Optimal scheduling method for hybrid energy storage system based on AI intelligent control

Abstract

The present invention discloses a hybrid energy storage system optimization and scheduling method based on AI intelligent control, which relates to the field of hybrid energy storage technology and includes the following steps: collecting real-time data and cross-validating data consistency through multi-source data fusion technology; using adaptive threshold detection combined with statistical analysis methods to identify data anomalies caused by sensor failures, communication delays, and environmental interference, and eliminating abnormal data. The present invention optimizes energy storage scheduling through data cleaning, time series prediction, and reinforcement learning, improving intelligent management and dynamic adaptability. Multi-source data fusion and anomaly detection are used to ensure data accuracy, and energy consumption is predicted with the help of LSTM and Transformer to optimize energy storage strategies in advance. Combined with reinforcement learning, the system dynamically adjusts charging and discharging, improves the photovoltaic absorption rate, reduces electricity purchase costs, and intelligently controls the SOC within the range of 40%-80%, extending battery life. At the same time, anomaly detection and correction improve system stability and reduce maintenance costs.

Description

Technical Field

[0001] The present invention relates to the field of hybrid energy storage technology, and in particular to a hybrid energy storage system optimization scheduling method based on AI intelligent control. Background Art

[0002] Optimal scheduling of hybrid energy storage systems based on AI intelligent control refers to the use of artificial intelligence technology to intelligently optimize and manage hybrid energy storage systems (typically including battery energy storage and supercapacitors) in photovoltaic and energy storage base stations to improve energy efficiency, reduce operating costs, and enhance system stability. Through real-time analysis of photovoltaic power generation, load demand, grid electricity prices, and energy storage device status through AI algorithms (such as machine learning and deep reinforcement learning), intelligent decisions are made on when to store and discharge energy, as well as how to reasonably allocate energy among different energy storage media, thereby achieving optimal energy scheduling. This technology can effectively smooth out the volatility of photovoltaic power generation, reduce abandoned solar power, improve the reliability of base station power supply, and reduce electricity costs during peak and valley electricity price scheduling.

[0003] The existing technology has the following shortcomings: In the process of optimizing the scheduling of hybrid energy storage systems based on AI intelligent control, the failure of scheduling strategies caused by data noise and outliers may have serious consequences. The energy storage system relies on sensors to collect real-time data such as photovoltaic power generation, load demand, and battery status. If sensor failure, communication delays, or environmental interference cause data anomalies, the AI scheduling algorithm may misidentify the system status and then execute the wrong energy allocation strategy. For example, misjudging the remaining battery capacity may lead to overcharging or over-discharging, affecting battery life and even causing safety accidents. In addition, if the AI model does not cover enough abnormal situations during training, the scheduling system may lose robustness in the face of sudden anomalies, resulting in unstable power supply to the base station and affecting the normal operation of the communication network.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a hybrid energy storage system optimization scheduling method based on AI intelligent control, which optimizes energy storage scheduling through data cleaning, time series prediction and reinforcement learning, so that the system has intelligent management and dynamic adaptability. Multi-source data fusion and anomaly detection technology are used to ensure the accuracy of data such as photovoltaic power generation, load demand, and battery SOC, and with the help of LSTM and Transformer, future energy consumption is predicted, and energy storage strategies are optimized in advance to prevent unstable power supply. Combined with reinforcement learning, the system can dynamically adjust the charging and discharging plan, improve the photovoltaic absorption rate, reduce the grid power purchase cost, and intelligently control the SOC within the range of 40%-80%, thereby extending battery life. At the same time, the abnormal scene detection and correction mechanism can cope with extreme load fluctuations, battery aging and abnormal ambient temperature, improve system stability and reduce long-term maintenance costs, so as to solve the problems in the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a hybrid energy storage system optimization scheduling method based on AI intelligent control, comprising the following steps:

[0007] Collect real-time data and cross-validate data consistency through multi-source data fusion technology; use adaptive threshold detection combined with statistical analysis methods to identify data anomalies caused by sensor failures, communication delays, and environmental interference, and eliminate abnormal data;

[0008] A fusion algorithm based on historical data regression analysis and physical modeling is used to correct marked abnormal data and interpolate and compensate using data from adjacent time windows to reduce the impact of missing data on scheduling decisions.

[0009] Use time series forecasting models to predict future photovoltaic power generation, load demand, and energy storage status, establish a dispatch warning mechanism, and adjust energy allocation strategies in advance when sudden anomalies are predicted to reduce the impact of anomalies on the stability of the energy storage system;

[0010] Combined with reinforcement learning algorithms, a reward function is constructed to optimize the energy scheduling strategy based on a distributed reinforcement learning framework, enabling the energy storage system to achieve dynamic optimal scheduling under different operating conditions;

[0011] During the AI scheduling process, confidence estimation and Bayesian optimization methods are combined to dynamically monitor the reliability of scheduling decisions. When the confidence level falls below the set threshold, the expert system-based rule correction mechanism is triggered to make reasonable adjustments to the scheduling results to ensure robustness under abnormal circumstances.

[0012] A federated learning architecture is adopted to enable multiple base stations to share abnormal data patterns and collaboratively optimize AI models while protecting data privacy. The training set is dynamically expanded through an active learning mechanism to improve the model's adaptability to data noise and outliers, thereby enhancing the robustness of the overall scheduling system.

[0013] Preferably, real-time data is collected and cross-validated for consistency using multi-source data fusion technology; adaptive threshold detection combined with statistical analysis methods is used to identify data anomalies caused by sensor failures, communication delays, and environmental interference, and to eliminate abnormal data. The specific steps are as follows:

[0014] Collect key data of photovoltaics, batteries, and loads, and perform time alignment, format standardization, and interpolation processing to ensure data consistency and availability;

[0015] Adopt multi-source data fusion technology to cross-validate data from different sensors to improve data accuracy;

[0016] Combining adaptive threshold detection, statistical analysis, and time series prediction methods to dynamically identify data anomalies and improve the sensitivity and adaptability of anomaly detection;

[0017] Abnormal data is filled through sliding window filtering, regression analysis and deep learning prediction, and the compensation strategy is optimized in combination with physical constraints to ensure the stable operation of the energy storage system.

[0018] Preferably, a fusion algorithm based on historical data regression analysis and physical modeling is used to correct the marked abnormal data and interpolate and compensate using adjacent time window data to reduce the impact of data missing on scheduling decisions. The specific steps are as follows:

[0019] Classify abnormal data, analyze its impact on scheduling strategies, and select the optimal compensation strategy to reduce the negative impact of data loss;

[0020] Use regression models to predict and correct abnormal data to align with historical trends and improve data accuracy and reliability;

[0021] Physical modeling is used to constrain the rationality of regression correction data to ensure that the corrected data conforms to physical laws and avoids erroneous compensation affecting system operation.

[0022] Linear interpolation, spline interpolation and Kalman filtering methods are used to fill short-term missing data, and combined with physical constraint optimization compensation to ensure data continuity and scheduling stability.

