Wind-solar-storage system daily and daily economic dispatching method

Abstract

The application discloses a wind-solar-storage system day-ahead and day-ahead economic dispatching method, relates to the wind-solar-storage technical field, and comprises the following steps: based on the historical operation data of the battery, a dynamic aging model is established, the actual attenuation characteristics of the battery are fitted through machine learning, the battery health state under different operation states and the change law thereof with time are acquired, and the basis of forming a dynamic adjustment strategy is formed.The application optimizes the charging and discharging of the energy storage through the health factor, the system monitors parameters such as SOH and SOC in real time, dynamically adjusts the task, reduces the battery aging, improves the stability, combines the deep reinforcement learning to optimize the SOC strategy, prolongs the service life, reduces the maintenance cost, adopts the rolling optimization scheduling and the experience playback, the system can dynamically adjust the strategy, improves the wind-solar consumption rate, optimizes the power grid balance, reduces the dependence on thermal power, reduces the scheduling cost, enhances the adaptability to extreme weather, and promotes the green and low-carbon development of electric power.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind-solar-storage, and particularly relates to a day-ahead and day-ahead economic dispatch method for a wind-solar-storage system. BACKGROUND

[0002] The day-ahead and day-ahead economic dispatch of the wind-solar-storage system refers to a joint optimization dispatch method based on wind energy, photovoltaic and energy storage system in the operation of the power system. Due to the great uncertainty of wind-solar power generation, the traditional day-ahead dispatch plan often cannot accurately match the actual power generation and load demand, so it is necessary to adjust and optimize in the day. Specifically, first, an optimized dispatch scheme is formulated based on weather prediction, power grid load prediction and energy storage state in the day-ahead (i.e. the previous day), but due to the possible deviation between the predicted value and the actual value, in the operation process of the next day, the actual situation of wind-solar power generation and load is monitored in real time, and the dispatch strategy is continuously adjusted and optimized. This method relies on the idea of rolling optimization, that is, based on the latest real-time data, the day-ahead dispatch result is corrected, so that the final dispatch scheme is closer to the actual operation state, thereby reducing the control error caused by the deviation and improving the economy and stability of the system operation.

[0003] The prior art has the following disadvantages: the cumulative error of the dynamic aging effect of the energy storage system in the day-ahead and day-ahead dispatch is a problem that is easily overlooked but may have serious consequences. The existing wind-solar-storage dispatch model usually assumes that the charging and discharging efficiency of the energy storage battery is constant and is dispatched according to the fixed charging and discharging curve, while ignoring the nonlinear aging effect of the battery caused by factors such as temperature change and current impact in the process of high-frequency charging and discharging. This aging not only reduces the actual available capacity of the battery, but also may cause the energy balance error in the dispatch optimization calculation to gradually accumulate, especially after a long time of operation, the available capacity of the battery and the dispatch plan may deviate, causing the energy storage to fail to release power as planned at critical moments, affecting the stability of the system. More seriously, in extreme cases, excessive discharge may cause internal short circuit or thermal runaway of the battery, increasing the risk of equipment damage and even safety accidents.

[0004] The above information disclosed in the background section is only used to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a wind-solar-storage system day-ahead and day-ahead economic dispatch method. Through the health factor driven energy storage charging and discharging optimization method, the system can monitor the parameters such as SOH, SOC, temperature and current in real time, dynamically adjust the charging and discharging task, reduce the battery unbalanced aging, improve the system stability, combine with the deep reinforcement learning optimization SOC management strategy, prolong the battery life, reduce the maintenance cost. In addition, by using the rolling optimization scheduling and experience playback mechanism, the system can dynamically adjust the scheduling strategy based on real-time load demand, wind-solar prediction error and other factors, improve the wind-solar consumption rate, optimize the power grid energy balance, reduce the dependence on thermal power, reduce the scheduling cost, enhance the adaptability of extreme weather, promote the green and low-carbon development of power system, so as to solve the problems in the above background technology.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: a wind-solar-storage system day-ahead and day-ahead economic dispatch method, comprising the following steps:

[0007] Based on the historical operation data of the battery, a dynamic aging model is established, the actual attenuation characteristics of the battery are fitted through machine learning, the battery health status under different operating conditions and its change rule with time are obtained, and the basis for forming a dynamic adjustment strategy is formed;

[0008] In the day-ahead scheduling stage, combined with the wind-solar power generation prediction data, the power grid load prediction data and the health status of the energy storage system, a multi-objective optimization method is used to calculate the optimal charging and discharging plan of the energy storage, to ensure that the wind-solar consumption is maximized while reducing the impact of high-rate charging and discharging on battery life, and to provide a benchmark scheme for day-ahead scheduling;

[0009] In the day-ahead scheduling process, the actual output of wind-solar power generation, the real-time demand of power grid load and the actual available capacity of energy storage system are obtained by using high-precision sensors and state estimation method, and compared with the day-ahead scheduling plan to identify the deviation, so as to determine whether the energy storage charging and discharging plan needs to be adjusted;

[0010] Based on the battery health status parameters calculated by the energy storage dynamic aging model, the scheduling strategy in actual operation is adjusted, the charging and discharging plan is corrected by introducing the health factor, the overcharging and discharging of the battery is prevented, and the load distribution of different battery groups is optimized, so that the overall attenuation of the energy storage system is balanced, the service life is prolonged and the scheduling deviation caused by uncertainty is reduced;

[0011] Based on real-time monitoring data and aging compensation control mechanism, a rolling time domain optimization method is used to dynamically correct the day-ahead scheduling scheme, the optimal energy storage charging and discharging strategy in a future period of time is recalculated every certain time interval, so that it can adapt to the dynamic changes of wind-solar output and load demand, and reduce the energy balance problem caused by prediction error;

[0012] In combination with historical operation data and real-time scheduling feedback, the scheduling parameters are dynamically adjusted using a reinforcement learning algorithm, so that the energy storage system can adapt to the long-term changes in battery aging state, improve the utilization efficiency of energy storage, and optimize the scheduling strategy under different operating scenarios, thereby reducing the cumulative error in the long-term operation process, making the energy storage system intelligent and economical.

[0013] Preferably, based on the historical operation data of the battery, a dynamic aging model is established, the actual attenuation characteristics of the battery are fitted through machine learning, the battery health state under different operating conditions and its variation law over time are obtained, and the specific steps to form the basis of the dynamic adjustment strategy are as follows:

[0014] A battery operation database is constructed to collect and clean the key data of charging and discharging current, voltage and temperature, remove outliers and standardize the processing to ensure data integrity and consistency;

[0015] The key feature variables affecting battery aging are screened, correlation analysis is used to optimize variable selection, and the modeling accuracy is improved;

[0016] A machine learning algorithm is used to train the SOH prediction model, the hyperparameters are optimized and the prediction accuracy is evaluated to ensure that the model accurately reflects the changes in battery health state;

[0017] The SOH prediction model is integrated into the energy storage management system, a rolling update mechanism is used for continuous optimization, and the charging and discharging strategy is adjusted using an optimization algorithm to prolong the battery life and improve the system stability.

[0018] Preferably, in the day-ahead scheduling stage, a multi-objective optimization method is used to calculate the optimal charging and discharging plan of the energy storage, to ensure that the wind and light consumption is maximized while reducing the impact of high-rate charging and discharging on battery life, and to provide a benchmark solution for the intra-day scheduling. The specific steps are as follows:

[0019] Integrate wind and light power generation prediction, power grid load prediction and energy storage health state data, and use uncertainty modeling methods to reduce the impact of prediction errors on scheduling decisions;

[0020] A multi-objective optimization model is constructed to consider wind and light consumption, battery aging and economic cost, and a Pareto optimization algorithm is used to dynamically balance different optimization objectives;

[0021] Based on mixed integer linear programming, the charging and discharging time, power and path of the energy storage are optimized, and the scheduling adaptability is improved by combining the MPC rolling optimization method;

[0022] A benchmark scheduling scheme is developed, including dynamic SOC window adjustment and intelligent equalization charging and discharging strategy, and a rolling optimization algorithm is used to continuously correct and optimize in the intra-day scheduling stage.

