An energy storage demand control method and system based on load prediction and battery state evaluation

Abstract

The application discloses a kind of energy storage demand control method and system based on load prediction and battery state evaluation, including data acquisition module, load prediction module, model monitoring and updating module, battery state evaluation module, demand control module and report display module.The application is by real-time monitoring to single-step prediction error and is modified to prediction result when meeting condition, simultaneously, periodically evaluates and updates to load prediction model, so that demand control strategy can continuously adapt to the change of load characteristics and energy storage system operating state.Under the parallel power market environment of two-part electricity price and time-of-use electricity price, the comprehensive optimization of electricity cost is realized, and the stability and economy of energy storage system operation are improved.

Description

Technical Field

[0001] This invention belongs to the field of electrical technology, specifically relating to a method and system for controlling energy storage demand based on load prediction and battery state assessment. Background Technology

[0002] With the ongoing reform of the power system, the power market is gradually shifting from a traditional planned management model to a market-oriented operation model, and power users are increasingly involved in electricity consumption behavior and cost management. For industrial and commercial users, electricity costs account for a large proportion of their overall operating costs. How to achieve reasonable control of electricity costs while ensuring normal electricity demand has become a common technical concern for power users.

[0003] On the user side, configuring energy storage systems and implementing peak shaving and valley filling control in conjunction with electricity load characteristics is a common method for optimizing electricity costs. This type of technical solution is typically based on time-of-use pricing mechanisms, charging the energy storage system during periods of lower electricity prices and discharging it during periods of higher prices to reduce the user's electricity costs within the billing cycle.

[0004] Furthermore, in many countries and regions, a two-part electricity pricing mechanism is commonly used for large-scale industrial and commercial users, dividing electricity costs into two parts: a power consumption price and a demand price. The power consumption price is based on the user's actual electricity consumption within the billing cycle, while the demand price is based on the user's maximum power consumption or assessed demand within the billing cycle. Under this pricing mechanism, the user's overall electricity cost is affected not only by power consumption but also closely related to maximum power demand. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes an energy storage demand control method and system based on load forecasting and battery state assessment. This addresses the problems of existing technologies where demand control relies on fixed thresholds, struggles to adapt to dynamic load changes, and ignores the actual available battery capacity. By performing multi-step time-series forecasting of building electricity load and combining it with historical battery operating data to assess the actual available battery capacity, a dynamic demand control strategy is generated to effectively limit the maximum electricity demand on the user side.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: A method for controlling energy storage demand based on load forecasting and battery state assessment, comprising: Step 1: Data collection, including power data, battery status data, and corresponding timestamp information from the energy storage system; Step 2, Data Preprocessing and Load Forecasting: The collected data is preprocessed by the preprocessing module to obtain feature tensors. Then, the trained load forecasting model generates a multi-step load forecasting sequence for the target period based on the feature tensors, and the single-step forecasting error is monitored in real time. Step 3, Model Monitoring and Update: Periodically evaluate the prediction error of the multi-step load forecasting results and update the load forecasting model when preset conditions are met; Step 4: Battery Status Assessment: Estimate the actual capacity of the battery based on historical battery operating data, and assess the battery's health status based on the actual capacity. Step 5, Demand Control: Generate a charging and discharging control strategy for the energy storage system based on the predicted value at each moment in the multi-step load forecast sequence and the actual battery capacity at the corresponding moment, and control the energy storage converter to execute the charging and discharging control strategy.

[0007] As a further improvement, in step one, a Hadoop distributed data processing architecture is used for data acquisition, and power data and battery status data are processed in batches and in real time in terms of time dimension.

[0008] Further improvements include a time-series-based prediction model, which uses feature tensors to predict load values ​​at multiple future times, forming a multi-step load prediction sequence. The method for real-time monitoring of single-step prediction errors is as follows: the predicted load value is compared with the actual load value to calculate the error, and when the error reaches a preset correction condition, the original multi-step load prediction sequence is discarded, and the input feature tensor is updated again based on the actual detected data to generate a new load prediction sequence. The time-series-based prediction model includes the Informer time-series prediction model.

