A load balancing control method and system for energy storage device

Abstract

The present application discloses a load balancing control method and system for an energy storage device, and relates to the technical field of energy storage load regulation. The method first obtains the historical operating data of the target energy storage device, and pre-processes it through the first Gaussian processing method to generate the first operating data. Subsequently, the first modal decomposition method is used to further process these data to decompose the multi-scale time series components, thereby generating the second operating data. These data are further input into a hybrid neural network, which includes two parts: the first sub-network extracts the spatial features of the data, and the second sub-network performs time dependency analysis. Based on the first load control information generated by the network, a control strategy for the target energy storage device is formulated. The system can also collect environmental data in real time, and adjust the noise parameters based on these data to optimize data processing and prediction accuracy.

Description

Technical Field

[0001] The present application relates to the technical field of energy storage load regulation, and in particular to a load balancing control method and system for an energy storage device. Background Art

[0002] With the rapid development of renewable energy and the continuous expansion of power systems, energy storage technology has become a key factor in achieving sustainable energy development. Energy storage systems, especially battery energy storage systems, play an important role in balancing supply and demand, improving energy utilization efficiency and enhancing system stability in the power grid. However, due to the intermittent and unpredictable nature of renewable energy sources such as wind and solar energy, energy storage devices face complex load regulation challenges. Traditional load balancing control methods for energy storage devices mainly rely on preset control strategies and simple data processing techniques. These methods are usually unable to make full use of historical and real-time data to optimize load regulation decisions. In addition, existing control systems often lack in-depth analysis of the multi-dimensional characteristics of data, resulting in poor results when dealing with complex environmental changes and load demands.

[0003] For example, a Chinese patent with authorization announcement number CN114362125B discloses an energy storage power supply system and a control method thereof, wherein the energy storage power supply system includes a mains power network, N load power supply devices, each load power supply device is connected to the mains power network via an AC power supply network, and each load power supply device includes an energy storage module, a power supply conversion module, and a control module. The control modules of each load power supply device communicate with each other, can receive relevant power parameters of each load power supply device, and perform judgment and calculation on power regulation based on the relevant power parameters of all the load power supply devices. When it is detected according to the judgment and calculation results that the load power supply device corresponding to the control module needs to transfer power to the load power supply device with insufficient power, the energy storage module is controlled to output power to the AC power supply network, so as to supply energy to the load power supply device with insufficient power through the AC power supply network, thereby realizing the sharing of energy storage power of each load power supply device.

[0004] The above methods have the problems raised by this background technology. Therefore, there is an urgent need to develop a new load balancing control method for energy storage devices, which needs to be able to more comprehensively utilize historical and real-time data to optimize load control strategies to cope with the instability of renewable energy supply and other environmental challenges. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of this application is to provide a load balancing control method and system for an energy storage device, which can make full use of historical and real-time operation data, and improve the load control accuracy and response speed of the energy storage system through Gaussian processing, modal decomposition and hybrid neural network. The specific scheme is as follows:

[0006] In a first aspect, the present application provides a load balancing control method for an energy storage device, comprising:

[0007] Acquire historical operation data of the target energy storage device, and perform preprocessing based on a first Gaussian processing method to generate first operation data;

[0008] Based on a first modal decomposition method, processing the first operation data, decomposing multi-scale time series components in the first operation data, and generating second operation data;

[0009] The second operating data is input into a hybrid neural network for processing, and the first load control information is output; based on the first load control information, a control method for a target energy storage device is determined; wherein the hybrid neural network includes: using a first sub-network to extract spatial features of the second operating data and sending the spatial features to a second sub-network, and the second sub-network is used for time dependency analysis.

[0010] As a possible implementation manner, the historical operation data includes: energy data, environmental data, equipment operation data and historical load data; the preprocessing based on the first Gaussian processing method to generate the first operation data includes:

[0011] The functional expression of the first Gaussian processing method is:

[0012]

[0013] in, Indicates the current processing time point. represents each measurement time point in the historical operation data, It's at the time The dynamically adjusted smooth width, Based on time point The asymmetric factor of the dynamic adjustment of data characteristics in the first time period before and after, is a symbolic function, according to and The relative position is returned as +1 or -1.

[0014] As a possible implementation, the dynamically adjusted smoothing width The function expression is:

[0015]

[0016] in, It's in time The peak frequency of the data observed in the nearby second time period, is the threshold used to adjust the sensitivity of the smoothing width adjustment, is the adjustment coefficient, which is used to control the adjustment amplitude of the variance on the smoothing width. It's in time The data variance in the third time period nearby is used to indicate the fluctuation of the data.

[0017] As a possible implementation manner, the first Gaussian processing method further includes:

[0018] Based on the influence of historical data, the function expression of the first Gaussian processing method is adjusted to:

[0019]

[0020] in, is the functional expression of the first Gaussian processing method before adjustment, is a weighting factor used to adjust the impact of historical correlation, yes and historical time The data correlation between It's a time delay.

[0021] As a possible implementation manner, the processing the first operating data based on the first modal decomposition method, decomposing the multi-scale time series components in the first operating data, and generating the second operating data includes:

[0022] Decomposing the first operating data using an empirical mode decomposition method, extracting a first intrinsic mode function of the current sequence therefrom, and subtracting the extracted intrinsic mode function from the sequence;

[0023] Repeat the empirical mode decomposition for the remaining data after the first decomposition, and extract the first intrinsic mode function of the current sequence each time;

[0024] In response to the remaining data being unable to further perform the empirical mode decomposition, stopping the decomposition and generating a plurality of intrinsic mode functions and a residual term;

[0025] The plurality of intrinsic mode functions and residual terms are used as the second operating data.

[0026] As a possible implementation manner, the first modal decomposition method further includes:

[0027] Acquiring real-time environmental data of the target energy storage device;

[0028] A first target noise is added to the first operating data based on the real-time environmental data; wherein the first target noise is used to simulate the actual impact of environmental factors on the operation of the target energy storage device.

[0029] As a possible implementation manner, the adding a first target noise to the first operation data based on the real-time environment data includes:

[0030] Based on the real-time environmental data, determining the intensity of the first target noise;

[0031] The real-time environmental data includes humidity data and temperature data; the function expression of the intensity of the first target noise is:

[0032]

[0033] in, is the basic noise intensity, and are the adjustment coefficients for the effects of temperature and humidity on noise, and are the reference values ​​of temperature and humidity respectively, and They are real-time measurements of temperature and humidity respectively;

[0034] Based on the real-time environmental data, statistical characteristics of the first target noise are determined.