[0023] Preferably, during the day scheduling process, the actual output of wind and solar power generation, the real-time demand of the power grid load and the actual available capacity of the energy storage system are obtained by using high-precision sensors and state estimation methods, and are compared with the day-ahead scheduling plan to identify the existing deviation, so as to determine whether the specific steps of adjusting the energy storage charging and discharging plan are as follows:

[0024] The wind and solar output, load demand and energy storage system state are obtained by using high-precision sensors and state estimation methods, and the measurement accuracy is improved by data filtering and machine learning methods;

[0025] The key deviation indicators of wind and solar output, load demand and energy storage system are calculated by comparing the actual running state with the day-ahead scheduling plan, and the influence of the deviation on the system operation is evaluated;

[0026] The future charging and discharging strategy is recalculated based on the MPC rolling optimization method, the SOC window is optimized, and the economy and the service life of the energy storage system are improved in combination with the grid scheduling and market electricity price;

[0027] The optimized scheduling scheme is executed, the energy storage running state is controlled in real time, and the long-term scheduling strategy is optimized by using reinforcement learning technology to improve the intelligent level of the system.

[0028] Preferably, based on the battery health state parameters calculated by the energy storage dynamic aging model, the scheduling strategy in actual operation is adjusted, the charging and discharging plan is corrected by introducing the health factor to prevent excessive charging and discharging of the battery, and the load distribution of different battery groups is optimized to balance the overall attenuation of the energy storage system, prolong the service life and reduce the scheduling deviation caused by uncertainty. The specific steps are as follows:

[0029] Based on the key factors of battery SOH, SOC and cycle number, the HF value is calculated by the weighted comprehensive evaluation method to measure the battery health state and provide a basis for scheduling optimization;

[0030] According to the HF value, the charging and discharging task allocation of the battery group is optimized, the SOC working window is adjusted to ensure that the batteries with good health state bear more load, and the overall aging of the energy storage system is balanced;

[0031] The day-ahead-day rolling optimization mechanism is adopted to dynamically calculate the future scheduling strategy every same time, and the adaptive control algorithm is combined to make the energy storage scheduling more flexible and economical;

[0032] A closed-loop monitoring mechanism is established, and the long-term scheduling strategy is optimized by using reinforcement learning method, so that the energy storage system can continuously optimize in different operating environments, prolong the battery life and improve the overall operating efficiency.

[0033] Preferably, based on real-time monitoring data and aging compensation control mechanism, the rolling time domain optimization method is used to dynamically correct the day-ahead scheduling scheme, the optimal energy storage charging and discharging strategy in a future period of time is recalculated every set time interval, so as to adapt to the dynamic changes of wind and light output and load demand, and to reduce the energy balance problem caused by prediction error. The specific steps are as follows:

[0034] At the current time, based on wind and light power generation, grid load and energy storage system state, a rolling optimization objective function is established to minimize the aging loss of the energy storage system while ensuring system energy balance, within a prediction time domain T p , the wind and light power generation and load demand at future time are predicted, and the available power of the energy storage system and the optimal scheduling strategy are calculated in combination with the health factor of the battery. The optimization objective is defined as follows:

[0035]

[0036] , wherein J is the optimization objective function, which measures the comprehensive cost of the energy storage scheduling strategy, t k is the current time, T p is the prediction time domain, t is the optimization time, which represents each discrete time point in the rolling optimization process, P b (t) is the charging and discharging power of the energy storage system at time t, SOC(t) is the state of charge of the energy storage system at time t, SOC ref is the target SOC value, HF(t) is the battery health factor, which is used to measure the health status of the battery, λ1 is the energy storage power adjustment weight, which controls the power fluctuation of the energy storage system, λ2 is the SOC deviation penalty weight, and λ3 is the battery aging influence weight.

[0037] Based on the optimization objective function J, the constraint conditions are constructed to ensure that the optimal scheduling strategy is feasible and meets the system operation requirements. At each prediction time domain T p , the MPC adopts the rolling optimization strategy, that is, the optimal energy storage charging and discharging strategy is calculated at the current time t k , but only the charging and discharging instructions in the current optimization period are executed, and then the next optimization period is entered to update the prediction data and solve the optimization problem again. The energy balance equation is as follows:

[0038]

[0039] , wherein SOC(t+1) is the state of charge at the next time, η c is the charging efficiency, η d is the discharging efficiency, and are the optimal charging power and discharging power at time t, respectively, which are obtained by optimization of the objective function J, and C bis the rated capacity of the battery, which determines the SOC change rate.

[0040] Preferably, combined with historical operation data and real-time scheduling feedback, the scheduling parameters are dynamically adjusted using a reinforcement learning algorithm, so that the energy storage system can adapt to the long-term changes in battery aging state, improve the utilization efficiency of energy storage, and optimize the scheduling strategy under different operating scenarios, thereby reducing the cumulative error in the long-term operation process, making the energy storage system intelligent and economical. The specific steps are as follows:

[0041] First, define the state, action and reward function of reinforcement learning, and construct the state transition equation, so that the model can learn the optimal scheduling strategy of the energy storage system under different operating environments;

[0042] State S t includes the current battery health state SOH, remaining power SOC, battery temperature T b , wind and solar power prediction value P wp , grid load demand P load and market electricity price C m key information, defined as:

[0043] S t ={SOH t , SOC t , T b , P wp , P load , C m}

[0044] Action A t represents the scheduling decision of the energy storage system at the current time;

[0045] A t ={P ch , P dis , ΔSOC}

[0046] wherein P ch is the charging power, P dis is the discharging power and ΔSOC is the SOC adjustment strategy;

[0047] The state transition equation is based on the battery dynamic aging model to calculate the SOH and temperature change at the next time:

[0048] SOH t+1 = SOH t - α · f(T b , P ch , P dis )

[0049] wherein SOH t is the battery health state at time t, SOH t+1is the battery state of health at next time, a is the aging factor, f(T b , P ch , P dis ) represents the decay rate of the battery due to temperature, charge and discharge power, T b is the battery temperature, P ch is the charging power, P dis is the discharging power;

[0050] The reward function is designed to optimize the scheduling strategy, a composite reward function is designed, defined as:

[0051]

[0052] , wherein R t is the reward function, P curtail is the curtailment power, β1 is the battery life optimization weight, β2 is the wind and light consumption optimization weight, and β3 is the economic cost optimization weight, which measures the role of market electricity price and charging and discharging cost in the optimization target.

[0053] Preferably, after determining the state S t , the state A t and the reward function R t , a deep reinforcement learning based on Actor-Critic architecture is used for policy training, so that the energy storage system can adapt to the battery aging state and optimize the scheduling strategy under different scenarios.

[0054] The Actor network is responsible for generating the state A t from the state S t , and the formula is as follows:

[0055] A t =π θ (S t )+N t

[0056] , wherein π θ (S t ) is a parameterized policy, and N t is an exploration noise for balancing exploration and utilization.

[0057] The Critic network is used to evaluate whether the policy of the Actor network is optimal, and the Q value is calculated, and the calculation formula is as follows:

[0058] Q(S t , A t )=R t +γQ(S t+1 , A t+1 )

[0059] , wherein Q(S t , At ) is a state-action value function, i.e., Q-value, γ is a discount factor to control the importance of future rewards, Q(S t+1 , A t+1 ) is the Q-value estimation at next time t+1;

[0060] Experience replay stores past state-action-reward-state transition data into an experience replay pool and randomly samples from it for training to avoid gradient shock problems caused by data correlation;

[0061] Policy update, gradient descent is used to optimize the Actor and Critic networks, update the policy parameters θ to maximize the cumulative reward, the update formula is as follows:

[0062]

[0063] , wherein θ is the training parameter of the Actor network, η is the learning rate, is the parameter update direction based on the policy gradient.

[0064] In the above technical solution, the technical effects and advantages provided by the present application are:

[0065] The energy storage charge and discharge optimization method driven by the health factor (HF) can effectively reduce the uneven aging phenomenon of the energy storage battery and improve the long-term operation stability of the entire energy storage system. In the traditional scheduling method, all battery groups are often operated according to the same charge and discharge strategy, which cannot be dynamically adjusted according to the health status of a single battery, resulting in accelerated aging of some batteries due to high-rate discharge, long-term high-SOC operation or temperature abnormalities, and ultimately affecting the reliability of the entire system. The present application monitors the SOH, SOC, temperature, current and other parameters in real time, calculates the battery health factor (HF), and intelligently adjusts the charge and discharge task allocation according to the HF value to ensure that the batteries with good health status bear more power, and the batteries with severe aging reduce deep discharge and high-rate operation. This dynamic optimization strategy can prolong the overall life of the energy storage system, reduce scheduling errors caused by individual battery degradation, and improve the stability of the system. In addition, combined with deep reinforcement learning (DRL) technology, the system can continuously learn and optimize the SOC management strategy in long-term operation, such as dynamically adjusting the SOC window and optimizing the charge and discharge rate, to minimize the battery aging rate. This not only reduces the maintenance and replacement cost of the energy storage battery, but also improves the long-term investment return rate of the energy storage system, making the wind-solar-storage system have better economic feasibility and operation reliability.