[0023] Preferably, a time series forecasting model is used to predict photovoltaic power generation, load demand, and energy storage status in the future, and a scheduling warning mechanism is established to adjust the energy allocation strategy in advance when sudden anomalies are predicted. The specific steps to reduce the impact of anomalies on the stability of the energy storage system are as follows:

[0024] Organize and normalize PV power generation, load demand, and energy storage status data, using time window technology to capture long-term and short-term trends to ensure stable data quality;

[0025] Select a time series forecasting model for training, optimize hyperparameters, and perform cross-validation to improve the forecast accuracy of photovoltaic power generation, load demand, and energy storage status;

[0026] Based on the prediction results, early warning thresholds are set to dynamically adjust energy distribution before photovoltaic power generation decreases or load surges, and reinforcement learning is combined to optimize the scheduling strategy;

[0027] The energy storage scheduling strategy is optimized before an anomaly occurs, and the model adaptive adjustment mechanism is used to correct the prediction error, thereby improving the stability and predictive ability of the energy storage system.

[0028] Preferably, the specific steps of combining reinforcement learning algorithms, constructing reward functions, and optimizing energy scheduling strategies based on a distributed reinforcement learning framework to achieve dynamic optimal scheduling of energy storage systems under different working conditions are as follows;

[0029] Combined with the time series prediction results, the state space of reinforcement learning is defined to build a dynamic scheduling environment;

[0030] A reward function is designed based on economic efficiency, battery life, load stability, and renewable energy utilization to guide the agent to learn the optimal energy scheduling strategy and dynamically adjust the weights to adapt to different application scenarios.

[0031] Adopt reinforcement learning algorithms such as PPO / DDPG and use a distributed reinforcement learning framework to accelerate training, enabling the intelligent agent to learn the optimal energy scheduling strategy under different working conditions, improving learning efficiency and convergence speed;

[0032] Through online learning and model adaptive adjustment mechanisms, the reinforcement learning agent can continuously optimize the scheduling strategy and dynamically adapt to environmental changes and sudden anomalies, thereby improving the stability and explainability of the energy storage system.

[0033] Preferably, during the AI scheduling execution process, confidence estimation and Bayesian optimization methods are combined to dynamically monitor the reliability of scheduling decisions. When the confidence level is lower than the set threshold, the rule correction mechanism based on the expert system is triggered to reasonably adjust the scheduling results. The specific steps to ensure robustness under abnormal circumstances are as follows:

[0034] During the execution of AI scheduling, the confidence of the current scheduling strategy is first evaluated to determine the reliability of the decision. The mean and variance of the scheduling strategy are calculated through multiple forward propagations to quantify the uncertainty of the scheduling strategy. The formula is as follows:

[0035]

[0036] , where C t is the confidence of the scheduling decision, σ[P t ] is the uncertainty of the current scheduling decision, E[Pt ] is the expected value of the scheduling decision at the current moment, ∈ is a very small positive number to prevent the denominator from approaching zero;

[0037] When the confidence level C t Below the set threshold C thresh When , the rule correction mechanism is triggered to ensure the stability of the scheduling strategy;

[0038] When the confidence level is lower than the threshold C thresh When the current scheduling parameters are optimized to find the optimal scheduling value, the scheduling strategy is dynamically adjusted using Bayesian optimization, so that the optimized decision reduces the risk while improving the economy and stability of the energy storage system. The objective function of Bayesian optimization is defined as follows:

[0039]

[0040] , where is the optimized battery charge and discharge power, R(P) is the reward function, which takes into account factors such as grid electricity price, battery life, and load fluctuation, and λ is the risk weight coefficient, which controls the degree of penalty for uncertainty. is the value of P that maximizes the objective function.

[0041] Preferably, although Bayesian optimization provides optimized scheduling parameters However, it may still lead to unacceptable scheduling decisions in extreme cases. Therefore, a rule correction mechanism based on expert system is adopted to correct the Perform final corrections to ensure the safety and feasibility of the scheduling strategy. The rule correction formula is expressed as follows:

[0042]

[0043] , where is the final battery charge and discharge power, I(·) is the indicator function, which takes 1 if the condition is met and 0 otherwise, α is the smoothing weight factor, ΔP smooth It is a smooth adjustment value based on load fluctuations.

[0044] Preferably, a federated learning architecture is adopted to enable multiple base stations to share abnormal data patterns and collaboratively optimize the AI model while protecting data privacy. The specific steps for dynamically expanding the training set through an active learning mechanism to improve the model's adaptability to data noise and outliers and enhance the robustness of the overall scheduling system are as follows:

[0045] During the optimization and scheduling of hybrid energy storage systems, the data of a single base station is limited, and abnormal data patterns vary depending on factors such as geographic location, climate, and load characteristics. Therefore, to improve the generalization capability of the AI scheduling model, a federated learning architecture is adopted to enable multiple base stations to share abnormal data patterns while ensuring data privacy. The core calculation formula for federated model aggregation is as follows:

[0046]

[0047] , where θ (j+1) is the global model parameter of the j+1th round, N is the total number of base stations, w i is the data volume of base station i, E is the normalization factor of the total data volume of all base stations, is the model parameter of base station i in the jth round of training, γ is the weight coefficient of abnormal data on model update, M is the normalization factor of the total amount of abnormal data of all base stations, is the model gradient of base station i after local training;

[0048] Because the adaptability of the energy storage system scheduling AI model to abnormal data depends on the diversity and coverage of the training data, after optimizing the global model through federated learning, an active learning mechanism is further adopted to dynamically expand the training set and improve the model's robustness to noise and outliers. The sample selection for active learning is based on the principle of maximizing information gain, calculated as follows:

[0049]

[0050] , where S * is the selected active learning sample subset, S c Is the candidate data set, containing all samples to be screened, x represents the candidate sample set S c A data point in , p(y|x, θ (j+1) ) is the global model parameter θ (j+1) Under this condition, the predicted probability distribution of sample x is H(p(y|x,θ (j+1) )) is the prediction entropy of the current global model, K is the number of sub-models used to calculate uncertainty, represents the parameters of the k-th sub-model after j+1 rounds of training, The goal is to maximize information gain and select valuable samples.

[0051] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0052] Through data cleaning, time series prediction, and reinforcement learning-based scheduling optimization, this system enables intelligent management of energy storage systems in complex environments and improves their adaptability to dynamic changes. First, multi-source data fusion and anomaly detection technologies are employed to ensure the accuracy of key data such as photovoltaic power generation, load demand, and battery SOC, preventing issues such as sensor failure and communication delays from impacting scheduling decisions. Second, using time series prediction models such as LSTM and Transformer, the system can predict future photovoltaic power generation and load demand in advance and establish a scheduling warning mechanism to optimize energy storage strategies before anomalies occur, preventing sudden load shocks from causing unstable power supply. Furthermore, by combining reinforcement learning-based scheduling optimization strategies, the system can autonomously learn optimal scheduling solutions based on grid electricity prices, photovoltaic output, and load fluctuations. For example, it can reduce discharge during peak electricity prices to minimize battery loss while ensuring a stable power supply. Through intelligent optimization, the system can dynamically adjust its strategies under varying climate conditions, load patterns, and energy storage states, enhancing the adaptability of the energy storage system and improving the long-term stable operation of photovoltaic energy storage base stations.

