompose minute-level charging and discharging commands into millisecond-level specific execution commands based on the capabilities and response times of different energy storage units.

[0041] The communication coordination module is used to support the communication needs between energy storage units under different bandwidth and latency conditions.

[0042] As a further aspect of the present invention, the specific steps of the weight adjustment module in dynamically adjusting the weight parameters of each operating target of the energy storage power station are as follows:

[0043] S4.1: Set up multiple operating scenarios for energy storage power stations by manual operation or automatic use of historical data, and establish corresponding real-time identification datasets based on the current real-time electricity price, load rate change value, frequency offset amplitude, energy storage SOC deviation from the median value, and various state quantities of new energy power fluctuation indicators. Establish corresponding identification models based on the Softmax regression model architecture.

[0044] S4.2: Based on the data requirements of the real-time recognition dataset, extract the corresponding state variables and corresponding real scene labels from the historical running data as training samples. Then, input the training samples into the recognition model in batches. The recognition model passes each training sample layer by layer through forward propagation and outputs the predicted probability of each scene.

[0045] S4.3: The cross-entropy loss function is used to calculate the difference between the predicted probability and the true label for each scene. If the difference is higher than the preset threshold, the gradient descent method is used to iteratively update the parameters of the recognition model. The training and update are repeated until the difference converges to within the preset threshold.

[0046] S4.4: Input the real-time recognition dataset into the trained recognition model, classify the current state through the recognition model, and output the scenario with the highest probability value as the current dominant running scenario. Based on the identified dominant scenario, dynamically adjust the weights of each running sub-objective.

[0047] As a further aspect of the present invention, the specific steps of the rolling optimization module in constructing the energy storage power station charge and discharge command plan are as follows:

[0048] S5.1: During daily or day-ahead periods, combining electricity price forecasts, renewable energy output forecasts, and grid dispatch signals, automatically formulate target charging and discharging profiles and preliminary revenue strategies within a preset time interval based on historical schemes, and use the power profiles of each hour as the target baseline for mid-level optimization.

[0049] S5.2: Based on the latest forecast information and the current system status, within the preset constraints of the battery SOC range, the upper and lower limits of charge and discharge power, and the grid voltage and frequency boundary, adjust the parameters of the target charge and discharge profile and the preliminary revenue strategy to form a high-resolution mid-level charge and discharge power plan.

[0050] S5.3: Decompose the power instructions per minute in the charging and discharging power plan generated in the middle layer into controllable energy storage unit operation instructions at the level of per second or millisecond, and dynamically adjust their corresponding weights according to the remaining energy, safety margin, thermal state and response rate of each energy storage unit.

[0051] S5.4: After each layer of optimization is completed, the prediction and state estimate for the next time period are updated again based on real-time monitoring data, actual power execution and load change trends. At the same time, a corresponding update cycle is set. When each update cycle has elapsed, the current energy storage power station charging and discharging command plan is adjusted and updated.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0053] This invention constructs a total task demand based on the grid and market requirements, extracts indicators of the current state of each energy storage unit, calculates the matching degree, selects the optimal candidate subset, and constructs a virtual energy storage cluster. When task demand changes or the state is mismatched, an update mechanism is automatically triggered. Combined with data on wind and solar power output, load, environment, and SOC, a multi-objective machine learning prediction model is constructed to predict the operating status and power demand trend of energy storage units. At the same time, the operating scenarios are classified in real time, and the control strategy is dynamically adjusted. Combined with day-ahead electricity prices and dispatch signals, a charging and discharging profile plan is formulated. Through hierarchical optimization, the medium-term goals are transformed into high-frequency control commands, ultimately achieving efficient, safe, and economical coordinated control of energy storage resources. It can accurately assess the adaptability of each energy storage unit to the current grid task, ensuring timely task response, sufficient power, and adequate power. It can effectively cope with complex grid demand and resource fluctuations, improve prediction accuracy and foresight, achieve seamless connection from macro strategy to micro control, and improve response accuracy and control granularity. Attached Figure Description

[0054] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0055] Figure 1 This is a flowchart of a coordinated control method for energy storage power stations proposed in this invention.

[0056] Figure 2 This is a system block diagram of an energy storage power station energy storage coordination control system proposed in this invention. Detailed Implementation

[0057] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0058] Example 1

[0059] Reference Figure 1 This embodiment discloses a coordinated control method for energy storage in an energy storage power station. The specific steps of this control method are as follows:

[0060] During the operation of the energy storage power station, the status data of each energy storage unit of the energy storage power station is collected and preprocessed in real time through edge nodes and uploaded to the central coordinator.