[0066] The application combines rolling optimization scheduling (MPC) with an experience replay mechanism (ReplayBuffer), so that the system can dynamically adjust the charging and discharging strategy according to real-time load demand, market electricity price, wind and light prediction error and other factors, thereby improving the wind and light consumption rate and optimizing the energy balance of the power grid. Specifically, in the intraday scheduling stage, the system can perform rolling optimization calculation based on the latest load and wind and light output data every 15-30 minutes, ensuring that the scheduling scheme always remains consistent with the actual operating state and reducing scheduling deviations caused by prediction errors. In addition, through reinforcement learning technology, the system can continuously optimize the scheduling strategy under different weather conditions and load characteristics during long-term operation, improving the system's adaptability to extreme weather or sudden load fluctuations. For example, when encountering continuous low wind and low light weather, the system can use historical experience data to automatically switch to a more optimal energy storage management strategy, such as pre-charging, reducing discharging power, optimizing charging timing, etc., to ensure the stable operation of the power grid. Ultimately, this optimization scheme can improve the utilization rate of wind and light power generation, reduce the dependence on standby thermal power units, reduce the cost of power grid scheduling, and improve the penetration rate of new energy in the power system, promoting the development of the power system in a more green and low-carbon direction. BRIEF DESCRIPTION OF DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only represent some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0068] Figure 1 The method flow chart of the wind and light storage system day-ahead and intraday economic dispatching method of the present application. DETAILED DESCRIPTION

[0069] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the gist of each example to those skilled in the art.

[0070] The present application provides a wind and light storage system day-ahead and intraday economic dispatching method as shown in Figure 1 characterized in that it comprises the following steps:

[0071] Based on the historical operation data of the battery, a dynamic aging model is established, which includes factors such as temperature influence, cycle number, and charge-discharge rate. The actual degradation characteristics of the battery are fitted through machine learning to obtain the battery health state (SOH) and its variation law with time under different operating conditions, to form the basis for dynamic adjustment strategy;

[0072] Based on the historical operation data of the battery, a dynamic aging model is established, which includes factors such as temperature influence, cycle number, and charge-discharge rate. The actual degradation characteristics of the battery are fitted through machine learning to obtain the battery health state (SOH) and its variation law with time under different operating conditions, to form the basis for dynamic adjustment strategy's specific steps are as follows:

[0073] Construct a battery operation database, collect and clean the charge-discharge current, voltage, temperature key data, remove outliers and standardize processing to ensure data integrity and consistency;

[0074] First, build a battery operation database to collect long-term running battery data, including charge-discharge current, voltage, temperature, cycle number, charge-discharge rate, SOC (State of Charge), SOH (State of Health) and other key parameters. Data sources can be historical operation records, laboratory accelerated aging test data, or real-time monitoring data of large-scale energy storage systems. During data collection, outliers need to be detected and invalid data caused by sensor errors, communication failures, etc. are removed, and interpolation, smoothing filtering and other methods are used to fill in missing data to ensure data integrity and reliability. At the same time, standardize the data from different sources to ensure that all data are within the same scale range, making subsequent modeling more accurate.

[0075] Screening key characteristic variables affecting battery aging, such as internal resistance growth rate and capacity attenuation rate, using correlation analysis to optimize variable selection and improve modeling accuracy;

[0076] In data preprocessing, feature variables related to battery aging are extracted, such as battery internal resistance growth rate, maximum available capacity decay rate, charge-discharge efficiency change curve, and the impact of temperature change on capacity retention rate. Feature extraction can use statistical analysis methods such as correlation analysis and principal component analysis (PCA) to screen out the variables that have the greatest impact on SOH. In addition, gray correlation analysis or SHAP (Shapley Additive Explanations) and other explanatory AI technologies can be used to analyze the contribution of each variable to SOH changes, remove redundant variables, and only retain features that have a significant impact on battery health status. For example, some batteries may be particularly sensitive to temperature changes, while others may be more sensitive to discharge rates, so the selection of feature variables may vary depending on the application scenario.

[0077] Using machine learning algorithms to train SOH prediction models, optimize hyperparameters, and evaluate prediction accuracy to ensure that the model accurately reflects changes in battery health status;

[0078] Select machine learning algorithms suitable for processing time series data, such as long short-term memory networks (LSTM), random forest regression, or gradient boosting trees (GBDT), to train historical data and establish SOH prediction models. When modeling, the data set needs to be divided into training, validation, and test sets, and cross-validation methods are used to optimize model hyperparameters to improve generalization ability. After model training, the prediction error of the model can be evaluated based on the validation and test sets, such as mean squared error (MSE), root mean squared error (RMSE), and R 2 Determination coefficient, etc. If the prediction error of the model is large, the accuracy can be improved by adjusting the model hyperparameters, adding feature variables, or using more complex deep learning networks such as Transformer or CNN-LSTM hybrid architecture.

[0079] Integrate the SOH prediction model into the energy storage management system, use a rolling update mechanism to continuously optimize, and adjust the charge-discharge strategy in combination with optimization algorithms to extend battery life and improve system stability.

[0080] In the model deployment phase, the trained SOH prediction model needs to be integrated into the energy management system (EMS) to realize real-time monitoring and prediction of the battery health status. At the same time, a rolling update mechanism is adopted to periodically train the model online using the latest operation data, so that it can adapt to the long-term changes in the battery aging state. In addition, parameter optimization methods such as Bayesian optimization or genetic algorithm can be combined to dynamically adjust the charging and discharging strategy of the energy storage system according to the current SOH of the battery, such as reducing the frequency of high-rate charging and discharging or optimizing the charging and discharging window to prolong the service life of the battery. Ultimately, the model can not only be used for energy storage scheduling optimization, but also for early warning of possible battery failures to improve the safety and economy of the wind-solar-storage system.

[0081] In the day-ahead scheduling phase, combined with wind-solar power generation prediction data, grid load prediction data and the health status of the energy storage system, a multi-objective optimization method is used to calculate the optimal charging and discharging plan of the energy storage to ensure that the impact of high-rate charging and discharging on battery life is reduced while maximizing wind-solar consumption, and to provide a benchmark solution for intraday scheduling;

[0082] In the day-ahead scheduling phase, a multi-objective optimization method is used to calculate the optimal charging and discharging plan of the energy storage to ensure that the impact of high-rate charging and discharging on battery life is reduced while maximizing wind-solar consumption, and to provide a benchmark solution for intraday scheduling. The specific steps are as follows:

[0083] Integrate wind-solar power generation prediction, grid load prediction and energy storage health status data, and use uncertainty modeling methods to reduce the impact of prediction errors on scheduling decisions;

[0084] First, integrate wind-solar power generation prediction data, grid load prediction data and the health status data of the energy storage system to provide comprehensive information support for scheduling optimization. Wind-solar power generation prediction data is usually generated by weather models, historical generation data and short-term prediction algorithms such as time series ARIMA model, LSTM, etc. Grid load prediction is analyzed in combination with historical load curve, weather factors, electricity usage habits, etc. While the health status data of the energy storage system is based on the aforementioned SOH prediction model, through machine learning analysis of temperature, cycle number, charging and discharging rate, etc., the remaining available capacity, maximum allowable charging and discharging power and optimal charging and discharging window of the battery are dynamically evaluated. In addition, to deal with errors in wind-solar power generation and load prediction, methods such as Monte Carlo simulation and Kalman filtering can be used to model uncertainty, making the scheduling solution more robust and reducing the impact of prediction errors on charging and discharging decisions.

[0085] Build a multi-objective optimization model that takes into account wind-solar consumption, battery aging and economic cost, and use Pareto optimization algorithm to dynamically balance different optimization objectives;

[0086] After data fusion, a multi-objective optimization model needs to be built to balance the maximum wind-solar consumption, the minimum battery aging, and the optimal system operation cost, etc. The optimization objectives can be set as: ① maximize the self-consumption ratio of wind-solar power generation to avoid energy waste caused by curtailment; ② minimize battery aging loss by limiting high-rate charging and discharging, controlling the SOC working range, and balancing the load of battery packs to extend the life; ③ reduce system operation cost, including peak shaving to reduce power purchase cost and reducing maintenance costs caused by frequent start-stop. To achieve these objectives, a Pareto multi-objective optimization method based on weights can be used to dynamically balance each optimization objective under different operating scenarios. In addition, the constraint conditions need to consider the maximum charging and discharging power of the battery, the minimum SOC limit, the power balance constraint of the grid, etc. to ensure that the optimization results are feasible and meet the physical limitations.