[0009] Further improvements include preprocessing the collected data, which includes: Outlier handling: Perform negative value detection, box plot detection, and moving mean detection on the original data to remove data that exceeds the reasonable physical range and has an abnormal distribution; Missing value imputation: Time-based imputation and forward / backward imputation are used to fill in missing data to ensure the continuity and integrity of the data time series; Data normalization: The Z-Score normalization method based on statistical features is used to standardize the power data, improving the model convergence speed and inference efficiency. The formula is as follows:

[0010] in It is the mean of the original data. It is the standard deviation of the original data. It is the raw data. It is normalized data; Time feature encoding: Parse the timestamp and extract the month, day, week, hour, and minute features divided according to preset time intervals; Sliding time window algorithm: A feature tensor is constructed using the sliding time window algorithm. The feature tensor includes an input feature tensor and an output feature tensor. The input feature tensor and the output feature tensor form a sample pair.

[0011] Further improvements are made, and the specific steps of step three are as follows: The performance of the load forecasting model is judged based on the prediction error of the trained load forecasting model. When the total energy consumption error and MAPE both exceed 5% for 3 out of 5 consecutive days, the parameters of the load forecasting model are updated. The updated load forecasting model is obtained by retraining using the original training data and the data collected during the operation of the trained load forecasting model as the training set, and then the updated load forecasting model is deployed to the load forecasting module.

[0012] Further improvements are made, and the specific steps of step four are as follows: Identify battery discharge events from historical battery operating data, and then use a regression model. By fitting the capacity trend changes in historical discharge events, the actual capacity of the battery can be predicted. Finally, the battery health status is calculated according to the following formula:

[0013] Among them, SOH represents the battery health status. The actual total battery capacity is obtained through regression model evaluation. This refers to the battery's rated total capacity; where... The actual capacity of the battery , This refers to the number of battery cycles. This refers to the number of days the battery operates. State of Charge (SOC) is the change in SOC during battery discharge. It is defined as the difference between the initial SOC and the final SOC during discharge and is used to characterize the impact of battery depth of discharge on capacity decay. , , and All of these are the fitting coefficients to be fitted.

[0014] Further improvements are made, and the specific steps of step five are as follows: First, a multi-step load prediction sequence is obtained based on the trained load prediction module. Then, the discharge capacity is calculated using the actual battery capacity output by the battery state assessment module. Next, the peak shaving threshold is solved using the multi-step load prediction sequence and the discharge capacity. Finally, the real-time energy storage discharge power is obtained based on the real-time load power and the peak shaving threshold.

[0015] Further improvements have been made to increase the battery's discharge capacity. The calculation formula is as follows:

[0016] in Indicates the battery's discharge capacity. Indicates the actual capacity of the battery. This indicates the battery's state of charge.

[0017] Peak Shaving Threshold The solution is obtained by optimization using the following constraints:

[0018] The peak-shaving threshold satisfying the above equation is solved iteratively using the bisection method. ,in This indicates the number of prediction steps within the target time period. Indicates the first The predicted load value of the step, The time interval between individual prediction time steps, The rated power for energy storage discharge; The real-time discharge power of the energy storage is:

[0019] For real-time load power, This refers to the real-time discharge power of the energy storage system.

[0020] An energy storage demand control system based on load forecasting and battery state assessment, used to execute the above method, the system comprising: Data acquisition module: used to collect data, including power data, battery status data and corresponding timestamp information in the energy storage system; Load forecasting module: The load forecasting module includes a preprocessing module and a load forecasting model; Model monitoring and update module: periodically evaluates the prediction error of the multi-step load prediction results and updates the load prediction model when preset conditions are met; Battery status assessment module: Estimates the actual capacity of the battery based on historical battery operating data, and assesses the health status of the battery based on the actual capacity; Demand control module: Generates a charging and discharging control strategy for the energy storage system based on the load forecast results for the target period and the actual battery capacity, and controls the energy storage converter to execute the charging and discharging control strategy; Report display module: Provides a visual representation of system operating status, prediction results, and control information; During the operation of the energy storage system, when the demand control module receives the corrected load forecast result, it updates the peak shaving threshold based on the current remaining power of the energy storage battery.