[0061] Specifically, the status data of each energy storage unit within a preset sliding time window is extracted in real time. The status data from different sampling frequencies and sources are time-aligned, and the status data of each energy storage unit are sorted from largest to smallest time. The status data at the 25th and 75th percentiles of each sorted status data are selected as the first and third quartiles of the corresponding energy storage unit status data. The distance between the first and third quartiles is taken as the interquartile range. Based on the first quartile, third quartile, and interquartile range, the normal threshold range of each energy storage unit status data is determined. Then, the status data within the normal threshold range is discarded, and the status data exceeding the upper and lower limits of the normal threshold range are removed. The missing values ​​of the status data of each energy storage unit extracted at different times are filled using linear interpolation. After filling, the noise information of the status data of each energy storage unit is smoothed using the exponential weighted moving average method. Finally, the smoothed energy storage units are subjected to minimum-maximum normalization.

[0062] The central coordinator dynamically constructs a virtual energy storage cluster based on current task requirements and the status of energy storage units, and performs hybrid control of the energy storage power station.

[0063] Specifically, the central coordinator collects current grid and market task requirements, including the required total power capacity, required duration, maximum allowable power fluctuation rate, and required response time. It integrates these requirements into a total task demand. Edge nodes extract parameters such as the maximum available power, available energy, efficiency coefficient, response time, and health score of each energy storage unit from the pre-processed status data. Based on these parameters, real-time status indicators are established for each energy storage unit. A comprehensive matching score is calculated between each energy storage unit and the total task demand, and these units are sorted from highest to lowest score. Units with scores below a preset matching score are removed. Based on the remaining units, a candidate energy storage subset is constructed. The subset is then verified to meet the total power and response requirements. If the sum of the maximum available power of all units in the candidate subset is lower than the required total power capacity, or if the available energy is lower than the required total power capacity, the candidate subset will be rejected. If the sum of the total power capacity is lower than the total power capacity within the required duration, the candidate energy storage subset is deemed not to meet the total power and response indicators required for the task. If the candidate energy storage subset does not meet the total power and response indicators required for the task, the matching score is reset, and energy storage units are re-selected to construct a new candidate energy storage subset. This process continues until the candidate energy storage subset meets the total power and response indicators required for the task. Then, the energy storage units in the candidate energy storage subset are logically aggregated into a virtual cluster, and the unified attributes of the virtual cluster are recorded, including cluster members, the current maximum available power of each energy storage unit, the current available energy, average response time, and average health factor. The central coordinator updates the total task demand in real time based on the current grid and market task requirements, and the edge nodes upload the current status indicators of each energy storage unit in real time. When the total task demand changes or the candidate energy storage subset does not meet the total power and response indicators required for the task, the update mechanism is triggered, and the comprehensive matching score between each energy storage unit and the current total task demand is recalculated to update the candidate energy storage subset.

[0064] It should be further explained that the specific formula for calculating the overall matching score is as follows:

[0065]

[0066] In the formula, Representing the The overall matching score of each energy storage unit; Representing the The current maximum available power of each energy storage unit; This represents the total required power capacity; Representing the The current available energy of each energy storage unit; This represents the required duration; Representing the Response time of each energy storage unit; This represents the required response time; Representing the Health score of each energy storage unit; , , as well as These represent the weighting coefficients for the capability dimension.

[0067] Based on real-time preprocessed status data, the system automatically determines the dominant operating scenario and dynamically updates the weights of each energy storage unit in the energy storage power station.

[0068] Short-term forecasts can be made on fluctuations in new energy output, load changes, and the evolution and health trends of the SOC of energy storage units.

[0069] Specifically, edge nodes preprocess historical state data of each energy storage unit in the current virtual cluster, and extract data on wind power output, photovoltaic irradiance, temperature, humidity, station load, cluster SOC weighted average, maximum SOC deviation, and disturbance factors from the preprocessed historical state data to construct corresponding training and validation sample sets. The energy storage power station prediction task is decomposed into multiple sub-objectives: new energy power prediction, short-term load change prediction, energy storage unit SOC evolution trajectory prediction, health indicator trend prediction, and local frequency disturbance sensitivity indicators. A corresponding learning model is selected based on each sub-objective, and the training sample set is used as the basis for further processing. The learning model corresponding to each sub-objective is trained independently and its performance index is recorded. Then, the error value of each learning model on the validation sample set is calculated. Then, the fusion weight is reversibly assigned according to the error of each learning model on the validation sample. When the error value of each learning model converges to the preset range during the training and validation process, the training is stopped. The state data of each energy storage unit in the energy storage power station is input into each learning model. Then, each learning model outputs the prediction result of the corresponding sub-objective. Based on the assigned fusion weight of each sub-objective, the prediction results of each sub-objective are weighted and fused to generate the final overall target prediction result. The overall target prediction results for continuous time periods are collected, and the corresponding state trend map of the energy storage power station is plotted.