[0087] Based on mixed integer linear programming, the charging and discharging time, power and path of the energy storage are optimized, and the rolling optimization method of MPC is used to improve the adaptability of scheduling;

[0088] Based on the multi-objective optimization model, the mixed integer linear programming (MILP) is used to solve the optimal scheduling scheme. In this process, the system needs to consider the following key factors: ① determine the optimal charging and discharging time window to ensure that the battery is charged when the electricity price is low and the photovoltaic output is high, and discharged when the load is high or the wind-solar output is low; ② limit high-rate charging and discharging, dynamically adjust the maximum charging and discharging power according to the SOH prediction model, and prevent rapid aging; ③ optimize the charging and discharging path of the energy storage system, for example, in a multi-battery pack energy storage system, preferentially use the battery with better health status to balance the aging and optimize the overall life. In the calculation process, the rolling time domain optimization method (MPC) can be used to recalculate the optimal charging and discharging strategy for a certain period of time in the future at certain time intervals to adapt to the dynamic changes of wind-solar prediction and improve the adaptability of the plan.

[0089] Develop a benchmark scheduling scheme including dynamic SOC window adjustment and intelligent balancing charging and discharging strategy, and combine the rolling optimization algorithm to continuously correct the optimization in the intra-day scheduling stage.

[0090] After the optimal charging and discharging plan is solved, it is used as a benchmark scheduling scheme to provide optimization reference for intraday scheduling. This benchmark scheme not only includes the charging and discharging time and power curve of the energy storage, but also includes the battery health management strategy, such as dynamic SOC window adjustment (flexibly adjusting the SOC working range according to the SOH state) and intelligent equalization charging and discharging strategy (allocating different charging and discharging tasks for battery packs with different aging degrees). In addition, in order to adapt to the uncertainty of intraday wind and light output and load, the benchmark scheme can be combined with the day-ahead-intraday rolling optimization algorithm to continuously update and optimize in the intraday scheduling stage, so as to reduce the control error caused by deviation. Finally, this optimization scheme not only can improve the wind and light consumption ratio and reduce the system operation cost, but also can prolong the service life of the energy storage system, providing long-term guarantee for the stable operation of the wind and light storage system.

[0091] In the intraday scheduling process, the actual output of wind and light generation, the real-time demand of grid load and the actual available capacity of energy storage system are obtained by using high-precision sensors and state estimation methods, and are compared with the day-ahead scheduling plan to identify the existing deviation, so as to determine whether the energy storage charging and discharging plan needs to be adjusted;

[0092] In the intraday scheduling process, the actual output of wind and light generation, the real-time demand of grid load and the actual available capacity of energy storage system are obtained by using high-precision sensors and state estimation methods, and are compared with the day-ahead scheduling plan to identify the existing deviation, so as to determine whether the energy storage charging and discharging plan needs to be adjusted; the specific steps are as follows:

[0093] The wind and light output, load demand and energy storage system state are obtained by using high-precision sensors and state estimation methods, and the measurement accuracy is improved by data filtering and machine learning methods;

[0094] Firstly, the actual output of wind and light generation, the real-time demand of grid load and the current running state of energy storage system, including battery SOC, SOH, temperature, voltage, current and other key parameters, are obtained by using high-precision sensors and data acquisition system (SCADA, PMU, etc.). These data are preprocessed by edge computing nodes or cloud computing platforms to eliminate outliers, filter noise, and improve measurement accuracy by using data assimilation techniques (such as Kalman filter, particle filter, etc.). In addition, combined with historical data and real-time monitoring data, a state estimation model is constructed to estimate the true value of unmeasured variables by machine learning or Bayesian inference method. For example, for a large-scale energy storage system, if some battery sensors are abnormal, the state of these sensors can be estimated by the data of other normal batteries and the aging model, to ensure the accuracy of scheduling decision.

[0095] The actual running state is compared with the day-ahead scheduling plan, the key deviation indicators of wind and light output, load demand and energy storage system are calculated, and the influence of deviation on system operation is evaluated;

[0096] After data collection and state estimation, the actual operating state needs to be compared with the day-ahead scheduling plan, and key deviation indicators need to be calculated, such as: ① wind and solar power output deviation (actual power generation vs. predicted power generation), ② load demand deviation (real-time load vs. predicted load), ③ energy storage system deviation (actual available capacity of battery vs. predicted available capacity). For different types of deviations, further analysis is needed: if the wind and solar power output deviation is large, the discharge power of the energy storage system may need to be increased to balance the system power; if the load demand deviation is large, the charging and discharging timing of the energy storage system may need to be adjusted; if the available capacity of the energy storage system is lower than expected, the charging and discharging rate may need to be adjusted or the battery pack with better health status may need to be switched to. In addition, in order to improve the robustness of the scheduling decision, probability analysis methods such as Monte Carlo simulation can be used to calculate the energy gap or excess that may be caused by the deviation, and to evaluate the risk level of different scheduling adjustment schemes, ensuring that the adjustment strategy meets the supply and demand balance while not exacerbating the aging of the energy storage system or increasing the operating cost.

[0097] Recalculate the future charging and discharging strategy based on the MPC rolling optimization method, optimize the SOC window, and combine the grid scheduling and market electricity price to improve the economy and the life of the energy storage system;

[0098] Based on the deviation analysis results, the real-time rolling optimization method (Model Predictive Control, MPC) is used to recalculate the optimal energy storage charging and discharging strategy for a certain period of time in the future. Specifically, this optimization process needs to consider: ① the current SOC, SOH and maximum allowed charging and discharging power of the energy storage system, to ensure that the adjustment strategy does not harm the battery health; ② the grid scheduling requirements, to avoid frequent adjustment of charging and discharging power leading to system instability; ③ market electricity price (if there is a power market), to ensure that the energy storage scheduling is optimal in terms of economy. This optimization calculation usually uses methods such as mixed integer linear programming (MILP) or deep reinforcement learning (DRL) to optimize the charging and discharging decision while considering battery health. For example, if it is found that the wind and solar power output is lower than expected, the system can charge extra at low electricity price period to compensate for the future power gap, or reduce charging at load valley to avoid excessive cycling of the battery. In addition, the optimization process can also dynamically adjust the SOC window, i.e. adjust the maximum and minimum SOC range according to the battery health status, to extend the service life of the energy storage system.

[0099] Execute the optimized scheduling scheme, real-time feedback control the energy storage operating state, and use reinforcement learning technology to optimize the long-term scheduling strategy to improve the intelligent level of the system.

[0100] After the optimization calculation is completed, a new energy storage charge and discharge plan is generated and issued to the energy storage management system (EMS) to control the energy storage device to operate according to the adjusted strategy. In order to ensure that the adjustment scheme can be effectively implemented, the system will perform feedback control in a short period (such as 15 minutes or 30 minutes) to monitor the response of the energy storage system in real time. If new deviations are found (such as the energy storage failing to discharge normally according to the adjusted scheme), the system will trigger an abnormal handling mechanism, such as re-optimizing the dispatch, adjusting the standby power strategy, or notifying manual intervention. In addition, reinforcement learning technology can be used to enable the system to continuously learn and optimize the dispatch strategy under different environments during long-term operation. For example, during long-term operation, the system can gradually learn the wind and light output patterns under specific weather conditions and the impact of different SOC management strategies on battery life, thereby continuously improving the intelligent level of dispatch and ultimately achieving efficient, stable, and low-cost economic dispatch of the wind-solar-storage system.

[0101] Based on the battery health state parameters calculated by the energy storage dynamic aging model, the dispatch strategy in actual operation is adjusted, the charge and discharge plan is corrected by introducing a health factor (HF), overcharging and discharging of the battery is prevented, and the load distribution of different battery groups is optimized to balance the overall degradation of the energy storage system, prolong the service life, and reduce the dispatch deviation caused by uncertainty.

[0102] Based on the battery health state parameters calculated by the energy storage dynamic aging model, the dispatch strategy in actual operation is adjusted, the charge and discharge plan is corrected by introducing a health factor (HF), overcharging and discharging of the battery is prevented, and the load distribution of different battery groups is optimized to balance the overall degradation of the energy storage system, prolong the service life, and reduce the dispatch deviation caused by uncertainty. The specific steps are as follows:

[0103] Based on the key factors of battery SOH, SOC, and cycle number, the HF value is calculated by a weighted comprehensive evaluation method to measure the battery health state and provide a basis for dispatch optimization.