[0021] Further improvements include a report display module that shows system operating status information, load forecasting results, battery health status, economic benefit analysis, and alarm logs, and supports generating operational analysis reports based on a time dimension. The beneficial effects of this invention are as follows: 1. By incorporating load forecasting results into demand control decisions, energy storage systems can respond to load demands in advance, thereby effectively suppressing peak electricity demand.

[0022] 2. By evaluating the actual capacity of the battery and using the evaluation results to generate control strategies, demand control failures caused by ignoring changes in the available battery capacity are avoided, thereby improving the reliability of the energy storage system.

[0023] 3. By establishing a real-time monitoring and correction mechanism for single-step forecasting errors, error feedback is created to improve the stability and adaptability of load forecasting results, enabling demand control strategies to better adapt to dynamic load changes.

[0024] 4. By periodically evaluating the prediction error of multi-step load forecasting results and updating the model when the prediction performance is poor, the forecasting model can be adaptively updated according to the dynamic changes in load characteristics. This avoids the decline in prediction accuracy due to changes in load patterns or long-term lack of model updates, thereby improving the stability and long-term effectiveness of demand control strategies.

[0025] 5. Achieving coordinated optimization of load forecasting, battery status assessment, and demand control helps reduce the overall electricity cost for power users and improve the economic operation level of energy storage systems. Attached Figure Description

[0026] The invention will be further illustrated with reference to the accompanying drawings, but the contents of the drawings do not constitute any limitation on the invention.

[0027] Figure 1 This is a diagram of the overall system architecture of the present invention.

[0028] Figure 2 This is a schematic diagram of the data preprocessing process.

[0029] Figure 3This is a flowchart of the load forecasting module.

[0030] Figure 4 This is a flowchart illustrating the model monitoring and update module.

[0031] Figure 5 Flowchart for the battery status assessment module Figure 6 This is a schematic diagram of constructing a feature tensor using the sliding time window algorithm. Detailed Implementation

[0032] To make the purpose, technical solution and advantages of the invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and examples, but the embodiments of the invention are not limited thereto.

[0033] This invention is applied to building power consumption scenarios equipped with user-side energy storage systems, including the load side, battery management system (BMS), energy storage converter (PCS), and energy management system (EMS). The overall system architecture of this invention is as follows: Figure 1 As shown, the modules interact with each other through multiple communication methods: the data acquisition module communicates with the load side, the energy storage converter, and the battery management system to collect building load power data, meter data, and battery status data, and provides the data to the load forecasting module and the battery status assessment module after time alignment; the load forecasting module is deployed on the local EMS, generates multi-step load forecasting results for the target period based on historical load characteristics, and sends the forecasting results to the demand control module; the model monitoring and update module is deployed on the AI ​​server to monitor the forecasting effect of the load forecasting module, updates the load forecasting model when preset conditions are met, and deploys the updated model to the load forecasting module; the battery status assessment module assesses the actual capacity and health status of the energy storage battery based on historical battery operating data and sends the assessment results to the demand control module; the demand control module integrates the load forecasting results and the battery status assessment results to generate a discharge control strategy for the energy storage system, and sends the control strategy to the energy storage converter for execution; the report display module is used to visualize the system operating status, load forecasting results, and energy storage control information.

[0034] Raw operational data from hardware devices such as sensors and metering equipment may suffer from data quality issues during collection and transmission due to communication abnormalities, equipment malfunctions, or environmental interference. These issues may include outliers, missing values, and inconsistent timestamps. To improve the accuracy and stability of subsequent load forecasting models, data preprocessing is necessary before model training and inference.