[0070] Based on the forecast results, the energy storage power station is continuously optimized and controlled, the health status of each energy storage unit is assessed in real time, and the charging and discharging strategies are dynamically adjusted.

[0071] Example 2

[0072] Reference Figure 2 This embodiment discloses an energy storage power station energy storage coordination control system, including a data acquisition and processing module, a state perception module, a scene recognition module, a cluster construction module, a weight adjustment module, a state prediction module, a rolling optimization module, a health assessment module, a decomposition execution module, and a communication coordination module;

[0073] The data acquisition and processing module is used to collect various types of data from the operation of the energy storage power station, and to filter, normalize, detect anomalies, and extract features from the collected data; the status awareness module evaluates the overall operating status of the energy storage power station in real time based on the pre-processed data.

[0074] The scene recognition module is used to identify the current operating scene based on the current operating status of the energy storage power station and historical data; the cluster construction module dynamically combines energy storage units of different technology types into a virtual energy storage cluster based on control objectives and real-time status; and the weight adjustment module dynamically adjusts the weight parameters of each operating objective of the energy storage power station based on the scene recognition results.

[0075] Specifically, multiple operating scenarios for energy storage power stations are set up manually or automatically using historical data. Based on current real-time electricity prices, load rate changes, frequency offset, deviation of energy storage SOC from the median, and various state variables of new energy power fluctuation indicators, corresponding real-time identification datasets are established. A corresponding identification model is built based on a Softmax regression model architecture. According to the data requirements of the real-time identification dataset, corresponding state variables and corresponding real-scene labels are extracted from historical operating data as training samples. These training samples are then input into the identification model in batches. The identification model passes each training sample layer by layer through forward propagation and outputs the predicted probability of each scenario. The cross-entropy loss function is used to calculate the difference between the predicted probability of each scenario and the real label. If the difference is higher than a preset threshold, the gradient descent method is used to iteratively update the identification model parameters. This training and update process is repeated until the difference converges to within the preset threshold. The real-time identification dataset is then input into the trained identification model. The identification model classifies the current state and outputs the scenario with the highest probability as the current dominant operating scenario. Based on the identified dominant scenario, the weights of each operating sub-objective are dynamically adjusted.

[0076] The status prediction module is used to make short-term predictions of the various operating states of the energy storage power station; the rolling optimization module is used to construct the charging and discharging command plan of the energy storage power station based on the prediction information, equipment status, grid constraints and market signals.

[0077] Specifically, during daily or pre-day periods, based on electricity price forecasts, renewable energy output forecasts, and grid dispatch signals, the system automatically formulates target charge / discharge profiles and preliminary revenue strategies for a preset time interval based on historical data. The hourly power profiles are used as the target baseline for mid-level optimization. Based on the latest forecast information and current system status, the system adjusts various parameters in the target charge / discharge profiles and preliminary revenue strategies within preset constraints such as battery SOC range, upper and lower limits of charge / discharge power, and grid voltage and frequency boundaries. This results in a high-resolution mid-level charge / discharge power plan. The power instructions per minute in the mid-level generated charge / discharge power plan are decomposed into controllable energy storage unit operation instructions at the second or millisecond level. Simultaneously, the weights of each energy storage unit are dynamically adjusted based on its remaining energy, safety margin, thermal state, and response rate. After each optimization layer, the forecasts and status estimates for the next time period are updated again based on real-time monitoring data, actual power execution, and load change trends. A corresponding update cycle is set, and the current energy storage power station charge / discharge instruction plan is adjusted and updated after each update cycle.

[0078] The health assessment module is used to continuously monitor the aging and operational health of each energy storage unit in the energy storage power station, and to evaluate the health factors of each energy storage unit; the decomposition and execution module is used to decompose minute-level charging and discharging commands into millisecond-level specific execution commands according to the different capabilities and response times of different energy storage units; the communication coordination module is used to support the communication needs between energy storage units under different bandwidth and latency conditions.