[0035] As a possible implementation manner, before inputting the second operation data into the hybrid neural network for processing and outputting the first load regulation information, the method further includes:

[0036] adding a second target noise to the second operation data based on the real-time environmental data;

[0037] Wherein, the second target noise includes:

[0038] Based on the sliding window method, determining the change trend of the real-time environmental data;

[0039] Based on the change trend of the real-time environmental data, a mean value and a variance of the second target noise are calculated.

[0040] In a second aspect, the present application also provides a load balancing control system for an energy storage device, comprising:

[0041] An acquisition unit, used for acquiring historical operation data of a target energy storage device, and performing preprocessing based on a first Gaussian processing method to generate first operation data;

[0042] A processing unit, based on a first modal decomposition method, processes the first operation data, decomposes multi-scale time series components in the first operation data, and generates second operation data;

[0043] The decision unit inputs the second operating data into the hybrid neural network for processing and outputs the first load control information; based on the first load control information, determines the control method for the target energy storage device; wherein the hybrid neural network includes: using the first sub-network to extract the spatial features of the second operating data and sending it to the second sub-network, and the second sub-network is used for time dependency analysis.

[0044] In a third aspect, an optional implementation of the present application further provides an electronic device, a processor, and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory, and when the machine-readable instructions are executed by the processor, the machine-readable instructions perform the steps of the above-mentioned first aspect, or any possible implementation of the first aspect.

[0045] In a fourth aspect, an optional implementation of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed, the steps of the above-mentioned first aspect, or any possible implementation of the first aspect are executed.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. Use the modified Gaussian smoothing technology to effectively reduce random noise and outliers in the data, making subsequent modal decomposition and neural network analysis more accurate and improving the prediction and control accuracy of the entire system.

[0048] 2. Target noise is added based on real-time environmental data at different stages. The first noise addition simulates the actual impact of environmental factors so that modal decomposition and subsequent neural network analysis can more accurately predict system response. The second noise addition helps the hybrid neural network better capture data fluctuations caused by environmental changes and optimize the model's time-dependent analysis capabilities.

[0049] 3. Use the empirical mode decomposition method to further process the data and decompose the multi-scale time series components, so that the energy storage system can identify and utilize the characteristics of different time scales in the data to effectively predict and control the load.

[0050] 4. The hybrid neural network used in this method not only extracts the spatial features of the data, but also performs temporal dependency analysis. This hybrid structure makes full use of spatial and temporal data, improving the accuracy of prediction and the adaptability of the control strategy.

[0051] Through the comprehensive application of these technologies, this application makes full use of historical and real-time data to optimize load regulation decisions, so as to improve the load balancing regulation capability of the battery energy storage system under the intermittent and unpredictable nature of renewable energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flow chart of a load balancing control method for an energy storage device provided in an embodiment of the present application;

[0053] Figure 2 A flowchart of adding noise to historical operation data provided in an embodiment of the present application;

[0054] Figure 3 A schematic diagram of the structure of a load balancing control system for an energy storage device provided in an embodiment of the present application;

[0055] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0057] The present application collects historical operating data of the energy storage device, and uses the first Gaussian processing method to process these data to generate first operating data. Then, the empirical mode decomposition method is used to process these data, decompose the multi-scale time series components, and generate second operating data. In addition, the system also collects real-time environmental data, such as temperature and humidity, and adds target noise to the first and second operating data based on these data. These noises are intended to simulate the real impact of environmental factors on the operation of the device.

[0058] The hybrid neural network receives the noise-adjusted second operating data and performs in-depth analysis. The network is divided into two parts: the first part extracts the spatial features of the data, and the second part is responsible for time-dependency analysis to generate comprehensive first load regulation information. This information provides a basis for the operation of the energy storage system, including the best time to charge and discharge, and how to distribute energy according to load demand.

[0059] See also Figure 1 As shown, it is a flow chart of a load balancing control method of an energy storage device provided in an embodiment of the present application, the method includes steps S101 to S103, wherein:

[0060] S101: Acquire historical operation data of a target energy storage device, and perform preprocessing based on a first Gaussian processing method to generate first operation data;

[0061] S102: Based on a first modal decomposition method, process the first operating data, decompose multi-scale time series components in the first operating data, and generate second operating data;

[0062] S103: Input the second operating data into the hybrid neural network for processing, and output the first load control information; based on the first load control information, determine the control method for the target energy storage device; wherein the hybrid neural network includes: using the first sub-network to extract the spatial features of the second operating data and sending it to the second sub-network, and the second sub-network is used for time dependency analysis.

[0063] In the specific implementation, the application first obtains the historical operation data of the target energy storage device, including energy data, environmental data, equipment operation data and historical load data, which also includes the relationship between input and output current and voltage. These data are pre-processed by the first Gaussian processing method to smooth the data and reduce noise, thereby facilitating more in-depth data analysis.

[0064] The preprocessed data will be sent to the first mode decomposition method, which can effectively decompose multi-scale time series data and identify and extract the intrinsic mode functions (IMFs). This step is to extract key features from complex time series and provide a clearer data perspective for subsequent analysis.

[0065] The second operating data obtained through modal decomposition is input into a hybrid neural network for in-depth analysis. The network includes a subnetwork for extracting spatial features and a subnetwork for time-dependency analysis. The output of the network is the first load control information, which will be used to formulate the optimal control strategy for the target energy storage device.

[0066] Among them, Gaussian processing can effectively reduce random fluctuations and outliers in the data, making subsequent analysis more stable and accurate; through modal decomposition, data can be analyzed hierarchically to identify behaviors and patterns at different time scales, which is very important for the response design of regulatory systems; the hybrid network architecture that combines spatial and temporal analysis can more comprehensively understand data characteristics and improve the accuracy of predictions and decisions, especially in dynamic and complex system operating environments.

[0067] For example, an energy storage system needs to optimize the charge and discharge cycle of its batteries to support the energy needs of an industrial park. First, the system collects daily battery performance data and energy consumption records of the park over the past year. These data are preprocessed by Gaussian smoothing method to remove outliers and smooth short-term fluctuations.

[0068] Next, these smoothed data are processed using empirical mode decomposition to extract the main energy consumption patterns and possible abnormal consumption patterns. The decomposition results are displayed in different IMFs, each of which represents a specific consumption frequency.

[0069] Finally, these data are input into a designed hybrid neural network, which predicts future energy demand and possible load peaks by analyzing spatial and temporal characteristics. Based on the control information output by the network, the system automatically adjusts the charging and discharging strategy to optimize energy utilization and cost efficiency.

[0070] In this way, the accuracy of energy management is improved and energy waste is reduced.