[0035] Data preprocessing workflow as follows Figure 2As shown, the data preprocessing process includes steps such as outlier detection and handling, missing value imputation, time alignment and unification, data normalization, and feature tensor construction.

[0036] Outlier Detection: First, considering that the actual building load power cannot be negative, any power value less than zero is identified as an outlier. Second, a box plot detection method based on the overall data distribution characteristics is used to identify outliers that significantly deviate from the normal distribution range by analyzing the upper and lower quartiles of the statistical data. Then, a moving average detection method is used to analyze load fluctuations within local time windows; when the power value at a certain moment deviates from the mean of its neighboring time windows by more than a preset threshold, it is marked as an outlier. Finally, all outliers are removed and treated as missing values.

[0037] Missing value imputation: In this embodiment, missing values ​​are handled using a combination of imputation strategies: for short-term consecutive missing data, time-series-based imputation can be used; for data at the beginning of the time series, forward or backward imputation can be used, i.e., using valid data from adjacent time points to fill in the gaps. By employing these methods, the impact of missing values ​​on the training and inference process of the load forecasting model is minimized while ensuring data continuity.

[0038] Timestamp unification and alignment: In the implementation column, a unified time granularity is set according to the system operation requirements. Data from different data sources are resampled or aligned according to the time granularity to construct a time-consistent data sequence.

[0039] Data normalization: In this embodiment, the Z-Score normalization method based on statistical features is used to standardize the power data, ensuring that the data distribution has zero mean and unit variance. This reduces numerical fluctuations during model training and improves model convergence speed and inference efficiency. The formula is shown below:

[0040] in It is the mean of the original data. It is the standard deviation of the original data. It is the raw data. It is the normalized data.

[0041] Time Feature Encoding: In load forecasting tasks, building electrical loads typically exhibit significant time-periodic characteristics. Introducing time feature encoding enables load forecasting models to fully learn the patterns of load variation over time. Multidimensional time features, including month, date, weekday, hour, and minute, are extracted based on the timestamp information of the data to characterize the load's variation characteristics at different time scales.

[0042] Feature Tensor Construction: A sliding time window algorithm is used to construct features from time series data. By selecting continuous historical data as the input window in the time dimension, power features, time features, etc., are combined to construct a feature tensor that the model can accept, which is then used for training and inference of the load forecasting model.

[0043] like Figure 3 As shown, the load forecasting module predicts the load for multiple future time steps based on historical load data and time feature information. In this embodiment, the load forecasting module employs the Informer time series forecasting model, whose overall structure includes an embedding module, an encoder module, and a decoder module. The embedding module converts historical power and time features into high-dimensional feature representations and concatenates them to enhance the model's ability to express load change trends and time periodicity. The encoder module models the historical load sequence through a multi-layer attention mechanism to learn the changing patterns and long-term dependencies of load data over a long period. Under the constraints of the feature representation output by the encoder, the decoder module combines future time feature information to predict the load values ​​for multiple future time steps. The load forecasting module uses a multi-step forecasting approach, outputting the predicted load sequence for multiple consecutive future time steps at once, providing a reference for the subsequent demand control module regarding future load change trends.

[0044] During system operation, the load forecasting module compares the real-time collected actual load values ​​with the corresponding predicted load values ​​to calculate the single-step forecasting error. This single-step forecasting error characterizes the model's prediction deviation at the current moment and serves as a crucial basis for judging the correction of the prediction results. In this embodiment, the system pre-sets prediction correction trigger conditions. When the single-step forecasting error meets these conditions, the prediction correction mechanism is triggered. For example, the trigger condition could be: the single-step forecasting error continuously exceeds a preset threshold within two consecutive monitoring times. When the prediction correction mechanism is triggered, the load forecasting module recalculates the predicted load values ​​for subsequent time steps by calling the load forecasting model and updates the corrected prediction results to the demand control module, thereby reducing the accumulation of prediction errors over time. For example, if the original predicted time period was (…), the prediction error would be calculated as follows: (…). , , ... If to The time error reaches the correction condition. At this point, the model will re-predict and generate ( , , ... The data from the previous step was used to initiate control, and new data was subsequently adopted.