Claims

1. A method for coordinated control of energy storage in an energy storage power station, characterized in that, The specific steps of this control method are as follows: Ⅰ. During the operation of the energy storage power station, the status data of each energy storage unit of the energy storage power station is collected and preprocessed in real time through edge nodes and uploaded to the central coordinator; II. The central coordinator dynamically constructs a virtual energy storage cluster based on the current task requirements and the status of the energy storage units, and performs hybrid control of the energy storage power station; Ⅲ. Based on the real-time preprocessed status data, automatically determine the current dominant operating scenario and dynamically update the weights of each energy storage unit in the energy storage power station; IV. To make short-term predictions on the fluctuations in new energy output, load changes, SOC evolution and health trends of energy storage power stations; V. Based on the forecast results, continuously optimize the control strategy of the energy storage power station, assess the health status of each energy storage unit in real time, and dynamically adjust the charging and discharging strategy. The specific steps for real-time acquisition and preprocessing of status data of each energy storage unit in the energy storage power station through edge nodes, as described in Step I, are as follows: S1.1: Extract the status data of each energy storage unit within the preset sliding time window in real time, align the status data of different sampling frequencies and sources in time, sort the status data of each energy storage unit in descending order of time, and select the status data at the 25% and 75% positions of each sorted status data as the first quartile and third quartile of the corresponding energy storage unit status data. S1.2: The distance between the first quartile and the third quartile is taken as the interquartile range. Then, based on the first quartile, the third quartile and the interquartile range, the normal threshold range of the state data of each energy storage unit is determined. Then, the state data within the normal threshold range is retained, and the state data exceeding the upper and lower limits of the normal threshold range is removed. S1.3: Use linear interpolation to fill in missing values ​​in the state data of each energy storage unit extracted at different times. After filling, use the exponential weighted moving average method to smooth the noise information of the state data of each energy storage unit. Then, perform minimum-maximum normalization on each smoothed energy storage unit.

2. The energy storage coordinated control method for an energy storage power station according to claim 1, characterized in that, The specific steps for dynamically constructing the virtual energy storage cluster described in step II are as follows: S2.1: The central coordinator collects current grid and market task requirements, including total required power capacity, required duration, maximum allowable power fluctuation rate, and required response time, and integrates the collected requirement information into a total task requirement; S2.2: The edge nodes extract the maximum available power, available energy, efficiency coefficient, response time, and health score of each energy storage unit from the preprocessed status data of each energy storage unit, and establish real-time status indicators for each energy storage unit based on the extracted parameters. S2.3: Based on the total current task requirements, calculate the comprehensive matching score between each energy storage unit and the total current task requirements, sort them from high to low according to the comprehensive matching score, filter out each energy storage unit with a score lower than the preset matching score, and construct the corresponding candidate energy storage subset based on the remaining energy storage units. S2.4: Verify whether the candidate energy storage subset can meet the total power and response indicators required by the task. If the sum of the current maximum available power of each energy storage unit in the candidate energy storage subset is lower than the required total power capacity, or the sum of the current available energy is lower than the sum of the total power capacity within the required duration, then it is determined that the candidate energy storage subset does not meet the total power and response indicators required by the task. S2.5: If the candidate energy storage subset does not meet the total power and response indicators required by the task, the matching score is reset and the energy storage units are re-selected to construct a candidate energy storage subset until the candidate energy storage subset meets the total power and response indicators required by the task. Then, the energy storage units in the candidate energy storage subset are logically aggregated into a virtual cluster, and the unified attributes of the virtual cluster are recorded, including cluster members, the sum of the current maximum available power of each energy storage unit and the sum of the current available energy, average response time and average health factor. S2.6: The central coordinator updates the total task demand in real time based on the current grid and market task requirements. Edge nodes upload the current status indicators of each energy storage unit in real time. When the total task demand changes or the candidate energy storage subset does not meet the total power and response indicators required by the task, the update mechanism is triggered, and the comprehensive matching score between each energy storage unit and the current total task demand is recalculated to update the candidate energy storage subset.

3. The energy storage coordinated control method for an energy storage power station according to claim 1, characterized in that, The specific steps for short-term forecasting described in step IV are as follows: S3.1: Edge nodes preprocess the historical state data of each energy storage unit in the current virtual cluster, and extract the wind power output, photovoltaic irradiance, temperature, humidity, station load, cluster SOC weighted average, maximum SOC deviation and disturbance factor of each new energy and energy storage state evolution characteristics from the preprocessed historical state data, so as to construct the corresponding training sample set and validation sample set. S3.2: Decompose the prediction task of energy storage power station into multiple sub-objectives, including new energy power prediction, short-term load change prediction, energy storage unit SOC evolution trajectory prediction, health index trend prediction, and local frequency disturbance sensitivity index, and select the corresponding learning model according to each sub-objective. S3.3: Based on the training sample set, the learning model corresponding to each sub-target is trained independently and the performance index is recorded. Then, the error value of each learning model on the validation sample set is calculated. Then, the fusion weight is reversibly allocated according to the error of each learning model on the validation sample. When the error value of each learning model converges to the preset range during the training and validation process, the training is stopped. S3.4: Input the state data of each energy storage unit in the energy storage power station into each learning model. Then, each learning model outputs the prediction results of the corresponding sub-objective. Based on the allocated fusion weights of each sub-objective, the prediction results of each sub-objective are weighted and fused to generate the final overall target prediction result. Collect the overall target prediction results for continuous time periods and draw the corresponding state trend map of the energy storage power station.