[0071] With respect to the above S101: acquiring historical operation data of the target energy storage device, and performing preprocessing based on a first Gaussian processing method to generate first operation data;

[0072] As a possible implementation, the historical operation data may further include: energy data, environmental data, equipment operation data, and historical load data; when preprocessing is performed based on the first Gaussian processing method to generate the first operation data, the function expression of the first Gaussian processing method is specifically:

[0073]

[0074] in, Indicates the time point of the current processing, which is selected dynamically, depending on the data characteristics that need to be analyzed. is a time point in the data sequence, representing each measurement time point in the historical operation data, Indicates at a point in time The dynamically adjusted smoothing width determines the width of the Gaussian function, thus affecting the local sensitivity of the smoothing process. It is an asymmetric factor that is dynamically adjusted based on data characteristics, which increases the flexibility of processing specific data points. This parameter allows the model to The change characteristics of the data in the first time period before and after adjust the asymmetry of the smoothing, is a symbolic function, according to and The relative position of the function is returned as +1 or -1, which is used to adjust the Gaussian function in The shape of the front and back is adapted to the asymmetry of the data. The specific selection of the first time period depends on the rate of change of the data, the noise level, and the feature scale that needs to be captured.

[0075] For example, you can select a time period based on the update frequency and change characteristics of the data. For rapidly changing data, select a shorter time period to more flexibly reflect immediate data changes; for slowly changing data, select a longer time period to smooth excessive fluctuations. You can also evaluate the noise level. For high-noise data, select a shorter time period for more frequent smoothing to reduce the impact of noise, while for low-noise data, select a longer time period to retain more signal details.

[0076] In specific implementation, the first Gaussian processing design of the present application aims to optimize the analysis effect of the energy storage system on historical operation data, and specifically makes targeted modifications to the nonlinear and asymmetric characteristics of the data.

[0077] Among them, the traditional Gaussian smoothing adopts a fixed width, while the width in this method is dynamically adjusted according to the data characteristics at each time point, making the processing more flexible and able to adapt to the change amplitude of the data at different time points.

[0078] In addition, in traditional Gaussian functions, smoothing is usually symmetrical. After introducing the asymmetric factor, the slope of the smoothing function is adjusted according to the actual fluctuation direction (increase or decrease) of the historical data, which is very important when processing data with obvious trend changes such as power load.

[0079] For example, at a specific sunrise time , the corresponding real-time temperature and humidity data show a significant change trend. At this time, the following parameters can be set for calculation:

[0080] =0.3, =30 minutes, =6:00 AM.

[0081] Based on these parameters, we can calculate the The corresponding Gaussian function value is used to further smooth the data. In this way, the energy output data in the morning time period is smoothed more accurately, revealing the true pattern of energy production.

[0082] In this way, the improved Gaussian smoothing method can more accurately capture the increasing trend of solar energy output after sunrise, and optimize the charging time and charging strategy of the energy storage device accordingly. This not only improves energy utilization efficiency, but also reduces energy waste.

[0083] For example, an industrial energy storage system is used to buffer sudden peaks in electricity demand. In this system, electricity consumption often has significant peaks during the morning and evening peak hours on weekdays, while it is relatively stable during other periods. One of the weekdays is selected for data processing, and the asymmetric factor is dynamically adjusted using the first Gaussian processing method. , to accurately capture and predict peak power consumption.

[0084] For example, determine peak hours 8:00 AM and 6:00 PM. When it is 8:00 AM, set Since electricity demand rises rapidly in the morning on weekdays, it is necessary to emphasize this rising trend by increasing the asymmetry factor.

[0085] In the evening rush hour When it is 6:00 PM, set This reflects the rapid decline in electricity demand after the evening peak, and this decline is highlighted by adjusting the asymmetry factor to a negative value.

[0086] In this way, by dynamically adjusting , the Gaussian smoothing shape can be made to better fit the actual changes in electricity demand during the morning and evening peaks. This adjustment allows the model to more sensitively capture the increase in demand before the peak and highlight the decrease in demand after the peak, thus providing more accurate data analysis results:

[0087] During the morning peak period, smoothing emphasizes the rise in demand, allowing the model to predict the start and end times of the demand peak, assisting decision makers in adjusting the charging or discharging strategy of the energy storage device in advance.

[0088] During the evening peak period, smoothing emphasizes the drop in demand through a negative asymmetry factor, helping the system optimize energy distribution and prevent overcharging or excessive call on storage resources.

[0089] As a possible implementation method, the width of Gaussian smoothing can be adjusted according to the data characteristics monitored in real time. , the specific function expression is:

[0090]

[0091] in, It's at the time The peak frequency of the data observed in the second time period nearby is used to reflect the main change trend of the data at that time point. is the threshold used to adjust the sensitivity of the smoothing width adjustment, setting a baseline to determine when a significant adjustment to the smoothing width is needed. is the adjustment coefficient, which is used to control the adjustment amplitude of the variance on the smoothing width. It's in time The data variance in the third time period nearby is used to indicate the fluctuation of the data.

[0092] The selection of the second time period and the third time period may refer to the selection method of the first time period, which will not be described in detail here.

[0093] In the specific implementation, this application sets up multiple monitoring points in the system and updates them regularly. , and record the corresponding data volatility , these monitoring points should be selected during time periods when data changes are sensitive, such as peak or valley energy consumption. and , these parameters can be optimized in an automated way using algorithms based on the performance of historical data. and After the data is obtained, the new smoothing width is calculated using the updated parameters .

[0094] This application adopts the method of adjusting the smoothing width based on the peak frequency of the data by Directly used to adjust the smoothing width, so that the smoothing process can flexibly adapt to the actual changes in the data, especially when the data shows significant periodicity or volatility. Dynamic adjustment of the smoothing width can better balance the detail preservation and noise reduction of the data, and improve the accuracy and effectiveness of data processing, especially in scenarios that need to respond quickly to external changes.

[0095] For example, energy storage systems can also be used to optimize market operations and price responses. During specific demand peaks, fluctuations in electricity prices can be significant, so accurate data processing methods are needed to predict and respond to these changes.

[0096] Select a peak demand period during the day, such as the high demand period in the afternoon, set to 3:00 PM. During this time period, market prices may change rapidly due to a sharp increase in demand. Observed data peak frequency For example, 3 times per hour.

[0097] Setting the Threshold = 2 times / hour, used to determine when significant adjustments to the smoothing width are required.

[0098] Setting the adjustment factor =1.0, adjust the smoothing width according to the volatility of the data.

[0099] Detecting data variance =0.05, for example, the price fluctuation variance at 3:00 PM is higher, reflecting the instability of prices.