[0053] Through intelligent control, the present invention enables the system to comprehensively consider factors such as grid electricity prices, photovoltaic power generation forecasts, and battery life, and dynamically adjust the charging and discharging plan, thereby reducing operating costs. For example, the system can prioritize charging during periods of low electricity prices and release stored energy during peak electricity prices, reducing the grid's electricity purchase expenditures while increasing the local absorption rate of photovoltaic power generation. In addition, to prevent battery aging due to frequent deep charging and discharging, the system optimizes scheduling through reinforcement learning, intelligently controls the battery SOC within the optimal operating range of 40%-80%, reduces overcharging and over-discharging, and improves battery health (SOH). At the same time, combined with abnormal scenario detection and intelligent correction mechanism, when extreme load fluctuations, battery aging, or abnormal ambient temperature are detected, the system can automatically adjust the scheduling strategy to prevent battery overheating or abnormal discharge, improve battery life, and reduce long-term operating and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0055] Figure 1 This is a flow chart of the method for optimizing the scheduling of hybrid energy storage systems based on AI intelligent control according to the present invention. DETAILED DESCRIPTION

[0056] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0057] The present invention provides Figure 1 The hybrid energy storage system optimization scheduling method based on AI intelligent control shown includes the following steps:

[0058] Collect real-time data such as photovoltaic power generation, load demand, and battery status, and cross-validate data consistency through multi-source data fusion technology; use adaptive threshold detection combined with statistical analysis methods to identify data anomalies caused by sensor failures, communication delays, and environmental interference, and eliminate abnormal data;

[0059] Collect real-time data such as photovoltaic power generation, load demand, and battery status, and cross-verify data consistency through multi-source data fusion technology. Use adaptive threshold detection combined with statistical analysis methods to identify data anomalies caused by sensor failures, communication delays, and environmental interference, and eliminate abnormal data. The specific steps are as follows:

[0060] Collect key data of photovoltaics, batteries, and loads, and perform time alignment, format standardization, and interpolation processing to ensure data consistency and availability;

[0061] In the process of optimizing and dispatching hybrid energy storage systems controlled by AI, key parameters such as photovoltaic power generation power, base station load demand, and battery state of charge (SOC) must first be collected. This process involves multiple sensors, such as photovoltaic current / voltage sensors, current / temperature sensors in the battery management system (BMS), smart meters, etc., and uploads data to the cloud or local edge computing devices through wireless or wired communication methods (such as LoRa, NB-IoT, Modbus). In order to ensure data integrity, the system needs to pre-process the collected raw data, including timestamp alignment, unit conversion, data format standardization, etc. For example, if the data refresh rates of some sensors are different, the system needs to perform interpolation processing to make all data comparable at the same point in time.

[0062] Adopt multi-source data fusion technology to cross-validate data from different sensors, and use redundant measurement, Kalman filtering and other methods to improve data accuracy;

[0063] Since energy storage systems involve multiple independent data sources, multi-source data fusion technology is needed to cross-validate data from different sensors to ensure data consistency. For example, redundant sensor data can be used for comparison, such as simultaneously obtaining the power generation data of the photovoltaic MPPT controller and the electricity meter, and calculating whether the error range of the two is within the set threshold. For different measurement methods of the same variable (such as battery SOC, which can be calculated by the BMS or estimated by the Coulomb integral method), Kalman filtering or Bayesian estimation methods can be used for fusion to obtain a more accurate state estimate. In addition, during the data fusion process, historical trend analysis can be introduced to detect whether the current data conforms to a reasonable change pattern, thereby improving data quality.

[0064] Combining adaptive threshold detection, statistical analysis, and time series prediction methods to dynamically identify data anomalies and improve the sensitivity and adaptability of anomaly detection;

[0065] After multi-source data is integrated, the system needs to use adaptive threshold detection combined with statistical analysis methods to identify and mark data anomalies. The adaptive threshold detection method can dynamically adjust the threshold based on the mean and standard deviation of the sliding window statistics to adapt to the data characteristics in different environments. For example, under normal operating conditions, the fluctuation range of battery voltage is small, while in extreme weather (such as high or low temperatures), battery performance may experience temporary anomalies. Therefore, the system can adjust the anomaly detection threshold according to the real-time environment. In addition, box plot analysis (IQR method) or Z-score method are used to calculate outliers. If a data point deviates from the mean by more than a certain standard deviation, it can be determined to be anomaly data. At the same time, time series analysis techniques (such as ARIMA or LSTM) are used to predict the normal data range and compare it with the actual data to further enhance the accuracy of anomaly identification.

[0066] Abnormal data is filled through sliding window filtering, regression analysis, and deep learning prediction, and compensation strategies are optimized in combination with physical constraints to ensure stable operation of the energy storage system.

[0067] For the detected abnormal data, the system needs to adopt appropriate elimination and compensation strategies. First, for short-term anomalies (such as voltage mutations caused by instantaneous sensor jitter), the sliding window mean filter method can be used for smoothing; for long-term anomalies (such as persistent data anomalies caused by sensor failure), the data point needs to be eliminated and intelligent compensation is performed using historical data. Intelligent compensation can combine physical modeling with machine learning methods, such as filling missing values through regression models or deep learning (such as Transformer prediction), while ensuring that the compensated data meets physical constraints (such as battery SOC cannot exceed 100%). In addition, the system can send alarms to operation and maintenance personnel when an anomaly occurs, and record the characteristics of the abnormal data to optimize future anomaly detection strategies and improve the overall stability of the system.

[0068] A fusion algorithm based on historical data regression analysis and physical modeling is used to correct marked abnormal data and interpolate and compensate using data from adjacent time windows to reduce the impact of missing data on scheduling decisions.

[0069] Using a fusion algorithm based on historical data regression analysis and physical modeling, we correct the marked abnormal data and use adjacent time window data for interpolation compensation to reduce the impact of missing data on scheduling decisions. The specific steps are as follows:

[0070] Classify abnormal data, analyze its impact on scheduling strategies, and select the optimal compensation strategy to reduce the negative impact of data loss;

[0071] After completing the collection of real-time data such as photovoltaic power generation, load demand, battery status, and cross-verifying the data consistency through multi-source data fusion technology, the system will use adaptive threshold detection and statistical analysis methods to identify and eliminate abnormal data. However, the elimination of abnormal data may lead to information loss, which will affect subsequent AI scheduling decisions. Therefore, the system first needs to classify the abnormal data and analyze its impact on the scheduling strategy. For example, if the abnormal data is short-term sensor jitter (such as instantaneous power surge), it can be directly processed by smoothing filtering; if it is long-term data loss (such as SOC loss caused by communication failure), more complex regression analysis and physical modeling are required for correction. By classifying abnormal data, the system can select the optimal compensation strategy for different types of data missing situations to minimize the impact of data missing on scheduling decisions.

[0072] Use regression models to predict and correct abnormal data to align with historical trends and improve data accuracy and reliability;

[0073] For outliers caused by missing or erroneous data, historical data can be used for regression analysis and correction. First, the system searches for similar historical data in the context time window of the abnormal data and establishes a regression model (such as linear regression, polynomial regression, or deep learning models such as LSTM) to predict the missing data. For example, when correcting abnormal photovoltaic power generation, the regression model can be trained using photovoltaic data under similar weather conditions in recent days, and multivariate fitting can be performed in combination with external factors such as temperature and irradiance to deduce reasonable correction values. At the same time, in order to enhance the stability of the prediction, the system can use the weighted moving average method (WMA) to perform a weighted combination of multiple predicted values, thereby reducing the impact of a single model error on the final data correction. This process ensures that the corrected data matches the historical trend, thereby improving the authenticity and reliability of the data.

[0074] Physical modeling is used to constrain the rationality of regression correction data to ensure that the corrected data conforms to physical laws and avoids erroneous compensation affecting system operation.

