High-power battery pack charging and discharging control method and system based on artificial intelligence

Abstract

The present invention relates to the technical field of battery charge and discharge control, and discloses a high-power battery pack charge and discharge control method and system based on artificial intelligence, which is used to improve the accuracy of high-power battery pack charge and discharge control based on artificial intelligence. The method includes: performing multi-dimensional parameter mapping on real-time state estimation and future state prediction information, generating a decision space matrix, and traversing and optimizing the decision space matrix to obtain a real-time charge and discharge plan and perform charge and discharge operations, collecting battery pack operation data of a high-power battery pack, performing parallel analysis on the battery pack operation data, obtaining multi-dimensional charge and discharge index data, dynamically calibrating the multi-dimensional charge and discharge index data, obtaining real-time safety state indicators and potential risk warning signals; performing multi-objective trade-off analysis on potential risk warning signals and real-time safety state indicators, generating a target decision space, traversing and analyzing the target decision space, and obtaining a battery pack management control instruction sequence.

Description

Technical Field

[0001] The present invention relates to the technical field of battery charge and discharge control, and in particular to a high-power battery pack charge and discharge control method and system based on artificial intelligence. Background Art

[0002] As a core component of new energy technology, high-power battery packs play an increasingly important role in electric transportation, renewable energy storage, and smart grids. Traditional battery management systems mainly rely on preset charge and discharge curves and fixed safety thresholds for control. These systems usually use methods such as constant current and constant voltage or multi-stage charging, and protect the battery by monitoring parameters such as voltage, current, and temperature. With the development of technology, some advanced management systems have begun to introduce adaptive control algorithms and data-driven predictive models to improve charging and discharging efficiency and battery life. For example, some researchers have proposed charging strategies based on fuzzy logic, or used machine learning algorithms for battery state estimation.

[0003] However, existing technologies still face many challenges. First, high-power battery packs are usually composed of a large number of single cells connected in series and parallel. Their internal states are complex and changeable. Existing methods make it difficult to achieve accurate management and global optimization of each single cell. Secondly, although high-frequency data acquisition can provide rich battery status information, it also brings challenges in data processing and real-time control. Furthermore, the performance and life of the battery are affected by many factors, such as ambient temperature, charge and discharge history, aging degree, etc. Existing static models are difficult to accurately capture these dynamic changes. In addition, in terms of safety management, fixed threshold methods often find it difficult to achieve the best balance between performance and safety. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides an artificial intelligence-based high-power battery pack charge and discharge control method and system, which are used to improve the accuracy of artificial intelligence-based high-power battery pack charge and discharge control.

[0005] The present invention provides a high-power battery pack charge and discharge control method based on artificial intelligence, comprising:

[0006] S1. Perform high-frequency sampling on a high-power battery pack to obtain sampled data, and perform segmented processing on the sampled data to obtain a multi-dimensional time series data set, wherein the sampled data includes voltage, current, temperature and internal resistance;

[0007] S2. Performing dimensionality reduction and multi-scale decomposition on the multi-dimensional time series data set to obtain a low-dimensional feature vector capable of characterizing the dynamic characteristics of a high-power battery pack;

[0008] S3, performing time series analysis and data fusion on the low-dimensional feature vector to obtain real-time state estimation and future state prediction information of the high-power battery pack;

[0009] S4, performing multi-dimensional parameter mapping on the real-time state estimation and the future state prediction information to generate a decision space matrix, and traversing and optimizing the decision space matrix through a dynamic programming algorithm to obtain a real-time charging and discharging plan;

[0010] S5. Controlling the high-power battery pack to perform charging and discharging operations according to the real-time charging and discharging scheme, collecting battery pack operation data of the high-power battery pack in real time, and performing parallel analysis on the battery pack operation data to obtain multi-dimensional charging and discharging index data, and dynamically calibrating the multi-dimensional charging and discharging index data through an adaptive threshold algorithm to obtain real-time safety status indicators and potential risk warning signals;

[0011] S6. Perform a multi-objective trade-off analysis on the potential risk warning signal and the real-time safety status indicator to generate a target decision space, and perform a traversal analysis on the target decision space through a heuristic search algorithm to obtain a battery pack management control instruction sequence.