[0104] Health factor (HF) is a dynamic adjustment parameter to measure the state of battery health, which is calculated based on multiple key parameters such as SOH (State of Health), SOC (State of Charge), cycle number, temperature, and charge-discharge rate. First, through the dynamic aging model of energy storage, the battery health status is monitored and predicted in real time, and the core variables that affect battery life are extracted. For example, some batteries may grow faster in internal resistance due to long-term high-rate discharge, while others may have lithium metal precipitation due to long-term high SOC, affecting safety. Then, the weighted comprehensive evaluation method (such as entropy weight method, AHP) is used to calculate the HF value of each battery, and the HF range is usually set between 0 and 1. The lower the HF value, the higher the degree of battery aging, and the need to reduce its charge-discharge load, while the battery with higher HF value can undertake more charge-discharge tasks to prolong the service life of the entire energy storage system.

[0105] According to the HF value, the charge-discharge task allocation of the battery pack is optimized, the SOC working window is adjusted, and the battery with good health status is ensured to undertake more load, achieving overall aging balance of the energy storage system;

[0106] After calculating the HF value, the charge-discharge task allocation needs to be optimized based on the health status of different battery packs to prevent individual battery packs from over-aging and achieve overall health balance of the energy storage system. Specifically, the dispatching system allocates charge-discharge power according to the HF value, and the battery with high HF value undertakes larger power first, and the battery with low HF value appropriately reduces power output or reduces the number of deep charge-discharge. In addition, the SOC working window can also be dynamically adjusted, for example, for the battery with poor health status, a narrower SOC range (such as 30%-70%) is set to reduce the impact of high SOC residence time and deep charge-discharge, while for the battery with good health status, a larger SOC range (such as 20%-90%) is allowed to improve the overall scheduling flexibility of the system. This process can use nonlinear optimization methods (such as particle swarm optimization PSO, mixed integer linear programming MILP) or reinforcement learning algorithms (such as deep reinforcement learning DRL) to optimize the load allocation strategy, and continuously adjust during operation to make the system adapt to scheduling requirements under different health status.

[0107] A day-ahead-day-in rolling optimization mechanism is adopted to dynamically calculate the future scheduling strategy every same time, combined with an adaptive control algorithm to make the energy storage scheduling more flexible and economical;

[0108] On the basis of optimizing load distribution, it is also necessary to combine the day-ahead and intra-day rolling optimization mechanism to adjust the charging and discharging plan in real time to adapt to the uncertainty of wind and light generation and the fluctuation of load demand. Specifically, the dispatching system will recalculate the optimal energy storage charging and discharging strategy for the future dispatching period every certain time (such as 15 minutes, 30 minutes), and dynamically optimize based on the HF value. During the rolling optimization process, the dispatching system needs to: ① adjust the charging plan in combination with the latest wind and light generation prediction data to avoid the duration of high SOC state being too long; ② adjust the discharging plan according to the grid load demand to prevent the battery from bearing too much load in the low SOH state; ③ set the battery rotation strategy to ensure the balanced use frequency of each battery pack and avoid the accelerated capacity decay of individual batteries due to long-term non-participation in charging and discharging. During this process, adaptive control algorithms (such as fuzzy control, Bayesian optimization) can be used to enable the system to self-learn and continuously optimize the dispatching strategy based on historical data and real-time operating status, so as to maximize the economic benefits and life of the energy storage system.

[0109] A closed-loop monitoring mechanism is established to optimize the long-term dispatching strategy using reinforcement learning methods, so that the energy storage system can continuously optimize in different operating environments, prolong the battery life and improve the overall operating efficiency.

[0110] After the execution of the adjusted charging and discharging strategy, the system needs to monitor the changes in the battery health status in real time to form a closed-loop control mechanism and ensure that the optimization effect meets the expectations. If it is found that the SOH of a certain battery pack is still declining too fast, or the system is over-dispatched in certain load scenarios, the HF calculation method needs to be further optimized, or the load distribution strategy of the battery pack needs to be adjusted. In addition, reinforcement learning techniques (such as deep Q learning DQN) can be used to enable the system to continuously iterate and optimize during long-term operation, improving the adaptive ability of the energy storage system in different operating environments. For example, the system can analyze the optimal HF dispatching strategy under different seasons and different wind and light output modes, and continuously optimize parameters such as SOC window and charging and discharging rate based on historical data to ensure that the system is always in the best health status during long-term operation. Ultimately, this continuous optimization mechanism can reduce energy storage aging while improving wind and light consumption rate, optimizing economic efficiency, and ensuring the safe and stable operation of the power grid.

[0111] Based on real-time monitoring data and aging compensation control mechanism, the rolling time domain optimization (Model Predictive Control, MPC) method is used to dynamically correct the day-ahead dispatching scheme, recalculate the optimal energy storage charging and discharging strategy for a certain period of time every certain time interval, so as to adapt to the dynamic changes of wind and light output and load demand, and reduce the energy balance problem caused by prediction error;

[0112] Based on real-time monitoring data and an aging compensation control mechanism, the rolling horizon optimization (Model Predictive Control, MPC) method is used to dynamically modify the day-ahead dispatch plan. The optimal energy storage charging and discharging strategy for the future is recalculated at set intervals to adapt to the dynamic changes in wind and solar power output and load demand. The specific steps to reduce energy balance problems caused by prediction errors are as follows:

[0113] At the current moment, based on wind and solar power generation, grid load and energy storage system status, a rolling optimization objective function is established to minimize the aging loss of the energy storage system while ensuring the energy balance of the system. p The system predicts future wind and solar power generation and load demand, and combines the battery health factor to calculate the available power of the energy storage system and the optimal scheduling strategy. The optimization objectives are defined as follows:

[0114]

[0115] , where J is the optimization objective function, which measures the comprehensive cost of the energy storage scheduling strategy, t k is the current moment, that is, the time when the MPC calculation starts, T p It is the prediction time domain, that is, the future time range considered when optimizing the scheduling plan. p Between 1 hour and 24 hours, it depends on the accuracy of wind and solar forecasts and the response speed of the energy storage system. t is the optimization moment, which represents each discrete time point in the rolling optimization process. b (t) is the charge and discharge power of the energy storage system at time t, SOC(t) is the state of charge of the energy storage system at time t, SOC ref is the target SOC value, usually set between 50% and 70% to reduce the deep charge and discharge of the battery and improve its service life. HF(t) is the battery health factor, which is used to measure the health status of the battery. The range is usually 0-1: HF(t) = 1 indicates that the battery is in the best health state and the maximum charge and discharge power is allowed; HF(t) < 1 indicates that the battery is aged and the charge and discharge power needs to be reduced; when HF(t) ≈ 0, the battery may have been severely aged or damaged and the charge and discharge need to be minimized. λ1 is the energy storage power regulation weight, which controls the power fluctuation of the energy storage system. λ2 is the SOC deviation penalty weight, which controls the SOC to maintain the target value SOC. ref Nearby, λ3 is the weight of battery aging, which controls the charge and discharge power when HF is low;

[0116] The optimization objective function aims to minimize the fluctuation of energy storage power, reduce the deviation of SOC from the reference value, and dynamically adjust the charge and discharge power to adapt to batteries with different degrees of aging.

[0117] Based on the optimization objective function J, the constraint conditions are constructed, including SOC constraint, battery power constraint, energy balance equation, etc., to ensure that the optimized scheduling strategy is feasible and meets the system operation requirements. In each prediction time domain T p , the MPC adopts a rolling optimization strategy, i.e. at the current time t k , the optimal energy storage charging and discharging strategy is calculated, but only the charging and discharging instructions within the current optimization period are executed, and then the next optimization period is entered, the prediction data is updated, and the optimization problem is solved again. The energy balance equation is as follows:

[0118]

[0119] , where SOC(t+1) is the state of charge at the next time, SOC(t+1) is the state of charge of the energy storage system calculated at time t+1, represents the influence of the optimized scheduling strategy on the battery energy state, η c is the charging efficiency, which represents the energy conversion efficiency of the battery during charging, usually η c <1, η d is the discharging efficiency, which represents the energy conversion efficiency of the battery during discharging, usually η d <1, and are the optimal charging power and discharging power at time t, respectively, which are obtained by optimization of the objective function J, satisfying (only charging or discharging is allowed at the same time), C b is the rated capacity of the battery, which determines the SOC change rate.