[0075] Although regression analysis can provide reasonable correction values for abnormal data, relying solely on statistical methods may cause the corrected data to violate physical laws. Therefore, this step introduces physical modeling for constraint correction. For example, in the energy storage system, the change in battery SOC is affected by factors such as charge and discharge efficiency, current, and voltage. Therefore, a battery dynamic model (such as the Coulomb integral method or the Thevenin equivalent circuit model) can be constructed to limit the reasonable range of the correction value. Suppose the regression analysis predicts that the SOC should be 60%, but according to the physical modeling calculation, the SOC can only reach a maximum of 55% under the current load conditions, then the correction value should be adjusted to 55%. In addition, if the predicted value of photovoltaic power generation is higher than the maximum power output allowed by the physical model, the system should be limited according to the battery charge and discharge curve to avoid overestimating the power generation. This strategy ensures that the corrected data is consistent with historical trends and meets physical constraints, avoiding the impact of incorrect compensation on the stable operation of the energy storage system.

[0076] Linear interpolation, spline interpolation, and Kalman filtering methods are used to fill short-term missing data, and combined with physical constraint optimization compensation to ensure data continuity and scheduling stability;

[0077] For missing data within a short time window, the system uses interpolation compensation technology to reduce the impact of data gaps on the AI scheduling model. First, select an appropriate interpolation method based on the time span of the missing data, for example:

[0078] Linear interpolation: It is suitable for situations where data changes smoothly in a short period of time, such as compensating for the loss of photovoltaic power generation when there is no sudden weather change.

[0079] Spline Interpolation: Applicable to situations where data changes are more complex, such as compensation of the changing curve of battery SOC during dynamic charging and discharging.

[0080] Kalman Filter: Applicable to situations where multiple variables change in coordination, such as joint compensation of voltage, current and SOC.

[0081] Use time series prediction models (such as LSTM or Transformer) to predict future PV power generation, load demand, and energy storage status. Establish a dispatch warning mechanism to adjust energy allocation strategies in advance when unexpected anomalies are predicted, thereby reducing the impact of abnormalities on the stability of the energy storage system.

[0082] Using a time series prediction model (such as LSTM or Transformer) to predict future PV power generation, load demand, and energy storage status, a scheduling warning mechanism is established. When an unexpected anomaly is predicted, the energy allocation strategy is adjusted in advance. The specific steps to reduce the impact of anomalies on the stability of the energy storage system are as follows:

[0083] Organize and normalize PV power generation, load demand, and energy storage status data, using time window technology to capture long-term and short-term trends to ensure stable data quality;

[0084] After completing data anomaly detection and correction, the system has obtained high-quality, continuous data on photovoltaic power generation, load demand, and energy storage status. However, this data only reflects the current and historical operating status and cannot provide a forward-looking decision-making basis for scheduling. Therefore, the goal of this step is to construct an input data set for the time series prediction model so that it can learn historical data patterns and predict energy status for a period of time in the future. First, the system selects key feature variables, including photovoltaic power generation power, ambient temperature, solar irradiance, battery SOC, grid electricity price, load demand, etc., and normalizes the data to eliminate the numerical dimension differences between different variables. Subsequently, the system constructs a time window and uses data from the past period of time (such as the past 24 hours or 7 days) as model input to capture long-term and short-term time series characteristics. In addition, for the missing parts after the anomaly data is corrected, the system can use sliding window filling technology to ensure that the model's overall learning ability is not affected by missing data at individual time points.

[0085] Select a time series forecasting model for training, optimize hyperparameters, and perform cross-validation to improve the forecast accuracy of photovoltaic power generation, load demand, and energy storage status;

[0086] After constructing the input dataset, the system needs to train a time series forecasting model to predict future PV power generation, load demand, and energy storage status. Common time series forecasting models include long short-term memory (LSTM), Transformer, and ARIMA (autoregressive integrated moving average). LSTM is suitable for scenarios with long time series dependencies and can capture cyclical variations in PV power generation and load demand. Transformer is more adept at handling complex time series relationships and is suitable for long-term forecasting. ARIMA is suitable for short-term trend forecasting and has a lower computational cost. During model training, the system uses historical data for supervised learning and uses the mean squared error (MSE) or root mean squared error (RMSE) as the loss function to optimize model parameters. Furthermore, the system can use Bayesian optimization methods to adjust hyperparameters (such as the learning rate and time window size) to improve the model's forecasting accuracy. Furthermore, to enhance the model's generalization capabilities, the system performs cross-validation. This involves dividing the dataset into training and validation sets and training them over different time periods to ensure the model's adaptability to various operating conditions.

[0087] Based on the prediction results, early warning thresholds are set to dynamically adjust energy distribution before photovoltaic power generation decreases or load surges, and reinforcement learning is combined to optimize the scheduling strategy;

[0088] Based on a trained time-series prediction model, the system can predict key parameters such as photovoltaic power generation, load demand, and battery SOC over a period of time. This allows the system to establish a scheduling warning mechanism to respond to unexpected abnormal situations. For example, if the system predicts a significant drop in photovoltaic power generation (such as the onset of rainy weather) or a surge in load demand (such as during peak hours for communication base stations), it can proactively adjust energy allocation strategies. Specifically, the system sets warning thresholds. For example, if the predicted photovoltaic power generation drops by more than 30% within the next hour, or if the load demand increases by more than 20% within 30 minutes, an alert is triggered. Furthermore, the system can leverage reinforcement learning methods and the prediction results to dynamically adjust battery charging and discharging schedules, such as pre-charging during periods of low electricity prices to ensure sufficient power supply during future periods of high demand. Furthermore, the system builds a knowledge base based on historical abnormalities. By comparing current predictions with historical abnormal patterns, the system can further optimize the warning mechanism and more accurately predict potential abnormal conditions.

[0089] Optimize energy storage scheduling strategies before anomalies occur, and use a model adaptive adjustment mechanism to correct prediction errors, thereby improving the stability and predictive capabilities of the energy storage system;

[0090] After predicting a sudden anomaly, the system needs to quickly adjust its scheduling strategy and make real-time corrections when the anomaly occurs. For example, if it predicts that photovoltaic power generation will decrease by 50% in the next hour, the system can increase the battery charging power in advance before the anomaly occurs, or adjust the load-side power supply strategy to reduce dependence on the external power grid. In addition, to ensure that the scheduling strategy is always effective during execution, the system can adopt a model adaptive adjustment mechanism, that is, re-evaluate the prediction error after the scheduling is executed. If the actual situation deviates too much from the predicted value, the prediction model parameters are adjusted to improve the model's accuracy in the future. For example, if the load prediction error is persistently high, the system can adjust the time window length or introduce new characteristic variables (such as temperature changes) to optimize the prediction model. In addition, after an abnormal event occurs, the system can use a self-learning mechanism to incorporate new abnormal patterns into the training data to improve the prediction and scheduling capabilities of similar anomalies in the future.

[0091] Incorporating reinforcement learning algorithms, a reward function is constructed that incorporates factors such as battery life, grid electricity prices, and load fluctuations. Based on a distributed reinforcement learning framework, energy scheduling strategies are optimized to achieve dynamic optimal scheduling of energy storage systems under different operating conditions.