[0045] The model monitoring and update module process is as follows: Figure 4 As shown, the model monitoring and update module uses a daily evaluation cycle to statistically analyze daily prediction errors and comprehensively evaluate prediction errors over several consecutive cycles to determine if model performance has declined. The module then determines whether the model update trigger conditions are met based on the prediction error evaluation results. In this embodiment, when the prediction error exceeds a preset error threshold over multiple consecutive evaluation cycles, it is determined that the current load prediction model can no longer effectively adapt to changes in load characteristics, thus triggering the model update process. Once the model update process is triggered, the module retrieves the latest historical load data from the database and retrains or updates the parameters of the load prediction model. After model training is complete, the updated model undergoes performance testing. When the model performance meets preset requirements, the model update is considered successful. The updated load prediction model is sent to the cloud via a communication interface and then distributed by the cloud to the corresponding energy management system to replace the original load prediction model. This centralized model management and distribution mechanism achieves consistent updates of load prediction models across multiple energy management systems, reducing manual maintenance costs.

[0046] The above-mentioned processes of prediction error assessment, model update triggering, model retraining, and model deployment constitute an automatic monitoring and update closed-loop mechanism for the load forecasting model. This enables the load forecasting model to continuously and adaptively adjust itself as load characteristics change, thereby improving the robustness and predictive reliability of the system during long-term operation.

[0047] In this embodiment, the battery state assessment module is used to evaluate the actual total available capacity and state of health (SOH) of the batteries in the energy storage system, providing a basis for the operation control, lifespan management, and scheduling decisions of the energy storage system. Unlike real-time online models such as load forecasting, changes in battery capacity and state of health have a slow evolutionary characteristic, typically only showing significant changes after undergoing deep charge-discharge cycles. Therefore, in this embodiment, the battery state assessment module adopts an intermittently triggered, on-demand assessment mechanism, performing capacity assessment calculations only when preset trigger conditions are met, rather than performing continuous online predictions.

[0048] The battery status assessment module process is as follows: Figure 5 As shown: Data Acquisition and Fusion: Battery operation data, including timestamps (ts) and the corresponding state of charge (SOC), is obtained from the Battery Management System (BMS) data table; discharge-related data, including timestamps (ts) and cumulative discharge capacity (discharge), is obtained from the Power Conversion System (PCS) data table. Since the BMS and PCS are different hardware systems, their data sampling frequencies and time bases may differ. Therefore, in this embodiment, data from different data sources is aligned based on timestamps to ensure the accuracy of subsequent feature calculations and model inference.

[0049] Detecting historical discharge events: By detecting historical discharge events, the system can collect important data related to battery degradation, such as the number of cycles, providing a basis for subsequent feature construction. In this embodiment, considering the actual situation of the project, it is defined that a valid historical discharge event is defined as one in which the battery SOC change at the start of the event is greater than or equal to 70% and the discharge time occurs between 18:00 and 22:00, and this event is included in subsequent analysis.

[0050] Constructing historical discharge characteristics: For each detected historical discharge event, this embodiment extracts and constructs multi-dimensional battery degradation characteristics, including: Cycle count—the cumulative number of charge-discharge cycles the battery has undergone, used to reflect the degree of battery cycle aging; Operating days—the number of days the battery has been in operation since it was put into operation, used to characterize the time-related calendar aging effect; ΔSOC—the change in SOC during a single discharge, used to reflect the depth of discharge; Degraded capacity—calculated based on the change in discharge SOC and the actual discharge capacity on the PCS side, reflecting the capacity loss of the battery during use; Actual capacity—calculated by subtracting the degraded capacity from the rated capacity, representing the currently available effective capacity.