[0012] The present invention also provides a high-power battery pack charge and discharge control system based on artificial intelligence, comprising:

[0013] A sampling module, used to perform high-frequency sampling on a high-power battery pack to obtain sampling data, and to perform segmented processing on the sampling data to obtain a multi-dimensional time series data set, wherein the sampling data includes voltage, current, temperature and internal resistance;

[0014] A decomposition module, used to perform dimensionality reduction and multi-scale decomposition on the multi-dimensional time series data set to obtain a low-dimensional feature vector capable of characterizing the dynamic characteristics of the high-power battery pack;

[0015] A fusion module, used for performing time series analysis and data fusion on the low-dimensional feature vector to obtain real-time state estimation and future state prediction information of the high-power battery pack;

[0016] A mapping module, used to perform multi-dimensional parameter mapping on the real-time state estimation and the future state prediction information, generate a decision space matrix, and traverse and optimize the decision space matrix through a dynamic programming algorithm to obtain a real-time charging and discharging plan;

[0017] A calibration module, used to control the high-power battery pack to perform charging and discharging operations according to the real-time charging and discharging scheme, collect battery pack operation data of the high-power battery pack in real time, and perform parallel analysis on the battery pack operation data to obtain multi-dimensional charging and discharging index data, and dynamically calibrate the multi-dimensional charging and discharging index data through an adaptive threshold algorithm to obtain real-time safety status indicators and potential risk warning signals;

[0018] The traversal module is used to perform a multi-objective trade-off analysis on the potential risk warning signal and the real-time safety status indicator, generate a target decision space, and perform a traversal analysis on the target decision space through a heuristic search algorithm to obtain a battery pack management control instruction sequence.

[0019] In the technical solution provided by the present invention, high-frequency sampling and multi-dimensional time series data processing technology significantly improve the accuracy and real-time performance of data acquisition, can capture the transient changes of the battery pack during the charging and discharging process, and help to timely discover potential safety hazards. The use of dimensionality reduction and multi-scale decomposition technology greatly reduces the complexity of data processing and improves the calculation efficiency. By extracting low-dimensional feature vectors, not only the data scale is compressed, but also the key dynamic characteristic information of the battery pack is retained, making the subsequent state estimation and prediction more accurate and efficient. The combination of multi-dimensional parameter mapping and dynamic programming algorithm realizes the real-time optimization of the charging and discharging scheme. It is possible to quickly generate the optimal charging and discharging strategy while considering multiple objectives (such as charging efficiency, battery life and safety), thereby improving the efficiency and life of the battery pack. The introduction of the adaptive threshold algorithm greatly enhances the safety and reliability of the system. By dynamically adjusting the safety threshold, the control parameters can be adjusted in time according to the actual state of the battery pack and environmental changes, effectively preventing safety issues such as overcharging, over-discharging and temperature abnormalities. The application of multi-objective trade-off analysis and heuristic search algorithms can quickly find the optimal solution in a complex decision space and generate a sophisticated control instruction sequence. It significantly improves the utilization efficiency of the battery pack, extends its service life, and also improves energy utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 This is a flow chart of a high-power battery pack charge and discharge control method based on artificial intelligence in an embodiment of the present invention.

[0022] Figure 2Schematic diagram of an artificial intelligence-based high-power battery pack charging and discharging control system in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0025] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0026] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , Figure 1 is a flow chart of a high-power battery pack charge and discharge control method based on artificial intelligence according to an embodiment of the present invention. Figure 1 As shown, the following steps are included:

[0027] S1. Perform high-frequency sampling on a high-power battery pack to obtain sampling data, and perform segmented processing on the sampling data to obtain a multi-dimensional time series data set, wherein the sampling data includes voltage, current, temperature and internal resistance;

[0028] S2. Perform dimensionality reduction and multi-scale decomposition on the multi-dimensional time series data set to obtain a low-dimensional feature vector that can characterize the dynamic characteristics of the high-power battery pack;

[0029] S3, performing time series analysis and data fusion on the low-dimensional feature vector to obtain real-time state estimation and future state prediction information of the high-power battery pack;

[0030] S4, perform multi-dimensional parameter mapping on the real-time state estimation and future state prediction information, generate a decision space matrix, and traverse and optimize the decision space matrix through a dynamic programming algorithm to obtain a real-time charging and discharging plan;

[0031] S5. Control the high-power battery pack to perform charging and discharging operations according to the real-time charging and discharging plan, collect the battery pack operation data of the high-power battery pack in real time, and perform parallel analysis on the battery pack operation data to obtain multi-dimensional charging and discharging index data. Dynamically calibrate the multi-dimensional charging and discharging index data through an adaptive threshold algorithm to obtain real-time safety status indicators and potential risk warning signals;

[0032] S6. Perform a multi-objective trade-off analysis on potential risk warning signals and real-time safety status indicators to generate a target decision space. Perform a traversal analysis on the target decision space through a heuristic search algorithm to obtain a battery pack management control instruction sequence.

[0033] It should be noted that the voltage, current, temperature and internal resistance data of the battery pack are obtained through high-frequency sampling. The sampling frequency is usually set between 1kHz and 10kHz to capture transient changes. The collected raw data is filtered and segmented to form a multi-dimensional time series data set. Subsequently, these data are reduced in dimension and multi-scale decomposed, and low-dimensional feature vectors that can characterize the dynamic characteristics of the battery pack are extracted using techniques such as principal component analysis (PCA) and wavelet transform. Time series analysis and data fusion are performed on the low-dimensional feature vectors, and deep learning models such as long short-term memory networks (LSTM) are used in combination with Kalman filters to obtain real-time state estimation and future state prediction information of the battery pack. This information includes key indicators such as state of charge (SOC), state of health (SOH) and remaining service life (RUL).