[0100] Calculate the smoothing width for a specific period of time :

[0101]

[0102] Calculated It will be used to adjust the width of Gaussian smoothing to make data processing more adaptable to the actual price fluctuations.

[0103] As an optional implementation, historical correlation (lagcorrelation) can be introduced into the first Gaussian processing method as a new factor to adjust the shape of the Gaussian smoothing function. By considering the historical correlation in the time series data, the Gaussian smoothing method is optimized to more accurately reflect the time-dependent characteristics of the data. The function expression of the adjusted first Gaussian processing method is:

[0104]

[0105] in, is a weighting factor used to adjust the impact of historical correlation, is based on the Gaussian smoothing function defined above, yes and historical time The data correlation between is the lag time.

[0106] In the specific implementation, the lag time It is the time span that measures the impact of historical data on current forecasts.

[0107] Autocorrelation analysis can be performed to identify significant periodicity in the data. For example, by calculating the autocorrelation of hourly electricity consumption data, finding significant lag times, such as 24-hour cycles, and analyzing intra-day and inter-day variations in electricity demand in past data to identify key historical time points that affect current electricity demand.

[0108] Adjustment Factor Controls the influence of historical correlation in Gaussian smoothing. The specific value can be changed by changing and observe the changes in the prediction results to find the optimal This can be done by simulating different This is done by looking at the impact of the value on the prediction accuracy.

[0109] Alternatively, algorithms such as genetic algorithms or gradient descent can be used to automatically adjust , the goal is to minimize the prediction error.

[0110] Lagged correlation is a statistical measure of the dependency between historical data points and the current forecast point. Correct calculation is critical to model performance. It can be calculated using a fixed-size moving window. and The correlation coefficient between .

[0111] It is also possible to use time series analysis tools in statistical packages, such as R or Python's pandas library, to automatically calculate and update lagged correlations.

[0112] For example, electricity demand is often closely correlated with historical consumption patterns at specific times (eg, holidays, weekdays, special event days).

[0113] Select peak hours during weekdays (e.g. 5:00 PM to 6:00 PM) as the key forecast time points . Set the delay time 7 days and 365 days, reflecting the effects of weekly and annual recurrence patterns, respectively. Set to 0.3 based on the analysis of the impact of historical data on model accuracy. Calculated using historical electricity consumption data and Tianhe For example, the historical correlation coefficients found were 0.6 and 0.8, respectively, indicating that electricity demand has similar consumption patterns at the same time point every week and every year.

[0114] The original Gaussian smoothing function can be adjusted according to the historical correlation results. The function expression of the adjusted first Gaussian processing method is:

[0115]

[0116] This will increase the responsiveness of the model to historical consumption patterns, making forecasts more closely aligned with actual demand fluctuations.

[0117] In this way, by introducing historical correlation to adjust Gaussian smoothing, important trends and patterns in historical data can be captured more accurately. This approach can identify and enhance important periodic features, such as daily load cycles, seasonal changes, or load fluctuations caused by specific events, thereby improving the accuracy of load forecasts. The improvement in forecast accuracy directly affects the dispatch strategy of the energy storage system, making it more effective in responding to expected load changes. Accurate load forecasting enables energy storage systems to manage energy storage and release more effectively. For example, when peak loads are predicted, the system can adjust the charging strategy of the energy storage device in advance to ensure that sufficient energy is available; similarly, during low load periods, the system can optimize the storage of electrical energy to avoid unnecessary energy waste. This optimization directly improves the economy and efficiency of energy use.

[0118] With respect to the above S102: based on the first modal decomposition method, processing the first operating data, decomposing the multi-scale time series components in the first operating data, and generating second operating data;

[0119] As an optional implementation, based on the first modal decomposition method, processing the first operating data, decomposing the multi-scale time series components in the first operating data, and generating the second operating data includes the following steps S21 to S24:

[0120] S21: Decomposing the first running data using an empirical mode decomposition method, extracting the first intrinsic mode function of the current sequence therefrom, and subtracting the extracted intrinsic mode function from the sequence;

[0121] S22: Repeat the empirical mode decomposition for the remaining data after the first decomposition, and extract the first intrinsic mode function of the current sequence each time;

[0122] S23: In response to the remaining data being unable to further perform the empirical mode decomposition, stopping the decomposition, and generating a plurality of intrinsic mode functions and a residual term;

[0123] S24: Using the plurality of intrinsic mode functions and residual terms as the second operating data.

[0124] In a specific implementation, with respect to the above S21:

[0125] The first operating data may include, but is not limited to, historical operating data of the energy storage device, such as power consumption and production data. First, the first operating data is processed using an empirical mode decomposition method. The empirical mode decomposition method is an adaptive decomposition method that can extract components of different frequencies from the data. In this process: the first operating data is selected as input data. The empirical mode decomposition process is performed to extract the first intrinsic mode function (IMF), which captures the highest frequency component in the data. The extracted IMF is subtracted from the original data to obtain the remaining data.

[0126] There are many ways to use the empirical mode decomposition method, such as through the "end" function in MATLAB, and the "PyEMD" library in Python to perform EMD and its variants such as ensemble empirical mode decomposition (EEMD). This library can be installed through Python's package manager pip. The benefit of using Python for EMD processing is that it can be easily integrated into larger data processing and analysis pipelines.

[0127] Regarding the above S22:

[0128] Repeat the EMD process for the remaining data, and continue to apply the EMD method to decompose the remaining data and extract new IMFs. Repeat this process, each time extracting the first IMF in the current sequence and subtracting it from the remaining data. This step separates and identifies the various frequency features of the data by successively extracting the IMFs of different frequency components.

[0129] Regarding S23 above:

[0130] In the continuous decomposition process, when the energy or change amplitude of the remaining data is small to a certain extent, it indicates that the main features in the data have been extracted and the remaining part cannot be further decomposed. At this time, the EMD process is stopped. Finally, a set of IMFs are generated, which represent different frequency components, and a residual term (Residue), which represents the low-frequency or trend part that cannot be further decomposed.

[0131] Regarding the above S24:

[0132] All extracted IMFs and residual terms are combined as the second operating data. These data contain the load change characteristics of different frequencies during the operation of the energy storage device:

[0133] By analyzing these multi-scale time series components, the load of energy storage devices can be predicted and regulated more accurately.

[0134] In specific applications, operators can formulate optimized energy storage control strategies based on these characteristics, such as rapid response during periods of high-frequency load changes and long-term planning during periods of low-frequency or long-term trend changes.

[0135] Exemplarily, energy storage devices are used to balance the load of the power grid and optimize its control strategy: power load data for the past month can be collected and used as first operating data after being processed by a first Gaussian processing method.