[0051] Model Construction: In this embodiment, considering that this module adopts an instantaneous computation and evaluation mechanism, the model uses a multiple linear regression model that is easy to train and infer. Capacity estimation is performed. A sample weighting strategy based on battery cycle count is introduced, and weighted least squares (WLS) is used to solve the model parameters to enhance the model's ability to fit the characteristics of battery degradation in later stages. The input features of the regression model include: battery cycle count; battery operating days; and SOC change (ΔSOC). During capacity estimation, ΔSOC is set to a predetermined full discharge value (e.g., 100%) to calculate the corresponding actual total battery capacity. The regression model is constructed and solved in real-time based on the latest historical discharge data each time the deep discharge trigger condition is met, without long-term storage of model parameters, thus avoiding model failure due to changes in battery operating conditions.

[0052] In this embodiment, after training the model based on historical running data, the following set of fitting results were obtained:

[0053] in: The intercept term a = 344.8; The coefficient b = -0.045 corresponds to the number of iterations; The coefficient c = -0.008 corresponds to the number of operating days; The coefficient corresponding to the change in SOC (depth of discharge) is d = -0.12.

[0054] The above results indicate that as the number of battery cycles, operating time, and depth of discharge increase, the actual battery capacity shows a downward trend, which is consistent with the actual operating law of energy storage battery capacity decay.

[0055] After obtaining the current actual total capacity of the battery, this embodiment further calculates the battery's state of health (SOH), which is calculated as follows:

[0056] in, The actual total battery capacity is obtained through regression model evaluation. This refers to the battery's rated total capacity. SOH (State of Harm) is usually expressed as a percentage and is used to quantify the degree of degradation of the battery relative to its initial state. Generally, if the SOH is below 80%, the battery should be replaced.

[0057] In this embodiment, the demand control module is used to generate a discharge control strategy for the energy storage system by utilizing multi-step load forecast results, actual battery capacity, and SOC information, while meeting the constraints of power safety and battery operation. The control strategy is then sent to the energy storage converter for execution, thereby achieving effective control of power demand during the target period.

[0058] The flow of the demand control module is as follows: Figure 6 As shown: Load forecast sequence: Obtain the multi-step load forecast sequence output by the load forecast module within the target control period. ,in, This indicates the number of prediction steps within the target time period. Indicates the first The predicted load value of the step, The time interval between individual prediction time steps.

[0059] Dischargeable capacity calculation: Obtain the actual total usable capacity of the battery from the battery state assessment module. The current state of charge (SOC) of the battery is obtained from the battery management system. Based on the current battery SOC, the discharge capacity of the battery is calculated without violating safety operating constraints. The calculation formula is as follows:

[0060] Peak reduction threshold calculation: By setting the geometric area of ​​the peak reduction equal to the battery's discharge capacity, and using a bisection method to iteratively solve the following constraints:

[0061] in, This represents the peak reduction threshold within the target time period. This indicates the number of prediction steps within the target time period. Indicates the first The predicted load value of the step, The time interval between individual prediction time steps, This is the rated power for energy storage discharge.

[0062] Demand control logic: In this embodiment, the demand control module dynamically calculates the real-time discharge power of the energy storage system based on the relationship between real-time load power and peak shaving threshold, and sends the calculation to the energy storage converter for execution. The formula for calculating the real-time discharge power of the energy storage system is as follows:

[0063] in, For real-time load power, This refers to the real-time discharge power of the energy storage system. The demand control module will calculate the real-time discharge power. As a control command, the power is sent to the energy storage converter (PCS) to control the energy storage system to output the corresponding discharge power to offset the load power exceeding the threshold, thereby limiting the equivalent power consumption on the user side to within the peak shaving threshold.

[0064] In this embodiment, the demand control module not only generates a discharge control strategy based on the initial multi-step load forecast results, but also dynamically updates the peak shaving threshold when the load forecast results are corrected, so as to improve the accuracy and robustness of demand control.

[0065] During the control period, the load forecasting module monitors the single-step forecasting error in real time. When the forecasting error meets the preset correction conditions, the load forecasting module outputs the corrected load forecasting result and sends the corrected value to the demand control module. After receiving the corrected load forecasting result, the demand control module recalculates the peak shaving threshold based on the updated forecasted power sequence and the remaining available capacity of the battery. Subsequently, following the same demand control logic as the initial control strategy, it dynamically adjusts the discharge power of the energy storage system, thereby achieving real-time optimization of the demand control strategy.