[0092] Integrating reinforcement learning algorithms, we construct a reward function that incorporates factors such as battery life, grid electricity prices, and load fluctuations. Using a distributed reinforcement learning framework, we optimize the energy scheduling strategy to achieve dynamic optimal scheduling for the energy storage system under different operating conditions. The specific steps are as follows:

[0093] Combined with time series forecast results, the state space of reinforcement learning is defined, including key parameters such as photovoltaic power generation, load demand, battery SOC, and grid electricity price, to build a dynamic scheduling environment;

[0094] After completing the training of the time series prediction model and the establishment of the scheduling warning mechanism, the system is able to predict future photovoltaic power generation, load demand and energy storage status, and adjust the energy allocation strategy in the event of sudden anomalies. However, the scheduling strategy is still based on empirical rules or simple optimization methods, and lacks global optimality. Therefore, the goal of this step is to build a dynamically optimized energy scheduling strategy based on reinforcement learning. First, it is necessary to define the reinforcement learning environment, in which the state space (StateSpace) should contain key parameters such as photovoltaic power generation forecast value, load forecast value, battery SOC, grid electricity price, historical charging and discharging behavior, etc., to fully describe the current state of the energy storage system. At the same time, the state variables need to be standardized to ensure that features of different numerical magnitudes have a balanced impact on the model. In addition, the system can use dimensionality reduction techniques (such as PCA or Autoencoder) to reduce the redundancy of state variables and improve learning efficiency.

[0095] A reward function is designed based on economic efficiency, battery life, load stability, and renewable energy utilization to guide the agent to learn the optimal energy scheduling strategy and dynamically adjust the weights to adapt to different application scenarios.

[0096] In the reinforcement learning framework, the agent continuously optimizes the energy scheduling strategy through trial and error learning, and the reward function determines the goal of system optimization. Therefore, the core of this step is to design a reward function that balances multiple factors so that the energy storage system can achieve dynamic optimal scheduling under different operating conditions. The reward function can include the following key factors:

[0097] 1. Economic efficiency: Based on the electricity price signal of the power grid, charging is prioritized during low electricity price periods and discharging is prioritized during high electricity price periods to maximize economic benefits.

[0098] 2. Battery life: Consider the battery's DOD (depth of discharge) and number of charge and discharge cycles to avoid overcharging and discharging to extend battery life.

[0099] 3. Stability: Reduce the impact of load fluctuations on the energy storage system, ensure that the SOC is within a reasonable range, and avoid overcharging or over-discharging.

[0100] 4. Utilization rate of renewable energy: Improve the local consumption rate of photovoltaic power generation and reduce the phenomenon of abandoned light.

[0101] Adopt reinforcement learning algorithms such as PPO / DDPG and use a distributed reinforcement learning framework to accelerate training, enabling the intelligent agent to learn the optimal energy scheduling strategy under different working conditions, improving learning efficiency and convergence speed;

[0102] After building the reinforcement learning environment and designing the reward function, the system needs to select an appropriate reinforcement learning algorithm and perform training based on a distributed architecture to accelerate policy optimization. Common reinforcement learning algorithms include Deep Q-Network (DQN), Proximal Policy Optimization (PPO), and Deep Deterministic Policy Gradient (DDPG). DQN is suitable for discrete action spaces, while PPO and DDPG are suitable for continuous action spaces. In energy scheduling problems, battery charge and discharge power is typically a continuous variable, making PPO or DDPG more advantageous. Furthermore, to improve training efficiency, the system can employ a distributed reinforcement learning framework (such as Ape-X or RayRLlib) to train multiple agents in parallel on multiple computing nodes, allowing them to simultaneously explore optimal policies under different conditions. For example, one agent can learn a scheduling policy for high-light conditions, while another can learn a policy for low-light conditions at night. Ultimately, federated learning is used to integrate the optimal policies across different conditions. Furthermore, the system can utilize Experience Replay and Target Network technologies to improve the stability and convergence speed of reinforcement learning.

[0103] Through online learning and model adaptive adjustment mechanisms, the reinforcement learning agent can continuously optimize the scheduling strategy and dynamically adapt to environmental changes and sudden anomalies, improving the stability and explainability of the energy storage system.

[0104] After the reinforcement learning agent completes training, the system must continuously optimize and adjust its scheduling strategy during actual operation to adapt to new environmental changes and unexpected anomalies. For example, if the system predicts a sharp drop in photovoltaic power generation within the next hour, the scheduling strategy needs to adjust the battery charging and discharging schedule in advance to ensure stable power supply. Furthermore, the system can utilize online learning mechanisms to enable the agent to continuously update its strategy during actual operation to adapt to dynamic changes such as grid price fluctuations and the addition of new equipment. For example, when battery aging causes increased SOC estimation errors, the system can retrain the prediction model using the reinforcement learning model adaptive adjustment mechanism to optimize energy scheduling. Furthermore, to enhance the interpretability of the strategy, the system can employ reinforcement learning methods based on the attention mechanism, enabling the agent to analyze which state variables have the greatest impact on decision-making, thereby improving the transparency and controllability of the strategy.

[0105] During the AI scheduling process, confidence estimation and Bayesian optimization methods are combined to dynamically monitor the reliability of scheduling decisions. When the confidence level falls below the set threshold, the expert system-based rule correction mechanism is triggered to make reasonable adjustments to the scheduling results to ensure robustness under abnormal circumstances.

[0106] During the AI scheduling process, confidence estimation and Bayesian optimization methods are combined to dynamically monitor the reliability of scheduling decisions. When the confidence level falls below the set threshold, the expert system-based rule correction mechanism is triggered to make reasonable adjustments to the scheduling results. The specific steps to ensure robustness in abnormal situations are as follows:

[0107] During the AI scheduling process, the confidence of the current scheduling strategy is first evaluated to determine the reliability of the decision. The confidence estimation uses the Bayesian neural network (BNN) or Monte Carlo (Monte Carlo Dropout) method to calculate the mean and variance of the scheduling strategy through multiple forward propagations, thereby quantifying the uncertainty of the scheduling strategy. The formula is as follows:

[0108]

[0109] , where C t is the confidence of the scheduling decision, with a value range of [0, 1]. The larger the value, the higher the decision reliability. t ] is the uncertainty (standard deviation) of the current scheduling decision, which is calculated by the Bayesian neural network, E[P t ] is the expected value of the scheduling decision at the current moment, that is, the average decision value in multiple forward propagations, and ∈ is a very small positive number to prevent the denominator from approaching zero;

[0110] When the confidence level C t Below the set threshold C thresh When (such as C t <0.7), the rule correction mechanism is triggered to ensure the stability of the scheduling strategy;

[0111] Calculate the confidence C of the scheduling decision t , if it is lower than the set threshold C thresh , it is necessary to further optimize the scheduling decision to enhance the robustness of the system.

[0112] When the confidence level is lower than the threshold C thresh When the current scheduling parameters (such as battery charging and discharging power P t ) is optimized to find the optimal dispatch value. Bayesian Optimization (BO) is used to dynamically adjust the dispatch strategy so that the optimized decision reduces risk while improving economy and energy storage system stability. The objective function of Bayesian optimization is defined as follows:

[0113]

[0114] , where is the optimized battery charge and discharge power, R(P) is the reward function, which takes into account factors such as grid electricity price, battery life, and load fluctuation, and λ is the risk weight coefficient, which controls the degree of penalty for uncertainty. is the value of P that maximizes the objective function;

[0115] Bayesian optimization builds a proxy model through Gaussian Process Regression (GPR) and uses the expected improvement criterion (EL) or the probability improvement criterion (PI) to select the optimal scheduling parameters. After optimization, the system will As the revised scheduling value, it is input into the next step of the rule correction mechanism. Summary: Using Bayesian optimization method to adjust scheduling parameters While optimizing the reward function, the scheduling uncertainty is reduced, and the reliability and economy of the scheduling scheme are improved.