[0136] Execute the above steps S21 to S24 to perform empirical mode decomposition on the data and extract several IMFs and a residual term.

[0137] By analyzing the extracted IMFs, it is found, for example, that the first IMF represents the high-frequency changes in the load within the day, the second IMF represents the medium-frequency changes within the week, and the residual term represents the monthly trend.

[0138] Based on the analysis results, the rapid response capability of the energy storage device can be optimized during high-frequency change periods during the day (such as daytime peak hours); the charging strategy of the energy storage device can be adjusted during medium-frequency change periods during the week (such as low electricity consumption on weekends); and long-term energy storage planning can be carried out based on monthly trends to ensure sufficient energy storage capacity when the load increases in the long term.

[0139] This will allow for more accurate forecasting of power demand and optimization of energy storage device operation, thereby improving power system efficiency and reliability.

[0140] As a possible implementation method, the effect of the first modal decomposition method can be enhanced by introducing real-time environmental data and target noise. The purpose of this method is to make data processing more realistic and accurate by adding target noise that simulates the influence of environmental factors before modal decomposition, thereby improving the effect of load regulation. The method includes:

[0141] Acquiring real-time environmental data of the target energy storage device;

[0142] A first target noise is added to the first operating data based on the real-time environmental data; wherein the first target noise is used to simulate the actual impact of environmental factors on the operation of the target energy storage device.

[0143] In a specific implementation, it is necessary to first obtain real-time environmental data, such as collecting real-time data of the environment where the energy storage device is located, such as temperature, humidity, etc. Based on the real-time environmental data, the target noise is added to the first operation data. The target noise is used to simulate the impact of environmental factors on the operation of the energy storage device, so that the data is closer to the actual operation state.

[0144] Real-time environmental data can be obtained by deploying environmental sensors to monitor environmental parameters such as temperature and humidity around the energy storage device in real time.

[0145] Based on real-time environmental data, target noise is generated to simulate environmental impacts. For example, temperature changes may cause battery performance fluctuations, and humidity changes may affect the stability of electrical connections.

[0146] Noise generation can be determined based on a relationship model between historical data and environmental parameters. For example, for every 1 degree Celsius increase in temperature, the battery output efficiency may decrease by a certain percentage, and such a relationship can be used to generate target noise.

[0147] The generated target noise is superimposed on the historical operation data of the energy storage device to obtain the adjusted first operation data. This processing can introduce more realistic environmental influences into the data, so that the subsequent modal decomposition can better reflect the actual situation.

[0148] For example, a power storage system needs to optimize its load regulation strategy, taking into account the impact of environmental factors on battery performance:

[0149] First, deploy temperature and humidity sensors to monitor the environmental changes at the location of the energy storage device in real time. For example, during a high temperature period in summer, the temperature data recorded by the sensor is: 35℃, 36℃, 37℃, etc.

[0150] According to historical data, for example, the battery output efficiency decreases by 0.5% for every 1°C increase in temperature. Therefore, when the temperature increases from 35°C to 45°C, the target noise generated is a 5% reduction in battery output efficiency. What is described here is the addition of noise, and the specific method for determining the noise intensity will be described below.

[0151] The generated noise is added to the historical operation data of the energy storage device. For example, the original battery output efficiency at a certain point in time is 100 units, which is adjusted to: , the adjusted data can better reflect the actual impact of high temperature on battery performance.

[0152] The adjusted data are processed using the empirical mode decomposition method to extract multiple intrinsic mode functions and a residual term.

[0153] Analyze these intrinsic mode functions to optimize the control strategy of the energy storage device. For example, during high temperature periods, reduce the load in advance or increase heat dissipation measures to ensure stable operation of the system.

[0154] In this way, the introduction of real-time environmental data and target noise can help data processing to be closer to actual operating conditions, thereby improving the accuracy and effectiveness of load regulation.

[0155] As an optional implementation manner, the method of adding the first target noise to the first operation data based on the real-time environment data may be specifically as follows:

[0156] Determining the intensity of the first target noise based on the real-time environmental data;

[0157] The real-time environmental data includes humidity data and temperature data. The function expression of the intensity of the first target noise is:

[0158]

[0159] in, is the basic noise intensity, and are the adjustment coefficients for the effects of temperature and humidity on noise, and are the reference values ​​of temperature and humidity respectively, and They are real-time measurements of temperature and humidity respectively;

[0160] In a specific implementation, the noise intensity is dynamically adjusted through real-time temperature and humidity data, so that the impact of simulated environmental factors on the operation of the energy storage device is more accurate. First, the real-time temperature and humidity data of the environment where the target energy storage device is located are collected. Based on the real-time environmental data, the intensity of the target noise is calculated using the function expression of the intensity of the above-mentioned first target noise. The calculated target noise is added to the first operating data to simulate the actual impact of environmental factors.

[0161] Among them, real-time environmental data includes temperature data and humidity data .

[0162] In a specific implementation, the calculated target noise is superimposed on the first operation data of the energy storage device. For example, if the current ambient temperature is higher than the reference temperature, the target noise will increase to simulate the negative impact of high temperature on the operation of the device.

[0163] In this way, the adjusted data will better reflect the changes in the actual operating environment, making the subsequent modal decomposition and load control more accurate and reliable.

[0164] In a specific implementation, the basic noise intensity It is a reflection of the intensity of the basic noise contained in the operating data of the energy storage device under standard environmental conditions. The energy storage device can be operated and the output data can be recorded under standard laboratory environmental conditions (such as 25°C and 50% humidity). Through multiple measurements and statistical analysis, the data fluctuation range under standard environmental conditions is calculated to determine the basic noise intensity. Analyze historical data under standard environmental conditions and calculate the average fluctuation range of the data as the basic noise intensity reference value.

[0165] Adjustment factor and Respectively reflect the degree of influence of temperature and humidity on noise intensity. The influence of temperature and humidity changes on the fluctuation of energy storage device output data can be determined by correlation analysis of historical operation data and environmental data. For example, the percentage of battery output efficiency reduction for every 1°C increase in temperature can be analyzed to determine the temperature adjustment coefficient. The implementation method can be to gradually change the temperature and humidity in a controlled environment, record the output data fluctuations of the energy storage device, and determine the temperature and humidity adjustment coefficients through regression analysis. and .

[0166] Reference value and It is the temperature and humidity values ​​measured under standard environmental conditions, which are used to calculate the deviation of real-time environmental data from the standard environment. Usually, the reference value is set to the standard environmental conditions recommended by the laboratory or manufacturer, such as 25°C and 50% humidity. In addition, the typical environmental conditions for the operation of energy storage devices in most cases can be determined through statistical analysis of the long-term operation data of energy storage devices, which can be used as reference values.