[0066] In this embodiment, the report display module presents the economic benefits generated by the energy storage system's participation in peak shaving and demand control to users via a web interface. Specifically, it statistically analyzes peak-valley arbitrage and demand reduction benefits over different time periods and displays the trends and proportions of these benefits in chart form. Simultaneously, the report display module records and displays demand control events to reflect the actual effectiveness of the energy storage system in reducing maximum demand.

[0067] By implementing this invention, the maximum demand on the user side can be precisely controlled, taking into full account the fluctuating characteristics of building electricity load and the actual available capacity of energy storage batteries. This effectively reduces demand-based electricity pricing costs while also balancing the electricity price benefits from peak shaving and valley filling, thus improving the overall operational economy of the energy storage system. Furthermore, by introducing a monitoring and updating mechanism for the predictive model, the model can continuously adapt to changes in load characteristics, enhancing the stability and reliability of the demand control strategy during long-term operation.

[0068] In summary, this invention achieves intelligent and dynamic management of user-side energy storage systems by integrating load forecasting, battery status assessment, and demand control strategies. It provides electricity users with a cost optimization solution that balances economy and safety, and has good engineering application value and promotion prospects.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for controlling energy storage demand based on load forecasting and battery state assessment, characterized in that, include: Step 1: Data collection, including power data, battery status data, and corresponding timestamp information from the energy storage system; Step 2, Data Preprocessing and Load Forecasting: The collected data is preprocessed by the preprocessing module to obtain feature tensors. Then, the trained load forecasting model generates a multi-step load forecasting sequence for the target period based on the feature tensors, and the single-step forecasting error is monitored in real time. Step 3: Model monitoring and updating: The prediction error of the multi-step load forecasting results is evaluated periodically, and the load forecasting model is updated when preset conditions are met; Step 4: Battery Status Assessment: Estimate the actual capacity of the battery based on historical battery operating data, and assess the battery's health status based on the actual capacity. Step 5, Demand Control: Generate a charging and discharging control strategy for the energy storage system based on the predicted value at each moment in the multi-step load forecast sequence and the actual battery capacity at the corresponding moment, and control the energy storage converter to execute the charging and discharging control strategy.

2. The energy storage demand control method based on load forecasting and battery state assessment according to claim 1, characterized in that, In step one, the Hadoop distributed data processing architecture is used for data acquisition, and power data and battery status data are processed in batches and in real time in terms of time dimension.

3. The energy storage demand control method based on load forecasting and battery state assessment according to claim 1, characterized in that, The trained load prediction model is a time-series-based prediction model. The time-series-based prediction model uses feature tensors to predict load values ​​at multiple future times, forming a multi-step load prediction sequence. The method for real-time monitoring of single-step prediction errors is as follows: the predicted load value is compared with the actual load value to calculate the error. When the error reaches a preset correction condition, the original multi-step load prediction sequence is discarded, and the input feature tensor is updated again based on the actual detected data to generate a new load prediction sequence. The time-series-based prediction model includes the Informer time-series prediction model.

4. The energy storage demand control method based on load forecasting and battery state assessment according to claim 1, characterized in that, The steps for preprocessing the collected data include: Outlier handling: Perform negative value detection, box plot detection, and moving mean detection on the original data to remove data that exceeds the reasonable physical range and has an abnormal distribution; Missing value imputation: Time-based imputation and forward / backward imputation are used to fill in missing data to ensure the continuity and integrity of the data time series; Data normalization: The Z-Score normalization method based on statistical features is used to standardize the power data, improving the model convergence speed and inference efficiency. The formula is as follows: ； in It is the mean of the original data. It is the standard deviation of the original data. It is the raw data. It is normalized data; Time feature encoding: Parse the timestamp and extract the month, day, week, hour, and minute features divided according to preset time intervals; Sliding time window algorithm: A feature tensor is constructed using the sliding time window algorithm. The feature tensor includes an input feature tensor and an output feature tensor. The input feature tensor and the output feature tensor form a sample pair.