[0120] Through MPC rolling optimization, the system can real-time correct prediction errors, improve the adaptability of scheduling, and optimize the battery health state while meeting the energy balance, prolonging the service life of the energy storage system.

[0121] By combining historical operation data and real-time scheduling feedback, using reinforcement learning algorithm (such as deep reinforcement learning DRL) to dynamically adjust the scheduling parameters, the energy storage system can adapt to the long-term changes of battery aging state, improve the utilization efficiency of energy storage, and optimize the scheduling strategy under different operating scenarios, thereby reducing cumulative errors in the long-term operation process, making the energy storage system intelligent and economical.

[0122] By combining historical operation data and real-time scheduling feedback, using reinforcement learning algorithm (such as deep reinforcement learning DRL) to dynamically adjust the scheduling parameters, the energy storage system can adapt to the long-term changes of battery aging state, improve the utilization efficiency of energy storage, and optimize the scheduling strategy under different operating scenarios, thereby reducing cumulative errors in the long-term operation process, making the energy storage system intelligent and economical. The specific steps are as follows:

[0123] Firstly, the state, action and reward function of reinforcement learning are defined, and the state transition equation is constructed to enable the model to learn the optimal scheduling strategy of the energy storage system under different operating environments.

[0124] State S t including the current battery health state SOH, the remaining capacity SOC, the battery temperature T b , the wind and solar power prediction value P wp , the grid load demand P load and the market electricity price C m Key information, defined as:

[0125] S t ={SOH t , SOC t , T b , P wp , P load , C m}

[0126] Action A t represents the scheduling decision of the energy storage system at the current time;

[0127] A t ={P ch , P dis , ΔSOC}

[0128] , wherein P ch is the charging power, P dis is the discharging power and ΔSOC is the SOC adjustment strategy;

[0129] The state transition equation is based on the battery dynamic aging model to calculate the SOH and temperature change at the next time:

[0130] SOH t+1 = SOH t - α·f(T b , P ch , P dis )

[0131] wherein SOH t is the battery health state at time t, SOH t+1 is the battery health state at the next time, α is the aging factor, f(T b , P ch , P dis ) represents the decay rate of the battery due to temperature, charging and discharging power, T b is the battery temperature, P ch is the charging power, and P dis is the discharging power;

[0132] The reward function is designed to optimize the scheduling strategy. A composite reward function is designed, including battery life optimization, power balance of the power grid, and wind and light consumption rate, and is defined as:

[0133]

[0134] wherein R t is the reward function, P curtail is the abandoned wind and light power, indicating the wind and light power that cannot be consumed due to insufficient energy storage or load, β1 is the battery life optimization weight, measuring the importance of the change in battery state of health (SOH) in the reward function, β2 is the wind and light consumption optimization weight, measuring the influence of wind and light power utilization rate on the scheduling strategy, and β3 is the economic cost optimization weight, measuring the role of market electricity price and charging and discharging cost in the optimization target.

[0135] Through this reward function, the reinforcement learning system can learn how to adjust the charging and discharging strategy in different environments to achieve life optimization, economic optimization, and wind and light consumption maximization.

[0136] After determining the state S t , the state A t , and the reward function R t , a deep reinforcement learning (Deep Deterministid Policy Gradient, DDPG) based on Actor-Critic architecture is used for policy training, so that the energy storage system can adapt to the battery aging state and optimize the scheduling strategy in different scenarios.

[0137] The Actor network (policy network) is responsible for generating the state A t from the state S t , and the formula is as follows:

[0138] A t =π θ (S t )+N t

[0139] wherein π θ (S t ) is the parameterized policy, and N t is the exploration noise (such as Ornstein-Uhlenbeck noise) for balancing exploration and utilization.

[0140] The Critic network (value network) is used to evaluate whether the policy of the Actor network is optimal, and calculates the Q value, and the calculation formula is as follows:

[0141] Q(S t , A t )=R t +γQ(St+1 , A t+1 )

[0142] where Q(S t , A t ) is the state-action value function, i.e., the Q-value, representing the expected value of future cumulative reward after taking action A t in state S t , the Critic network is used to estimate this value and guide the Actor network to update the policy so as to maximize the future return, γ is the discount factor used to control the importance of future rewards, Q(S t+1 , A t+1 ) is the Q-value estimate at the next time t+1, representing the long-term cumulative return of the system after taking a new action A t+1 in a new state S t+1 ;

[0143] Experience Replay (Replay Buffer) To improve the stability of training, the past state-action-reward-state transition data is stored in the experience replay pool, and is randomly sampled from it for training, avoiding the problem of gradient shock caused by data correlation;

[0144] Policy Update, the Actor and Critic networks are optimized using gradient descent to update the policy parameters θ to maximize the cumulative reward, and the update formula is as follows:

[0145]

[0146] where θ is the training parameter of the Actor network, including the weights and bias terms of the neural network, η is the learning rate, is the parameter update direction based on the policy gradient.

[0147] Finally, through reinforcement learning training, the system can dynamically adjust the scheduling parameters according to different operating scenarios, so that the energy storage system can adapt to the battery aging state, improve the utilization efficiency of the energy storage system, and optimize the long-term economy of the wind-solar-storage system.

[0148] Specific implementation 1: In the economic dispatching process of wind-solar-storage systems, accurate prediction of wind-solar output and reasonable arrangement of charging and discharging strategies of the energy storage system are important means to optimize energy utilization efficiency and improve economic efficiency. However, due to the uncertainty of wind-solar power generation, the dynamic changes of power grid load and the aging characteristics of the energy storage system itself, the traditional fixed rule dispatching method often cannot adapt to the complex operating environment. Therefore, this embodiment proposes an intelligent dispatching system based on deep reinforcement learning (Deep Reinforcement Learning, DRL), which can combine historical operation data and real-time dispatching feedback, autonomously learn the optimal charging and discharging strategy under different operating scenarios, and dynamically adjust the dispatching parameters of the energy storage system to optimize the overall operating effect of the wind-solar-storage system.

[0149] In the day-ahead dispatching stage, the system first obtains wind-solar power generation prediction data, power grid load prediction data and health status information of the energy storage system, and uses a multi-objective optimization method to calculate the optimal energy storage charging and discharging plan. The optimization objectives include maximizing wind-solar consumption, minimizing battery aging rate, reducing power grid purchase cost, etc., to ensure a balance between economy and equipment life. The calculated dispatching scheme will be used as a reference scheme for the intra-day dispatching stage.

[0150] In the intra-day dispatching stage, the system monitors wind-solar output, power grid load changes, energy storage SOC (state of charge), SOH (state of health) and battery temperature in real time through high-precision sensors, and uses state estimation technology (such as Kalman filtering) to process the collected data to eliminate noise and outliers and improve data accuracy. When the actual state of the energy storage system deviates significantly from the day-ahead dispatching plan, the system will trigger a dynamic adjustment mechanism to recalculate the energy storage charging and discharging strategy for a certain period of time in the future to reduce dispatching errors. For example, when the actual wind-solar output is much lower than the predicted value, the system can actively supplement charging during low electricity price periods to avoid future power shortages, and when the wind-solar output is higher than the predicted value, the system can increase the discharging power to fully utilize the excess clean energy.

[0151] To enable the system to continuously optimize the scheduling strategy, deep reinforcement learning (DRL) is introduced into the scheduling process, enabling the system to continuously learn the best charging and discharging decisions from historical data and real-time feedback. DRL adopts a state-action-reward mechanism, and whenever the system executes a certain scheduling strategy, it calculates the benefits of the current decision based on key indicators such as battery life changes, wind and light consumption ratios, and power costs, and optimizes future scheduling decisions based on long-term benefits. For example, if a certain charging and discharging strategy causes the battery SOH to rapidly decline in a short period of time, the system will reduce the selection of similar scheduling decisions in subsequent training to optimize battery life. As training continues, the system will form a set of adaptive intelligent scheduling strategies that can dynamically adjust charging and discharging power and SOC management strategies to maximize system operating efficiency, even under different environmental conditions.

[0152] Ultimately, the system not only improves the economic efficiency and safety of the wind-solar-storage system, but also significantly reduces scheduling deviations caused by prediction errors, making the energy storage system more adaptive and long-term optimized.

[0153] Specific implementation 2: During long-term operation, the aging effect of energy storage batteries is a key issue affecting scheduling accuracy and system life. Traditional scheduling methods are usually based on fixed charging and discharging strategies, ignoring the dynamic changes in battery health status, which can cause some batteries to age prematurely, affecting the overall efficiency of the energy storage system. To solve this problem, this implementation proposes a Health Factor (HF) based energy storage charging and discharging optimization method that monitors battery health status in real time and dynamically adjusts the load distribution of different battery groups to balance the overall degradation of the energy storage system and improve long-term stability of the equipment.