[0167] As an optional implementation, based on the real-time environmental data, the statistical characteristics of the first target noise can also be determined, for example, the statistical characteristics such as the mean and variance of the target noise can be determined, so as to more accurately simulate the impact of environmental factors on the operation of the energy storage device.

[0168] In a specific implementation, the environmental data within a period of time may be used to calculate the average value of the noise intensity within the period of time.

[0169] Mean The calculation formula is:

[0170]

[0171] in, is the number of data points in the time period, For the The noise intensity of each data point.

[0172] The variance of the noise intensity over the same time period is calculated to measure the volatility of the noise.

[0173] variance The calculation formula is:

[0174]

[0175] Among them, the meaning of each parameter is the same as above.

[0176] Optionally, the skewness and kurtosis of the noise can also be calculated to further describe the characteristics of the noise distribution.

[0177] For example, the temperature and humidity data of each day in a week can be collected through environmental sensors, and the daily noise intensity, as well as the mean and variance of the noise can be determined based on the above method. The higher the mean of noise in high temperature and high humidity environments, the greater the impact of the environment on the energy storage system, which may lead to a decrease in battery output efficiency. The smaller the noise variance, the more stable the environmental impact, but attention still needs to be paid to the continued impact of high temperature and high humidity.

[0178] It is understandable that the time point for performing calculation and analysis here may also be every half day, or every hour, etc.

[0179] Based on the statistical characteristics of the above noise, the energy storage strategy of the energy storage device can be adjusted. For example, the charging amount can be increased during periods of low temperature and humidity (such as morning and evening) to ensure that the battery has sufficient energy reserves during high temperature periods to cope with possible high load demands.

[0180] During the hot and humid noon and afternoon hours, reduce high-power discharge operations to avoid further impact of high ambient noise on battery performance. Give priority to discharging to insensitive loads to ensure stable operation of critical loads;

[0181] Monitor temperature and humidity data in real time and adjust the charging and discharging strategy according to the latest environmental data and noise intensity. For example, when the temperature exceeds a certain threshold (such as 40°C), initiate additional cooling measures or reduce the load to reduce the thermal stress of the battery.

[0182] In addition, based on the long-term statistical characteristics of noise, a regular battery maintenance and inspection plan can be formulated to promptly detect and address battery performance degradation problems that may be caused by environmental influences.

[0183] As an optional implementation manner, before inputting the second operation data into the hybrid neural network for processing and outputting the first load regulation information, the method further includes:

[0184] adding a second target noise to the second operation data based on the real-time environmental data;

[0185] Wherein, the second target noise includes:

[0186] Based on the sliding window method, determining the change trend of the real-time environmental data;

[0187] Based on the change trend of the real-time environmental data, a mean value and a variance of the second target noise are calculated.

[0188] See also Figure 2 , is a flow chart of adding noise to historical operation data provided by an embodiment of the present application. In a specific implementation, before the second operation data is input into the hybrid neural network, noise based on real-time data can also be added. This can further enhance the dynamic performance of the data, especially for scenarios that require accurate time-dependent analysis, such as peak load prediction.

[0189] In the embodiment of the present application, there are two noise additions, which are respectively at different stages of the method disclosed in the present application. It should be noted that the purposes and effects of the two noise additions are different, and the detailed differences are as follows:

[0190] After the first Gaussian processing, the first target noise is added based on the real-time environmental data to simulate the impact of environmental factors on the actual operation of the energy storage device. It is designed to reflect environmental changes in real time and improve the adaptability and sensitivity of data preprocessing. By adding noise guided by real-time environmental data, the data can be adjusted more accurately to reflect immediate environmental impacts such as temperature fluctuations or load changes, so that modal decomposition and subsequent neural network analysis can more accurately predict system responses.

[0191] After modal decomposition, noise is added again based on real-time data to further enhance the dynamics and real-time nature of the data before it is input into the hybrid neural network. This helps the hybrid neural network better capture data fluctuations caused by environmental changes and optimizes the time-dependent analysis of the model. This step helps the system to adjust and respond to the latest environmental conditions in real time after the data processing and analysis phase, improving the response speed and accuracy of the entire load balancing control system.

[0192] The modal decomposition data may not integrate the latest environmental information due to time delay, and the input data processed by the hybrid neural network is not comprehensive enough to fully reflect the current system status and environmental conditions. This will lead to inaccurate load forecasting and control decisions based on these data, especially when environmental conditions have a greater impact on system performance.

[0193] In a specific implementation, a sliding window size (e.g., 1 hour) can be set, and at each time point, the data in the window is used to calculate the change trend of the environmental data. For each window, the change rate of temperature and humidity is calculated, which can be achieved by linear regression or difference method.

[0194] Based on the change trend in the sliding window, the mean value of the second target noise is calculated.

[0195] The calculated second target noise (mean and variance) is added to the second operation data to simulate the impact of environmental changes on system operation. Specifically, the noise value can be superimposed point by point on the corresponding point of the second operation data.

[0196] In this way, the adjusted second operation data can better reflect the real-time environmental changes, thereby making the subsequent hybrid neural network processing more accurate.

[0197] It is understandable that the specific calculation method of the statistical characteristics of the second target noise can refer to the calculation method of the first target noise, which will not be described in detail here.

[0198] With respect to the above S103: inputting the second operating data into a hybrid neural network for processing, and outputting first load control information; and determining a control method for a target energy storage device based on the first load control information;

[0199] Wherein, the hybrid neural network includes: using the first sub-network to extract the spatial features of the second operating data and sending it to the second sub-network, and the second sub-network is used for time dependency analysis.

[0200] In a specific implementation, by inputting the second operating data into a hybrid neural network for processing, load control information can be generated. Based on this control information, a control strategy suitable for the target energy storage device can be formulated. The design of the hybrid neural network includes two subnetworks: the first subnetwork is used to extract the spatial features of the data, and the second subnetwork is used to perform time dependency analysis. The purpose of this design is to combine spatial and temporal features to provide a more comprehensive and accurate load control solution.

[0201] The second operation data includes historical and real-time operation data after noise processing and modal decomposition.

[0202] In the specific implementation, a convolutional neural network (CNN) can be used as the first sub-network. CNN can effectively extract spatial features of data, such as local patterns and trends in load data. The feature map of the input data is extracted through multiple convolutional layers, and each convolutional layer can capture spatial features at different levels. The feature map is reduced in dimension using a pooling layer to retain important features while reducing the amount of computation. The spatial features extracted by the first sub-network are sent to the second sub-network. In time-dependent analysis, this step is conducive to the full utilization of the spatial features of the data.