[0116] Although Bayesian optimization provides optimized scheduling parameters However, it may still lead to unacceptable scheduling decisions in extreme cases (such as sudden load surge or battery SOC exceeding the safety range). Therefore, a rule correction mechanism based on expert system is adopted to correct the To make the final correction and ensure the safety and feasibility of the scheduling strategy, the expert system is based on a set of predefined rules, for example: if the battery SOC is lower than 20% and If it is still discharging, it is forced to Set to 0; if the grid electricity price is much lower than the historical average, it is encouraged to increase the charging power; if the load forecast value fluctuates violently in a short period of time, For smoothing processing, the rule correction formula is expressed as follows:

[0117]

[0118] , where is the final battery charge and discharge power, I(·) is the indicator function, which takes 1 if the condition is met and 0 otherwise, α is the smoothing weight factor, ΔP smooth It is a smooth adjustment value based on load fluctuation.

[0119] Finally, the system adopts It is used as a new scheduling strategy and fed back to the timing prediction module to update the calculation basis for future scheduling decisions.

[0120] A federated learning architecture enables multiple base stations to share abnormal data patterns and collaboratively optimize AI models while protecting data privacy. Active learning mechanisms dynamically expand the training set, improving the model's adaptability to data noise and outliers, and enhancing the robustness of the overall scheduling system.

[0121] A federated learning architecture is used to enable multiple base stations to share abnormal data patterns and collaboratively optimize AI models while protecting data privacy. The active learning mechanism dynamically expands the training set, improves the model's adaptability to data noise and outliers, and enhances the robustness of the overall scheduling system. The specific steps are as follows:

[0122] During the optimization and scheduling of hybrid energy storage systems, the data of a single base station is limited, and abnormal data patterns vary depending on factors such as geographic location, climate, and load characteristics. Therefore, to improve the generalization capability of the AI scheduling model, a federated learning (FL) architecture is adopted to enable multiple base stations to share abnormal data patterns while ensuring data privacy. Specifically, each base station maintains a local AI model and, after a round of local training, sends the model gradient rather than the original data to the central server. The server aggregates the model gradients of multiple base stations based on the FedAvg algorithm and updates the global model parameters θ. The core calculation formula for federated model aggregation is as follows:

[0123]

[0124] , where θ (j+1) is the global model parameter of the j+1th round, N is the total number of base stations, each base station trains a local model independently, and w i is the data volume of base station i, which represents the number of data samples collected by the base station, and W is the normalization factor of the total amount of data from all base stations. is the model parameter of base station i in the jth round of training, γ is the weight coefficient of abnormal data on model update (used to enhance the adaptability of the model to abnormal data), M is the normalization factor of the total amount of abnormal data of all base stations, is the model gradient of base station i after local training;

[0125] This step optimizes the global model through a dual-weighted mechanism (based on the amount of data and the proportion of abnormal data), so that base stations with more abnormal data contribute more to the global model, improving the model's adaptability to abnormal data patterns, while preventing the model with small data sets from having too much impact on the global optimization.

[0126] Because the adaptability of the energy storage system scheduling AI model to abnormal data depends on the diversity and coverage of the training data, after federated learning optimizes the global model, an active learning mechanism is further adopted to dynamically expand the training set and improve the model's robustness to noise and outliers. Specifically, after each round of training, the system calculates the uncertainty of the current model and selects the data samples with the most information to add to the training set, thereby improving the model's generalization ability in abnormal scenarios. The active learning sample selection is based on the principle of maximizing information gain (MIG), calculated as follows:

[0127]

[0128] , where S * is the selected active learning sample subset, S c Is the candidate data set, containing all samples to be screened, x represents the candidate sample set S c A data point in , p(y|x, θ (j+1) ) is the global model parameter θ (j+1) Under this condition, the predicted probability distribution of sample x represents the uncertainty of the sample, H(p(y|x,θ (j+1) )) is the prediction entropy of the current global model, K is the number of sub-models used to calculate uncertainty, represents the parameters of the k-th sub-model after j+1 rounds of training, The goal is to maximize information gain and select valuable samples.

[0129] The core idea of the above steps is to calculate the change in sample prediction entropy, that is, to select those samples that reduce the global model entropy the most and add them to the training set to maximize the information gain of the model. * The information will be fed back to each base station for local training, and the global model will be optimized again through the federated learning framework to form a dynamic cyclic optimization mechanism.

[0130] Implementation method 1: In photovoltaic energy storage base stations, the accuracy and continuity of data are crucial for intelligent scheduling. Since the energy storage system involves multiple data sources, including photovoltaic power generation, battery SOC (state of charge), load demand, etc., the data collection process may be affected by factors such as sensor failure, communication delay, environmental interference, etc., resulting in data anomalies. In order to ensure the stability of the scheduling system, this implementation method first uses multi-source data fusion and adaptive threshold detection technology to monitor the collected data in real time and eliminate anomalies. For example, the system can compare the data consistency of multiple sensors. If there is a significant deviation between the power output of the photovoltaic MPPT controller and the power measured by the meter, it may indicate that a sensor has failed. In addition, the system can also use statistical methods such as sliding window mean and Z-score analysis to identify anomalies, and combine time series prediction to detect short-term mutations to improve data quality.

[0131] After completing the data cleaning, the system needs to correct and compensate for abnormal data. For marked abnormal data, a fusion algorithm of historical data regression analysis and physical modeling can be used for correction. For example, in the case of SOC data loss, the system can use a regression model (such as polynomial regression or LSTM) to predict a reasonable SOC value based on the SOC change trend under similar load conditions in recent days. At the same time, physical constraint correction is performed in combination with the battery charge and discharge model (such as the Thevenin equivalent circuit model) to ensure that the corrected SOC value does not exceed the physically possible range. In addition, for missing data within a short time window, interpolation compensation methods such as linear interpolation or spline interpolation can be used to keep the data smooth and reduce the impact on scheduling decisions.

[0132] After the data quality is guaranteed, the system needs to predict photovoltaic power generation, load demand and battery SOC in the future to provide forward-looking guidance for scheduling decisions. The system can use time series prediction models such as LSTM (Long Short-Term Memory Network), Transformer or ARIMA (Autoregressive Integrated Moving Average Model). Among them, LSTM is suitable for capturing load change patterns with long-term dependencies, while Transformer can process complex time series data and improve prediction accuracy. When training the system, historical data can be used for supervised learning, and Bayesian optimization can be used to adjust hyperparameters to improve the generalization ability of the model. Based on the prediction results, the system establishes a scheduling warning mechanism to adjust the battery charging and discharging plan in advance when photovoltaic power generation drops sharply or the load surges to ensure the stable operation of the energy storage system.

[0133] The advantage of this implementation is that it not only improves data reliability but also proactively adjusts scheduling strategies through time series prediction, mitigating the impact of anomalies on energy storage system stability. This approach is suitable for scenarios with unstable photovoltaic power generation and large fluctuations in load demand, effectively improving the adaptability and operational efficiency of energy storage systems.

[0134] Implementation 2: After completing data cleaning and time series prediction, the system has high-quality input data and can predict future photovoltaic power generation, load demand, and battery SOC. However, traditional rule-based scheduling strategies often rely on manual experience and are difficult to adapt to complex dynamic environments. Therefore, this implementation uses reinforcement learning (RL) algorithms to optimize the energy storage system's energy scheduling strategy.