5. The energy storage demand control method based on load forecasting and battery state assessment according to claim 1, characterized in that, The specific steps of step three are as follows: The performance of the load forecasting model is judged based on the prediction error of the trained load forecasting model. When the total energy consumption error and MAPE both exceed 5% for 3 out of 5 consecutive days, the parameters of the load forecasting model are updated. The updated load forecasting model is obtained by retraining using the original training data and the data collected during the operation of the trained load forecasting model as the training set, and then the updated load forecasting model is deployed to the load forecasting module.

6. The energy storage demand control method based on load forecasting and battery state assessment according to claim 1, characterized in that, The specific steps of step four are as follows: Identify battery discharge events from historical battery operating data, and then use a regression model. By fitting the capacity trend changes in historical discharge events, the actual capacity of the battery can be predicted. Finally, the battery health status is calculated according to the following formula: ； Among them, SOH represents the battery health status. The actual total battery capacity is obtained through regression model evaluation. This refers to the battery's rated total capacity; where... The actual capacity of the battery , This refers to the number of battery cycles. This refers to the number of days the battery operates. State of Charge (SOC) is the change in SOC during battery discharge, defined as the difference between the initial SOC and the final SOC, and is used to characterize the impact of battery depth of discharge on capacity decay. , , and All of these are the fitting coefficients to be fitted.

7. The energy storage demand control method based on load forecasting and battery state assessment according to claim 1, characterized in that, The specific steps of step five are as follows: First, a multi-step load prediction sequence is obtained based on the trained load prediction module. Then, the discharge capacity is calculated using the actual battery capacity output by the battery state assessment module. Next, the peak shaving threshold is solved using the multi-step load prediction sequence and the discharge capacity. Finally, the real-time energy storage discharge power is obtained based on the real-time load power and the peak shaving threshold.

8. The energy storage demand control system based on load forecasting and battery state assessment according to claim 7, characterized in that, Battery discharge capacity The calculation formula is as follows: ； in Indicates the battery's discharge capacity. Indicates the actual capacity of the battery. Indicates the battery's state of charge; Peak Shaving Threshold The solution is obtained by optimization using the following constraints: ； The peak reduction threshold is solved iteratively using the bisection method. ,in This indicates the number of prediction steps within the target time period. Indicates the first The predicted load value of the step, The time interval between individual prediction time steps, The rated power for energy storage discharge; The real-time discharge power of the energy storage is: ； For real-time load power, This refers to the real-time discharge power of the energy storage system.

9. An energy storage demand control system based on load forecasting and battery state assessment, used to execute the method according to any one of claims 1-8, characterized in that, include: Data acquisition module: used to collect data, including power data, battery status data and corresponding timestamp information in the energy storage system; Load forecasting module: The load forecasting module includes a preprocessing module and a load forecasting model; Model monitoring and update module: periodically evaluates the prediction error of the multi-step load prediction results and updates the load prediction model when preset conditions are met; Battery status assessment module: Estimates the actual capacity of the battery based on historical battery operating data, and assesses the health status of the battery based on the actual capacity; Demand control module: Generates a charging and discharging control strategy for the energy storage system based on the load forecast results for the target period and the actual battery capacity, and controls the energy storage converter to execute the charging and discharging control strategy; Report display module: Provides a visual representation of system operating status, prediction results, and control information; During the operation of the energy storage system, when the demand control module receives the corrected load forecast result, it updates the peak shaving threshold based on the current remaining power of the energy storage battery.

10. The energy storage demand control system based on load forecasting and battery state assessment according to claim 9, characterized in that, The report display module displays the system's operating status information, load forecasting effect, battery health status, economic benefit analysis, and alarm logs, and supports generating operation analysis reports according to the time dimension.

Images