[0203] In the specific implementation, a long short-term memory network (LSTM) can be used as the second subnetwork. LSTM is good at processing sequence data and can capture long-term dependencies and dynamic changes in time series. The spatial features of the input are processed through multiple LSTM layers, the temporal features are gradually extracted, and the load control information with time dependency is output. On the basis of the LSTM layer, a fully connected layer is used to generate the final output. The fully connected layer can convert the extracted temporal features into specific control information. The first load control information is output, including specific instructions such as the charging and discharging strategy of the energy storage device and load distribution.

[0204] In specific implementations, hybrid neural networks require some targeted design in order to effectively handle different intrinsic mode functions (IMFs) generated by empirical mode decomposition (EMD). This is because each IMF represents a specific consumption frequency and pattern in the data, and these patterns vary greatly in frequency and importance, so the network needs to be able to adapt to these differences to optimize prediction and control strategies.

[0205] For example, using convolution kernels of different sizes in the convolutional neural network branch corresponds to processing high-frequency and low-frequency IMFs, and using different time window sizes in the long short-term memory network to capture long-term and short-term time series dependencies.

[0206] Adjust the depth and width of the network to ensure that the network has enough capacity to process and integrate the information extracted from each IMF. For example, increasing the depth of the network can help capture more complex patterns, while the width of the network can enhance the network's ability to process parallel information.

[0207] The loss function can also be customized to prioritize the IMFs that are most critical to forecasting load regulation. For example, a greater weight can be given to IMFs that are particularly critical to forecasting energy demand, and a weighted loss function can be used to distinguish the contributions of different IMFs according to their importance.

[0208] In the specific implementation, a specific control method is formulated based on the load control information output by the hybrid neural network. This method takes into account the current load demand, energy storage device status and environmental factors to optimize system operation.

[0209] The generated control method is applied to the actual operation of the energy storage device to adjust the charging and discharging behavior, balance the grid load, and ensure stable and efficient operation of the system.

[0210] For example, the grid load data and real-time environmental data of the past week can be input into the hybrid neural network. The peak and valley patterns of the load data are extracted through CNN, and the long-term change trend and dynamics of the load data are captured through LSTM. A specific charging and discharging strategy is generated. For example, if the load peak is predicted at noon tomorrow, the system recommends charging during the morning valley period. According to the generated strategy, the energy storage system is charged during the morning valley period to ensure sufficient power reserves during the peak period at noon.

[0211] Through this method, the power storage system can more accurately predict load changes, optimize charging and discharging strategies, and improve the overall stability and efficiency of the power grid.

[0212] The generated "first load regulation information" includes but is not limited to the following types of information: predicted load curve, optimal charging time period, optimal discharging time period, load peak shaving and valley filling strategy, emergency response strategy, etc. This information is obtained through deep learning and analysis of historical and real-time data, reflecting the load demand and power supply capacity of the energy storage system in different time periods.

[0213] Based on the generated "first load regulation information", a specific regulation method can be formulated. For example, when the predicted load curve shows that there will be a load peak at noon tomorrow, the system can formulate a strategy to charge during the morning valley period to ensure sufficient power reserves during the peak period. The optimal charging time period information is conducive to the energy storage device charging during the period when the electricity price is low and the load is small, thereby reducing costs; the optimal discharge time period information is conducive to the energy storage device discharging during the period when the electricity price is high and the load is large, thereby increasing profits. The load peak shaving and valley filling strategy can smooth the power load curve by regulating the charging and discharging time, thereby reducing the pressure on the power grid. The emergency response strategy provides a rapid regulation solution for sudden load changes or equipment failures to ensure stable operation of the system.

[0214] In this way, the "first load control information" generated by the hybrid neural network directly guides the charging and discharging operations of the energy storage system, ensuring that the system can operate efficiently and stably under various load and environmental conditions. This not only improves the economic benefits of the energy storage system, but also enhances the overall stability and reliability of the power grid.

[0215] Based on the same inventive concept, an embodiment of the present application also provides a load balancing control system for an energy storage device corresponding to a load balancing control method for an energy storage device. Since the principle of solving the problem by the system in the embodiment of the present application is similar to that of the load balancing control method for the above-mentioned energy storage device, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be repeated.

[0216] Reference Figure 3 FIG. 1 is a schematic diagram of a load balancing control system for an energy storage device provided in an embodiment of the present application, wherein the system comprises:

[0217] An acquisition unit 10 is used to acquire historical operation data of a target energy storage device and perform preprocessing based on a first Gaussian processing method to generate first operation data;

[0218] The processing unit 20 processes the first operation data based on a first modal decomposition method, decomposes a multi-scale time series component in the first operation data, and generates second operation data;

[0219] The decision unit 30 inputs the second operating data into the hybrid neural network for processing, and outputs the first load control information; based on the first load control information, determines the control method for the target energy storage device; wherein the hybrid neural network includes: using the first sub-network to extract the spatial features of the second operating data and sending it to the second sub-network, and the second sub-network is used for time dependency analysis.

[0220] The present application also provides an electronic device, such as Figure 4 FIG. 1 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application, including:

[0221] A processor 41 and a memory 42; the memory 42 stores machine-readable instructions executable by the processor 41, and the processor 41 is used to execute the machine-readable instructions stored in the memory 42. When the machine-readable instructions are executed by the processor 41, the processor 41 performs the following steps:

[0222] Acquire historical operation data of the target energy storage device, and perform preprocessing based on a first Gaussian processing method to generate first operation data;

[0223] Based on a first modal decomposition method, processing the first operation data, decomposing multi-scale time series components in the first operation data, and generating second operation data;

[0224] The second operating data is input into a hybrid neural network for processing, and the first load control information is output; based on the first load control information, a control method for a target energy storage device is determined; wherein the hybrid neural network includes: using a first sub-network to extract spatial features of the second operating data and sending the spatial features to a second sub-network, and the second sub-network is used for time dependency analysis.

[0225] The above-mentioned memory 42 includes internal memory 421 and external memory 422; the memory 421 here is also called internal memory, which is used to temporarily store the calculation data in the processor 41 and the data exchanged with the external memory 422 such as the hard disk. The processor 41 exchanges data with the external memory 422 through the internal memory 421.

[0226] The specific execution process of the above instructions can refer to the steps of a load balancing control method for an energy storage device described in an embodiment of the present disclosure, which will not be repeated here.

[0227] The present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a load balancing control method for an energy storage device described in the above method embodiment are executed. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0228] An embodiment of the present application also provides a computer program product, which carries a program code. The instructions included in the program code can be used to execute the steps of a load balancing control method for an energy storage device described in the above method embodiment. For details, please refer to the above method embodiment, which will not be repeated here.