[0135] First, a reinforcement learning environment needs to be constructed, defining the state space, action space, and reward function. The state space includes key parameters such as the predicted photovoltaic power generation value, the predicted load value, the battery state of charge (SOC), the grid electricity price, and historical charging and discharging behavior to comprehensively describe the current state of the energy storage system. The action space defines executable scheduling policies, such as charging power, discharging power, and the proportion of grid power purchased. The reward function forms the basis for the agent's optimization strategy and typically includes the following key factors:

[0136] Economic efficiency: Prioritize charging during low electricity price periods and discharging during high electricity price periods to maximize economic benefits.

[0137] Battery life: Consider the depth of charge and discharge (DOD) and the number of cycles to avoid overcharging and discharging to extend the battery life.

[0138] System stability: Reduce the impact of load fluctuations on the energy storage system, ensure that the SOC remains within a reasonable range, and prevent overcharging or over-discharging.

[0139] Utilization rate of renewable energy: Improve the local consumption rate of photovoltaic power generation and reduce the phenomenon of abandoned light.

[0140] In terms of algorithm selection, the system can adopt reinforcement learning methods such as PPO (Proximal Policy Optimization) or DDPG (Deep Deterministic Policy Gradient). PPO is suitable for high-dimensional continuous action spaces, while DDPG is suitable for low-latency real-time control tasks. In addition, to accelerate training, the system can adopt a distributed reinforcement learning framework (such as Ape-X or RayRLlib) to train the agent in parallel on multiple computing nodes, allowing it to simultaneously explore the optimal strategy under different working conditions. In addition, combined with online learning mechanisms, the system can continuously optimize the scheduling strategy during actual operation, allowing it to dynamically adapt to environmental changes and improve its intelligence level.

[0141] The advantage of this implementation is that it can automatically learn the optimal scheduling strategy without manual intervention, is suitable for complex and changeable photovoltaic energy storage scenarios, and can significantly improve the economy and stability of the energy storage system.

[0142] Implementation 3: During energy storage system operation, in addition to daily load fluctuations and PV power generation variations, unexpected anomalies such as extreme weather and battery failures may also occur. To address these situations, this implementation introduces an intelligent anomaly detection and correction mechanism to improve the system's adaptability and robustness.

[0143] First, the system uses confidence estimation and Bayesian optimization methods to evaluate the reliability of the scheduling strategy in real time. For example, during AI scheduling, if the system detects that the current load forecast deviates significantly from the actual load (exceeding twice the standard deviation of the historical average error), the prediction model may have failed, and the anomaly correction mechanism needs to be triggered. Furthermore, if the confidence level of the PV power generation forecast results is low (e.g., if the LSTM prediction variance is large), the system can reduce its reliance on the forecast data and adopt a more conservative scheduling strategy.

[0144] After the anomaly correction mechanism is triggered, the system can employ an expert system rule correction mechanism to optimize the scheduling strategy based on historical data and operational experience. For example, if prolonged rainy weather results in insufficient photovoltaic power supply, the system can preemptively increase the proportion of power purchased from the grid and reduce the power supply priority for non-essential loads. Furthermore, the system can incorporate a federated learning architecture, enabling multiple base stations to share anomaly data patterns and collaboratively optimize AI models without compromising data privacy, thereby improving the overall system's ability to respond to anomalies. For example, if a base station experiences extreme high temperatures, its anomaly handling experience can be used to optimize the scheduling strategies of other base stations, enhancing the intelligent scheduling capabilities of the entire network.

[0145] The advantage of this implementation is that it can effectively respond to sudden abnormal situations, avoid system instability caused by prediction errors or model failures, ensure the long-term reliable operation of the photovoltaic energy storage system, and is suitable for high-requirement, high-reliability energy management scenarios.

[0146] Through data cleaning, time series prediction, and reinforcement learning-based scheduling optimization, this system enables intelligent management of energy storage systems in complex environments and improves their adaptability to dynamic changes. First, multi-source data fusion and anomaly detection technologies are employed to ensure the accuracy of key data such as photovoltaic power generation, load demand, and battery SOC, preventing issues such as sensor failure and communication delays from impacting scheduling decisions. Second, using time series prediction models such as LSTM and Transformer, the system can predict future photovoltaic power generation and load demand in advance and establish a scheduling warning mechanism to optimize energy storage strategies before anomalies occur, preventing sudden load shocks from causing unstable power supply. Furthermore, by combining reinforcement learning-based scheduling optimization strategies, the system can autonomously learn optimal scheduling solutions based on grid electricity prices, photovoltaic output, and load fluctuations. For example, it can reduce discharge during peak electricity prices to minimize battery loss while ensuring a stable power supply. Through intelligent optimization, the system can dynamically adjust its strategies under varying climate conditions, load patterns, and energy storage states, enhancing the adaptability of the energy storage system and improving the long-term stable operation of photovoltaic energy storage base stations.

[0147] Through intelligent control, the present invention enables the system to comprehensively consider factors such as grid electricity prices, photovoltaic power generation forecasts, and battery life, and dynamically adjust the charging and discharging plan, thereby reducing operating costs. For example, the system can prioritize charging during periods of low electricity prices and release stored energy during peak electricity prices, reducing the grid's electricity purchase expenditures while increasing the local absorption rate of photovoltaic power generation. In addition, to prevent battery aging due to frequent deep charging and discharging, the system optimizes scheduling through reinforcement learning, intelligently controls the battery SOC within the optimal operating range of 40%-80%, reduces overcharging and over-discharging, and improves battery health (SOH). At the same time, combined with abnormal scenario detection and intelligent correction mechanism, when extreme load fluctuations, battery aging, or abnormal ambient temperature are detected, the system can automatically adjust the scheduling strategy to prevent battery overheating or abnormal discharge, improve battery life, and reduce long-term operating and maintenance costs.

[0148] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0149] 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.

[0150] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0151] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0152] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0153] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0154] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0155] 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.

[0156] 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.

[0157] 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. The hybrid energy storage system optimization scheduling method based on AI intelligent control is characterized by: The following steps are involved: Collect real-time data and cross-validate data consistency through multi-source data fusion technology; use adaptive threshold detection combined with statistical analysis methods to identify data anomalies caused by sensor failures, communication delays, and environmental interference, and eliminate abnormal data; A fusion algorithm based on historical data regression analysis and physical modeling is used to correct marked abnormal data and interpolate and compensate using data from adjacent time windows to reduce the impact of missing data on scheduling decisions. Use time series forecasting models to predict future photovoltaic power generation, load demand, and energy storage status, establish a dispatch warning mechanism, and adjust energy allocation strategies in advance when sudden anomalies are predicted to reduce the impact of anomalies on the stability of the energy storage system; Combined with reinforcement learning algorithms, a reward function is constructed to optimize the energy scheduling strategy based on a distributed reinforcement learning framework, enabling the energy storage system to achieve dynamic optimal scheduling under different operating conditions; During the AI scheduling process, confidence estimation and Bayesian optimization methods are combined to dynamically monitor the reliability of scheduling decisions. When the confidence level falls below the set threshold, the expert system-based rule correction mechanism is triggered to make reasonable adjustments to the scheduling results to ensure robustness under abnormal circumstances. A federated learning architecture enables multiple base stations to share abnormal data patterns and collaboratively optimize AI models while protecting data privacy. Active learning mechanisms dynamically expand the training set, improving the model's adaptability to data noise and outliers, and enhancing the robustness of the overall scheduling system. The specific steps for combining reinforcement learning algorithms, constructing reward functions, and optimizing energy scheduling strategies based on a distributed reinforcement learning framework to achieve dynamic optimal scheduling of energy storage systems under different operating conditions are as follows: Combined with the time series prediction results, the state space of reinforcement learning is defined to build a dynamic scheduling environment; A reward function is designed based on economic efficiency, battery life, load stability, and renewable energy utilization to guide the agent to learn the optimal energy scheduling strategy and dynamically adjust the weights to adapt to different application scenarios. Adopting the PPO / DDPG reinforcement learning algorithm and utilizing the distributed reinforcement learning framework to accelerate training, the intelligent agent learns the optimal energy scheduling strategy under different working conditions, improving learning efficiency and convergence speed; Through online learning and model adaptive adjustment mechanisms, the reinforcement learning agent can continuously optimize the scheduling strategy and dynamically adapt to environmental changes and sudden anomalies, thereby improving the stability and explainability of the energy storage system.