[0154] During system operation, the energy storage management system (EMS) monitors key parameters such as SOC, SOH, battery temperature, and charging and discharging rate of each battery group in real time, and calculates the health factor (HF) of each battery based on the energy storage dynamic aging model. The HF value is usually set between 0 and 1, with a higher value indicating a better battery health status and a higher ability to bear more charging and discharging loads, while a lower HF value indicates that the battery should reduce the frequency of deep charging and discharging to reduce the aging rate.

[0155] In the scheduling decision process, the system optimally allocates the battery load based on the HF value, ensuring that the battery with better health status bears more power, and the battery with poor health status reduces the load. For example, the system can dynamically adjust the SOC working window, and for the battery pack with low SOH, the SOC working range is narrowed (such as 30%-70%), the high SOC residence time is reduced to reduce the lithium deposition phenomenon, and for the battery with high SOH, a larger SOC range (such as 20%-90%) is allowed to improve the energy utilization. In addition, in a large-scale energy storage system, different battery packs may use different charging and discharging strategies, and the system can manage the battery packs according to the HF value, and preferentially use the battery pack with better health status to balance the battery aging rate and improve the overall life of the system.

[0156] Through this method, the uneven aging phenomenon of energy storage batteries can be effectively reduced, the risk of premature retirement of individual battery packs is reduced, and the scheduling flexibility and long-term operation stability of the entire wind-solar-storage system are improved. DETAILED DESCRIPTION 3

[0158] In the actual operation of the wind-solar-storage system, due to the continuous change of wind-solar output, grid load, market electricity price and other factors, a fixed scheduling strategy often cannot adapt to the complex dynamic environment. Therefore, in order to ensure the long-term optimal operation of the energy storage system, this embodiment proposes a method combining rolling optimization scheduling and experience replay mechanism, which updates real-time data and learns from historical experience to continuously optimize the scheduling strategy and improve the operation efficiency of the wind-solar-storage system.

[0159] In the scheduling process, the system uses the rolling time domain optimization (MPC) method to recalculate the optimal charging and discharging strategy for a certain period of time (such as 15 minutes) to ensure that the scheduling scheme always matches the actual system state. For example, in a short-term power market environment, the system can adjust the charging and discharging strategy according to the real-time market electricity price, charge during the period when the electricity price is low, and discharge during the period when the electricity price is high, to improve the economy. At the same time, in order to improve the long-term stability of the scheduling strategy, the system introduces the experience replay mechanism (ReplayBuffer) to store the state-action-reward data in the past operation process, and refers to the historical best scheduling strategy in future decision-making. For example, in extreme weather (such as continuous rainy days or high temperature weather), the system can call the best scheduling scheme under similar environment in the past, and adjust the current charging and discharging plan to make it more suitable for the current environmental change.

[0160] Finally, this scheme can improve the utilization efficiency of the energy storage system while reducing the scheduling error, and make the energy storage system realize the optimal economic scheduling strategy under different operating environments.

[0161] The application can effectively reduce the uneven aging phenomenon of energy storage batteries and improve the long-term operation stability of the entire energy storage system through the health factor (HF) driven energy storage charging and discharging optimization method. In the traditional scheduling method, all battery groups are often operated according to the same charging and discharging strategy, which cannot be dynamically adjusted according to the health status of individual batteries, resulting in accelerated aging of some batteries due to high-rate discharging, long-term high SOC operation or temperature abnormalities, ultimately affecting the reliability of the entire system. The present application monitors SOH, SOC, temperature, current and other parameters in real time, calculates the battery health factor (HF), and intelligently adjusts the charging and discharging task allocation according to the HF value, ensuring that the batteries with good health status bear more power, while the batteries with severe aging reduce deep discharge and high-rate operation. This dynamic optimization strategy can prolong the overall life of the energy storage system, reduce scheduling errors caused by individual battery degradation, and improve system stability. In addition, combined with deep reinforcement learning (DRL) technology, the system can continuously learn and optimize the SOC management strategy in long-term operation, such as dynamically adjusting the SOC window and optimizing the charging and discharging rate, to minimize the battery aging rate. This not only reduces the maintenance and replacement cost of energy storage batteries, but also improves the long-term investment return rate of the energy storage system, making the wind-solar-storage system more economically feasible and operationally reliable.

[0162] The application combines rolling optimization scheduling (MPC) with experience replay mechanism (ReplayBuffer) to enable the system to dynamically adjust the charging and discharging strategy according to real-time load demand, market electricity price, wind and light prediction error and other factors, thereby improving wind and light consumption rate and optimizing grid energy balance. Specifically, in the intra-day scheduling stage, the system can perform rolling optimization calculation based on the latest load and wind and light output data every 15-30 minutes to ensure that the scheduling scheme always matches the actual operating state and reduces scheduling deviation caused by prediction error. In addition, through reinforcement learning technology, the system can continuously optimize the scheduling strategy under different weather conditions and load characteristics during long-term operation to improve the system's adaptability to extreme weather or sudden load fluctuations. For example, when encountering continuous low wind and low light weather, the system can use historical experience data to automatically switch to a more optimal energy storage management strategy, such as early charging, reducing discharging power, optimizing charging timing, etc., to ensure the stable operation of the grid. Ultimately, this optimization scheme can improve the utilization rate of wind and light generation while reducing the dependence on standby thermal power units, reducing grid scheduling costs, and improving the penetration rate of new energy in the power system, promoting the development of the power system towards a more green and low-carbon direction.

[0163] The above formulas are dimensionless values calculated, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the latest real situation. The preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0164] It is apparent that for the person having ordinary skill in the art, many changes and modifications can be made to the embodiments described above without departing from the spirit and the scope of the application. Thus, the foregoing description is by way of example only, and is not intended to be limiting. The following drawings and description are included to provide a more complete understanding of the present application.

[0165] It should be noted that, as used in this document, the terms "first", "second", etc. are used only to distinguish one entity or action from another, and do not necessarily require or imply any actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0166] It should be understood that the size of the sequence number of the processes described above in various embodiments of the present application does not mean the order of execution, and the execution order of the processes should be determined according to their functions and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0167] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0168] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0169] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e. they can be located in one place or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the present embodiment according to actual needs.

[0170] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0171] 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.

[0172] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A method for day-ahead economic dispatch of a wind, solar, and energy storage system, characterized by: The following steps are involved: Based on the battery's historical operating data, a dynamic aging model is established. Through machine learning, the actual attenuation characteristics of the battery are fitted to obtain the battery health status under different operating conditions and its change pattern over time, forming the basis for dynamic adjustment strategies. During the day-ahead scheduling phase, a multi-objective optimization method is used to calculate the optimal charging and discharging plan for energy storage, combining wind and solar power generation forecast data, grid load forecast data, and the health status of the energy storage system. This ensures that the impact of high-rate charging and discharging on battery life is minimized while maximizing wind and solar power consumption. This also provides a benchmark solution for intraday scheduling. During the intraday dispatch process, high-precision sensors and state estimation methods are used to obtain the actual output of wind and solar power generation, the real-time demand of the grid load, and the actual available capacity of the energy storage system. These are then compared with the day-ahead dispatch plan to identify any deviations and determine whether the energy storage charging and discharging plan needs to be adjusted. Based on the battery health parameters calculated by the dynamic aging model of energy storage, the scheduling strategy in actual operation is adjusted. By introducing health factors to modify the charge and discharge plan, the battery is prevented from overcharging and overdischarging. At the same time, the load distribution of different battery groups is optimized to achieve balanced attenuation of the overall energy storage system, extend the service life, and reduce scheduling deviations caused by uncertainty. Based on real-time monitoring data and an aging compensation control mechanism, a rolling time domain optimization method is used to dynamically modify the day-ahead dispatch plan. The optimal energy storage charging and discharging strategy for the future is recalculated at set time intervals to adapt to the dynamic changes in wind and solar power output and load demand, thereby reducing energy balance issues caused by forecast errors. The specific steps are as follows: At the current moment, based on wind and solar power generation, grid load and energy storage system status, a rolling optimization objective function is established to minimize the aging loss of the energy storage system while ensuring the energy balance of the system. The system predicts future wind and solar power generation and load demand, and combines the battery health factor to calculate the available power of the energy storage system and the optimal scheduling strategy. The optimization objectives are defined as follows: , where is the optimization objective function, which measures the comprehensive cost of the energy storage scheduling strategy. It is the present moment. is the prediction time domain, Is the optimization moment, which represents each discrete time point in the rolling optimization process. The energy storage system is at the moment The charge and discharge power, The energy storage system is at the moment The state of charge, is the target value, Is the battery health factor, used to measure the health of the battery, is the energy storage power regulation weight, which controls the power fluctuation of the energy storage system. yes Deviation penalty weight, is the weight of battery aging; Combining historical operating data and real-time scheduling feedback, the reinforcement learning algorithm is used to dynamically adjust the scheduling parameters, so that the energy storage system can adapt to the long-term changes in the battery aging state, improve the energy storage utilization efficiency, and optimize the scheduling strategy under different operating scenarios, thereby reducing the cumulative error in the long-term operation process and making the energy storage system intelligent and economical.