[0229] The computer program product may be implemented in hardware, software or a combination thereof. In one optional embodiment, the computer program product is implemented as a computer storage medium. In another optional embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0230] It can 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.

[0231] It should be understood that determining B based on A does not mean determining B only based on A. B can also be determined based on A and / or other information.

[0232] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention 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 to be beyond the scope of the present invention.

[0233] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0234] The preferred embodiments of the present invention disclosed above are only used to help explain the present application. The preferred embodiments do not describe all the details in detail, nor do they limit the specific implementation methods of the present application. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present application, so that those skilled in the art can understand and use the present application well. The present application is limited only by the claims and their full scope and equivalents.

Claims

1. A load balancing control method for an energy storage device, characterized in that: include: Acquire historical operation data of the target energy storage device, and perform preprocessing based on a first Gaussian processing method to generate first operation data; Based on a first modal decomposition method, processing the first operation data, decomposing multi-scale time series components in the first operation data, and generating second operation data; inputting the second operation data into a hybrid neural network for processing, and outputting first load regulation information; Based on the first load control information, determining a control method for a target energy storage device; wherein the hybrid neural network comprises: using a first sub-network to extract spatial features of the second operating data and sending the spatial features to a second sub-network, wherein the second sub-network is used for time dependency analysis; The historical operation data includes: energy data, environmental data, equipment operation data and historical load data; the preprocessing based on the first Gaussian processing method to generate the first operation data includes: The functional expression of the first Gaussian processing method is: ； in, Indicates the current processing time point. represents each measurement time point in the historical operation data, It's at the time The dynamically adjusted smooth width, Based on time point The asymmetric factor of the dynamic adjustment of data characteristics in the first time period before and after, is a symbolic function, according to and The relative position returns +1 or -1; The dynamically adjusted smooth width The function expression is: ； in, It's in time The peak frequency of the data observed in the nearby second time period, is the threshold used to adjust the sensitivity of the smoothing width adjustment, is the adjustment coefficient, which is used to control the adjustment amplitude of the variance on the smoothing width. It's in time The data variance in the third time period nearby is used to indicate the fluctuation of the data; The first Gaussian processing method also includes: Based on the influence of historical data, the function expression of the first Gaussian processing method is adjusted to: ； in, is the functional expression of the first Gaussian processing method before adjustment, is a weighting factor used to adjust the impact of historical correlation, yes and historical time The data correlation between It's a time delay.

2. The load balancing control method of the energy storage device according to claim 1, characterized in that: The step of processing the first operation data based on the first modal decomposition method, decomposing the multi-scale time series components in the first operation data, and generating the second operation data comprises: Decomposing the first operating data using an empirical mode decomposition method, extracting a first intrinsic mode function of the current sequence therefrom, and subtracting the extracted intrinsic mode function from the sequence; Repeat the empirical mode decomposition for the remaining data after the first decomposition, and extract the first intrinsic mode function of the current sequence each time; In response to the remaining data being unable to further perform the empirical mode decomposition, stopping the decomposition and generating a plurality of intrinsic mode functions and a residual term; The plurality of intrinsic mode functions and a residual term are used as the second operating data.

3. The load balancing control method of the energy storage device according to claim 2, characterized in that: The first modal decomposition method further comprises: Acquiring real-time environmental data of the target energy storage device; A first target noise is added to the first operating data based on the real-time environmental data; wherein the first target noise is used to simulate the actual impact of environmental factors on the operation of the target energy storage device.

4. The load balancing control method of the energy storage device according to claim 3, characterized in that: The adding a first target noise to the first operation data based on the real-time environment data comprises: Based on the real-time environmental data, determining the intensity of the first target noise; The real-time environmental data includes humidity data and temperature data; the function expression of the intensity of the first target noise is: ； in, is the basic noise intensity, and are the adjustment coefficients for the effects of temperature and humidity on noise, and are the reference values ​​of temperature and humidity respectively, and They are real-time measurements of temperature and humidity respectively; Based on the real-time environmental data, statistical characteristics of the first target noise are determined.

5. The load balancing control method of the energy storage device according to claim 4, characterized in that: The method further includes inputting the second operation data into the hybrid neural network for processing and outputting the first load regulation information: adding a second target noise to the second operation data based on the real-time environmental data; Wherein, the second target noise includes: Based on the sliding window method, determining the change trend of the real-time environmental data; Based on the change trend of the real-time environmental data, a mean value and a variance of the second target noise are calculated.

6. A load balancing control system for an energy storage device, characterized in that: include: An acquisition unit, used for acquiring historical operation data of a target energy storage device, and performing preprocessing based on a first Gaussian processing method to generate first operation data; A processing unit, based on a first modal decomposition method, processes the first operation data, decomposes multi-scale time series components in the first operation data, and generates second operation data; A decision unit, inputting the second operation data into a hybrid neural network for processing, and outputting first load regulation information; Based on the first load control information, determining a control method for a target energy storage device; wherein the hybrid neural network comprises: using a first sub-network to extract spatial features of the second operating data and sending the spatial features to a second sub-network, wherein the second sub-network is used for time dependency analysis; The historical operation data includes: energy data, environmental data, equipment operation data and historical load data; the preprocessing based on the first Gaussian processing method to generate the first operation data includes: The functional expression of the first Gaussian processing method is: ； in, Indicates the current processing time point. represents each measurement time point in the historical operation data, It's at the time The dynamically adjusted smooth width, Based on time point The asymmetric factor of the dynamic adjustment of data characteristics in the first time period before and after, is a symbolic function, according to and The relative position returns +1 or -1; The dynamically adjusted smooth width The function expression is: ； in, It's in time The peak frequency of the data observed in the nearby second time period, is the threshold used to adjust the sensitivity of the smoothing width adjustment, is the adjustment coefficient, which is used to control the adjustment amplitude of the variance on the smoothing width. It's in time The data variance in the third time period nearby is used to indicate the fluctuation of the data; The first Gaussian processing method also includes: Based on the influence of historical data, the function expression of the first Gaussian processing method is adjusted to: ； in, is the functional expression of the first Gaussian processing method before adjustment, is a weighting factor used to adjust the impact of historical correlation, yes and historical time The data correlation between It's a time delay.

7. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and the processor is used to execute the machine-readable instructions stored in the memory. When the machine-readable instructions are executed by the processor, the processor executes the steps of a load balancing control method for an energy storage device as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program. When the computer program is executed by an electronic device, the electronic device executes the steps of a load balancing control method for an energy storage device as described in any one of claims 1 to 5.

Images