2. The hybrid energy storage system optimization scheduling method based on AI intelligent control according to claim 1 is characterized in that: Collect real-time data and cross-validate data consistency through multi-source data fusion technology. Use adaptive threshold detection combined with statistical analysis methods to identify data anomalies caused by sensor failures, communication delays, and environmental interference, and eliminate abnormal data. The specific steps are as follows: Collect key data of photovoltaics, batteries, and loads, and perform time alignment, format standardization, and interpolation processing to ensure data consistency and availability; Adopt multi-source data fusion technology to cross-validate data from different sensors to improve data accuracy; Combining adaptive threshold detection, statistical analysis, and time series prediction methods to dynamically identify data anomalies and improve the sensitivity and adaptability of anomaly detection; Abnormal data is filled through sliding window filtering, regression analysis and deep learning prediction, and the compensation strategy is optimized in combination with physical constraints to ensure the stable operation of the energy storage system.

3. The hybrid energy storage system optimization scheduling method based on AI intelligent control according to claim 1 is characterized in that: Using a fusion algorithm based on historical data regression analysis and physical modeling, we correct the marked abnormal data and use adjacent time window data for interpolation compensation to reduce the impact of missing data on scheduling decisions. The specific steps are as follows: Classify abnormal data, analyze its impact on scheduling strategies, and select the optimal compensation strategy to reduce the negative impact of data loss; Use regression models to predict and correct abnormal data to align with historical trends and improve data accuracy and reliability; Physical modeling is used to constrain the rationality of regression correction data to ensure that the corrected data conforms to physical laws and avoids erroneous compensation affecting system operation. Linear interpolation, spline interpolation and Kalman filtering methods are used to fill short-term missing data, and combined with physical constraint optimization compensation to ensure data continuity and scheduling stability.

4. The hybrid energy storage system optimization scheduling method based on AI intelligent control according to claim 1 is characterized in that: Using a time series forecasting model to predict future PV power generation, load demand, and energy storage status, a scheduling warning mechanism is established to proactively adjust energy allocation strategies when unexpected anomalies are predicted. The specific steps to mitigate the impact of these anomalies on energy storage system stability are as follows: Organize and normalize PV power generation, load demand, and energy storage status data, using time window technology to capture long-term and short-term trends to ensure stable data quality; Select a time series forecasting model for training, optimize hyperparameters, and perform cross-validation to improve the forecast accuracy of photovoltaic power generation, load demand, and energy storage status; Based on the prediction results, early warning thresholds are set to dynamically adjust energy distribution before photovoltaic power generation decreases or load surges, and reinforcement learning is combined to optimize the scheduling strategy; The energy storage scheduling strategy is optimized before an anomaly occurs, and the model adaptive adjustment mechanism is used to correct the prediction error, thereby improving the stability and predictive ability of the energy storage system.

5. The hybrid energy storage system optimization scheduling method based on AI intelligent control according to claim 1 is characterized in that: During the AI scheduling process, confidence estimation and Bayesian optimization methods are combined to dynamically monitor the reliability of scheduling decisions. When the confidence level falls below the set threshold, the expert system-based rule correction mechanism is triggered to make reasonable adjustments to the scheduling results. The specific steps to ensure robustness in abnormal situations are as follows: During the execution of AI scheduling, the confidence of the current scheduling strategy is first evaluated to determine the reliability of the decision. The mean and variance of the scheduling strategy are calculated through multiple forward propagations to quantify the uncertainty of the scheduling strategy. The formula is as follows: , where is the confidence of the scheduling decision, is the uncertainty of the current scheduling decision, is the expected value of the scheduling decision at the current moment, It is a very small positive number to prevent the denominator from approaching zero; When confidence Below the set threshold When , the rule correction mechanism is triggered to ensure the stability of the scheduling strategy; When the confidence level is lower than the threshold When the current scheduling parameters are optimized to find the optimal scheduling value, the scheduling strategy is dynamically adjusted using Bayesian optimization, so that the optimized decision reduces the risk while improving the economy and stability of the energy storage system. The objective function of Bayesian optimization is defined as follows: , where is the optimized battery charge and discharge power, is a reward function that takes into account factors such as grid electricity price, battery life, and load fluctuation. is the risk weight coefficient, which controls the degree of penalty for uncertainty. is the one that maximizes the objective function Get the value.

6. The hybrid energy storage system optimization scheduling method based on AI intelligent control according to claim 5 is characterized in that: Although Bayesian optimization provides optimized scheduling parameters However, it will still lead to unacceptable scheduling decisions in extreme cases. Therefore, a rule correction mechanism based on expert system is adopted to correct Perform final corrections to ensure the safety and feasibility of the scheduling strategy. The rule correction formula is expressed as follows: , where yes , Is an indicator function, which takes 1 if the condition is met, otherwise takes 0. is the smoothing weight factor, It is a smooth adjustment value based on load fluctuations.

7. The hybrid energy storage system optimization scheduling method based on AI intelligent control according to claim 1 is characterized in that: A federated learning architecture is used to enable multiple base stations to share abnormal data patterns and collaboratively optimize AI models while protecting data privacy. The active learning mechanism dynamically expands the training set, improves the model's adaptability to data noise and outliers, and enhances the robustness of the overall scheduling system. The specific steps are as follows: During the optimization and scheduling of hybrid energy storage systems, the data of a single base station is limited, and abnormal data patterns vary depending on factors such as geographic location, climate, and load characteristics. Therefore, to improve the generalization capability of the AI scheduling model, a federated learning architecture is adopted to enable multiple base stations to share abnormal data patterns while ensuring data privacy. The core calculation formula for federated model aggregation is as follows: , where It is The global model parameters of the wheel, is the total number of base stations, It is a base station The amount of data, is the normalization factor for the total amount of data from all base stations, It is a base station In the Model parameters during round training, is the weight coefficient of abnormal data on model update, is the normalization factor for the total amount of abnormal data of all base stations, It is a base station Model gradients after local training; Because the adaptability of the energy storage system scheduling AI model to abnormal data depends on the diversity and coverage of the training data, after optimizing the global model through federated learning, an active learning mechanism is further adopted to dynamically expand the training set and improve the model's robustness to noise and outliers. The sample selection for active learning is based on the principle of maximizing information gain, calculated as follows: , where is the selected active learning sample subset, Is the candidate data set, containing all samples to be screened, Representative candidate sample set A data point in is the global model parameter Next, for the sample The predicted probability distribution of is the predicted entropy of the current global model, is the number of submodels used to calculate uncertainty, Indicates the The sub-model The parameters after round training, The goal is to maximize information gain and select valuable samples.

Images