2. The method for day-ahead economic dispatch of a wind, solar, and energy storage system according to claim 1 is characterized in that Based on the historical operating data of the battery, a dynamic aging model is established. The actual attenuation characteristics of the battery are fitted through machine learning to obtain the battery health status under different operating conditions and its change pattern over time, so as to form the basis of the dynamic adjustment strategy. The specific steps are as follows: Build a battery operation database, collect and clean key data such as charge and discharge current, voltage, and temperature, remove outliers, and standardize the data to ensure data integrity and consistency; Screen key characteristic variables that affect battery aging, use correlation analysis to optimize variable selection, and improve modeling accuracy; Training with machine learning algorithms Predictive model, optimize hyperparameters and evaluate prediction accuracy to ensure the model accurately reflects changes in battery health status; The SOH prediction model is integrated into the energy storage management system, and a rolling update mechanism is used for continuous optimization. The charging and discharging strategies are adjusted in combination with the optimization algorithm to extend battery life and improve system stability.

3. The method for day-ahead economic dispatch of a wind, solar and energy storage system according to claim 1 is characterized in that During the day-ahead scheduling phase, a multi-objective optimization method is used to calculate the optimal charging and discharging plan for energy storage, ensuring that the impact of high-rate charging and discharging on battery life is reduced while maximizing wind and solar power consumption. The specific steps for providing a benchmark solution for intraday scheduling are as follows: Integrate wind and solar power generation forecasts, grid load forecasts, and energy storage health data, and use uncertainty modeling methods to reduce the impact of forecast errors on scheduling decisions; Construct a multi-objective optimization model that takes into account wind and solar power consumption, battery aging, and economic costs, and use the Pareto optimization algorithm to dynamically balance different optimization objectives; Based on mixed integer linear programming, the energy storage charging and discharging time, power and path are optimized, and the MPC rolling optimization method is combined to improve scheduling adaptability; Formulate dynamic The benchmark scheduling scheme of window adjustment and intelligent balanced charging and discharging strategy is combined with the rolling optimization algorithm to continuously revise and optimize during the intraday scheduling stage.

4. The method for day-ahead economic dispatch of a wind, solar, and energy storage system according to claim 1 is characterized in that During the intraday dispatch process, high-precision sensors and state estimation methods are used to obtain the actual output of wind and solar power generation, the real-time demand of the grid load, and the actual available capacity of the energy storage system. These are then compared with the day-ahead dispatch plan to identify any deviations and determine whether the energy storage charging and discharging plan needs to be adjusted. The specific steps are as follows: Use high-precision sensors and state estimation methods to obtain wind and solar power output, load demand, and energy storage system status, and improve measurement accuracy through data filtering and machine learning methods; Compare actual operating status with the day-ahead dispatch plan, calculate key deviation indicators of wind and solar power output, load demand, and energy storage systems, and assess the impact of deviations on system operation; Recalculate future charging and discharging strategies based on the MPC rolling optimization method, optimize the SOC window, and combine grid scheduling and market electricity prices to improve economic efficiency and energy storage system life; Execute the optimized scheduling plan, provide real-time feedback to control the energy storage operating status, and use reinforcement learning technology to optimize long-term scheduling strategies to improve the intelligence level of the system.

5. The method for day-ahead economic dispatch of a wind, solar and energy storage system according to claim 1 is characterized in that Based on the battery health status parameters calculated by the energy storage dynamic aging model, the scheduling strategy in actual operation is adjusted. By introducing the health factor to correct the charge and discharge plan, the battery is prevented from overcharging and discharging. At the same time, the load distribution of different battery groups is optimized to make the overall attenuation of the energy storage system balanced, extend the service life and reduce the scheduling deviation caused by uncertainty. The specific steps are as follows: Battery-based 、 , the key factors of the number of cycles are calculated by weighted comprehensive evaluation method Value, measure the battery health status and provide a basis for scheduling optimization; Optimize the battery pack's charge and discharge task allocation according to the HF value, adjust Working window, ensuring that batteries in good health bear more load and achieve balanced aging of the energy storage system as a whole; A day-ahead and day-intraday rolling optimization mechanism is used to dynamically calculate future dispatch strategies at equal intervals. Combined with an adaptive control algorithm, this makes energy storage dispatch more flexible and economical. Establish a closed-loop monitoring mechanism and use reinforcement learning methods to optimize long-term scheduling strategies, so that the energy storage system can be continuously optimized under different operating environments, extending battery life and improving overall operating efficiency.

6. The method for day-ahead economic dispatch of a wind, solar and energy storage system according to claim 1, characterized in that: Based on the optimization objective function , build constraints to ensure that the optimal scheduling strategy is feasible and meets the system operation requirements, in each prediction time domain , adopting the rolling optimization strategy, that is, at the current moment Calculate the optimal energy storage charging and discharging strategy, but only execute the charging and discharging instructions within the current optimization cycle. Then enter the next optimization cycle, update the prediction data, and re-solve the optimization problem. The energy balance equation is as follows: , where is the state of charge at the next moment, is the charging efficiency, is the discharge efficiency, and Separate moments The optimal charging power and discharging power are given by the objective function The optimization solution is: Is the rated capacity of the battery, which determines Rate of change.

7. The method for day-ahead economic dispatch of a wind, solar, and energy storage system according to claim 1 is characterized in that By combining historical operating data with real-time scheduling feedback, and using reinforcement learning algorithms to dynamically adjust scheduling parameters, the energy storage system can adapt to long-term changes in battery aging, improve energy storage utilization efficiency, and optimize scheduling strategies under different operating scenarios. This reduces cumulative errors during long-term operation, making the energy storage system intelligent and economical. The specific steps are as follows: First, the state, action, and reward functions of reinforcement learning are defined, and the state transition equation is constructed so that the model can learn the optimal scheduling strategy for the energy storage system under different operating environments. state Including the current battery health status , remaining power , battery temperature , wind and solar power generation forecast value , grid load demand and market electricity prices Key information, defined as: action Represents the dispatch decision of the energy storage system at the current moment; ,in, is the charging power, is the discharge power and yes Adjust strategies; The state transfer equation is based on the battery dynamic aging model and calculates the next moment And temperature changes: in, It's time Battery health status, is the battery health status at the next moment, It is the aging factor, Indicates the battery attenuation rate caused by temperature and charge and discharge power. is the battery temperature, is the charging power, is the discharge power; The reward function is designed to optimize the scheduling strategy and design a composite reward function, which is defined as: ,in, is the reward function, is the abandoned wind and solar power, is the battery life optimization weight, is the optimization weight of wind and solar power consumption, It is the economic cost optimization weight, which measures the role of market electricity price and charging and discharging cost in the optimization target.

8. The method for day-ahead economic dispatch of a wind, solar, and energy storage system according to claim 7 is characterized in that , in the determined state ,state and the reward function Afterwards, based on The architecture's deep reinforcement learning performs strategy training, enabling the energy storage system to adapt to battery aging conditions and optimize scheduling strategies in different scenarios. The network is responsible for Build Status , the formula is as follows: in, is a parameterized strategy, is exploration noise, used to balance exploration and exploitation; The network is used to evaluate Is the network strategy optimal? The value is calculated as follows: in, is the state-action value function, that is value, is a discount factor that controls the importance of future rewards, The next moment of Value estimation; To improve training stability, experience replay stores past state-action-reward-state transition data in an experience replay pool and randomly samples it for training to avoid gradient oscillation caused by data correlation. Strategy update, using gradient descent optimization and Network, update policy parameters To maximize the cumulative reward, the update formula is as follows: , where yes The training parameters of the network, is the learning rate, It is the parameter update direction based on policy gradient.

Images