4. An energy storage power station energy storage coordination control system, used to implement the energy storage coordination control method of any one of claims 1-3, characterized in that, It includes a data acquisition and processing module, a status awareness module, a scene recognition module, a cluster construction module, a weight adjustment module, a status prediction module, a rolling optimization module, a health assessment module, a decomposition and execution module, and a communication and coordination module. The data acquisition and processing module is used to collect various types of data from the operation of the energy storage power station, and to filter, normalize, detect anomalies, and extract features from the collected data. The status perception module assesses the overall operating status of the energy storage power station in real time based on the pre-processed data. The scene recognition module is used to identify the current operating scene based on the current operating status of the energy storage power station and historical data; The cluster construction module dynamically combines energy storage units of different technology types into a virtual energy storage cluster based on control objectives and real-time status. The weight adjustment module dynamically adjusts the weight parameters of each operating target of the energy storage power station based on the scene recognition results. The state prediction module is used to make short-term predictions of the various operating states of the energy storage power station. The rolling optimization module is used to construct a charging and discharging command plan for the energy storage power station based on forecast information, equipment status, grid constraints and market signals; The health assessment module is used to continuously monitor the aging and operational health of each energy storage unit in the energy storage power station, and to assess the health factors of each energy storage unit. The decomposition and execution module is used to decompose minute-level charging and discharging commands into millisecond-level specific execution commands based on the capabilities and response times of different energy storage units. The communication coordination module is used to support the communication needs between energy storage units under different bandwidth and latency conditions.

5. The energy storage coordinated control system for an energy storage power station according to claim 4, characterized in that, The specific steps for the weight adjustment module to dynamically adjust the weight parameters of each operating target of the energy storage power station are as follows: S4.1: Set up multiple operating scenarios for energy storage power stations by manual operation or automatic use of historical data, and establish corresponding real-time identification datasets based on the current real-time electricity price, load rate change value, frequency offset amplitude, energy storage SOC deviation from the median value, and various state quantities of new energy power fluctuation indicators. Establish corresponding identification models based on the Softmax regression model architecture. S4.2: Based on the data requirements of the real-time recognition dataset, extract the corresponding state variables and corresponding real scene labels from the historical running data as training samples. Then, input the training samples into the recognition model in batches. The recognition model passes each training sample layer by layer through forward propagation and outputs the predicted probability of each scene. S4.3: The cross-entropy loss function is used to calculate the difference between the predicted probability and the true label for each scene. If the difference is higher than the preset threshold, the gradient descent method is used to iteratively update the recognition model parameters. The training and update are repeated until the difference converges to within the preset threshold. S4.4: Input the real-time recognition dataset into the trained recognition model, classify the current state through the recognition model, and output the scenario with the highest probability value as the current dominant running scenario. Based on the identified dominant scenario, dynamically adjust the weights of each running sub-objective.

6. The energy storage coordinated control system for an energy storage power station according to claim 4, characterized in that, The specific steps of the rolling optimization module in constructing the energy storage power station charge and discharge command plan are as follows: S5.1: During daily or day-ahead periods, combining electricity price forecasts, renewable energy output forecasts, and grid dispatch signals, automatically formulate target charging and discharging profiles and preliminary revenue strategies within a preset time interval based on historical schemes, and use the power profiles of each hour as the target baseline for mid-level optimization. S5.2: Based on the latest forecast information and the current system status, within the preset constraints of the battery SOC range, the upper and lower limits of charge and discharge power, and the grid voltage and frequency boundary, adjust the parameters of the target charge and discharge profile and the preliminary revenue strategy to form a high-resolution mid-level charge and discharge power plan. S5.3: Decompose the power instructions per minute in the charging and discharging power plan generated in the middle layer into controllable energy storage unit operation instructions at the level of per second or millisecond, and dynamically adjust their corresponding weights according to the remaining energy, safety margin, thermal state and response rate of each energy storage unit. S5.4: After each layer of optimization is completed, the prediction and state estimate for the next time period are updated again based on real-time monitoring data, actual power execution and load change trends. At the same time, a corresponding update cycle is set. When each update cycle has elapsed, the current energy storage power station charging and discharging command plan is adjusted and updated.

Images