[0034] Based on the state estimation and prediction information, multi-dimensional parameter mapping is performed to generate a decision space matrix. The matrix is ​​traversed and optimized by the dynamic programming algorithm to obtain a real-time charging and discharging plan. This plan takes into account multiple goals such as battery safety, charging and discharging efficiency, and life extension. While executing the charging and discharging plan, the battery pack operation data is collected in real time and analyzed in parallel. The adaptive threshold algorithm is used to dynamically calibrate the multi-dimensional charging and discharging index data to obtain real-time safety status indicators and potential risk warning signals. This process can detect abnormal conditions in a timely manner, such as overcharging, over-discharging, or abnormal temperature. Finally, a multi-objective trade-off analysis is performed on the potential risk warning signals and real-time safety status indicators to generate a target decision space. A heuristic search algorithm, such as an improved ant colony algorithm, is used to traverse and analyze the target decision space, and finally a battery pack management control instruction sequence is obtained. These instruction sequences include specific operation instructions such as charging and discharging power adjustment, temperature control, and balancing management.

[0035] Take a lithium-ion battery pack for a large electric vehicle as an example. The battery pack consists of 500 cells connected in series, with a rated voltage of 800V and a capacity of 200kWh. During a fast charging process, the sampling frequency is set to 5kHz, generating 20,000 data points per second. Through PCA dimensionality reduction and wavelet transform, the original 500-dimensional data is compressed into a 20-dimensional feature vector. Based on these feature vectors, the LSTM model predicts the SOC change trend and temperature distribution in the next 2 hours. Based on these prediction results, the dynamic programming algorithm generates a three-stage charging plan: fast charging at 150kW for the first 30 minutes, reducing to 100kW for the middle 60 minutes, and slow charging at 50kW for the last 30 minutes. During the charging process, the adaptive threshold algorithm dynamically adjusts the temperature warning threshold from 55℃ to 52℃ because the temperature rise rate of some battery cells is detected to be abnormal. Finally, the heuristic search algorithm generates a detailed control instruction sequence based on these safety factors, including power adjustment instructions every 5 minutes and balancing management instructions every 2 minutes. This intelligent control method not only achieves fast charging, but also effectively protects the battery.

[0036] By executing the above steps, high-frequency sampling and multi-dimensional time series data processing technology significantly improve the accuracy and real-time performance of data acquisition, can capture the transient changes of the battery pack during the charging and discharging process, and help to timely discover potential safety hazards. The use of dimensionality reduction and multi-scale decomposition technology greatly reduces the complexity of data processing and improves computational efficiency. By extracting low-dimensional feature vectors, not only the data size is compressed, but also the key dynamic characteristics information of the battery pack is retained, making subsequent state estimation and prediction more accurate and efficient. The combination of multi-dimensional parameter mapping and dynamic programming algorithm realizes real-time optimization of charging and discharging schemes. It can quickly generate the optimal charging and discharging strategy while considering multiple objectives (such as charging efficiency, battery life and safety), thereby improving the efficiency and life of the battery pack. The introduction of the adaptive threshold algorithm greatly enhances the safety and reliability of the system. By dynamically adjusting the safety threshold, the control parameters can be adjusted in time according to the actual state of the battery pack and environmental changes, effectively preventing safety issues such as overcharging, over-discharging and temperature abnormalities. The application of multi-objective trade-off analysis and heuristic search algorithms can quickly find the optimal solution in a complex decision space and generate a sophisticated control instruction sequence. It significantly improves the utilization efficiency of the battery pack, extends its service life, and also improves energy utilization efficiency.

[0037] In a specific embodiment, the process of executing step S1 may specifically include the following steps:

[0038] (1) Perform high-frequency sampling on the voltage, current, temperature and internal resistance of the high-power battery pack to obtain sampled data, and perform noise removal on the sampled data to obtain filtered sampled data;

[0039] (2) Dividing the filtered sampling data into time windows to obtain multiple time window data segments, and extracting features from the multiple time window data segments to obtain preliminary feature data;

[0040] (3) Normalizing the preliminary feature data to obtain standardized feature data, and performing correlation analysis on the standardized feature data to obtain a feature correlation matrix;

[0041] (4) Perform feature selection on the standardized feature data according to the feature correlation matrix to obtain a candidate feature set, and perform time series alignment on the candidate feature set to obtain aligned feature data;

[0042] (5) Interpolate the aligned feature data to obtain equally spaced sampling data, and reconstruct the equally spaced sampling data to obtain reconstructed time series data;

[0043] (6) Perform data enhancement processing on the reconstructed time series data to obtain a multidimensional time series data set.

[0044] Specifically, the voltage, current, temperature and internal resistance of the high-power battery pack are sampled at high frequency. The sampling frequency is usually set between 1kHz and 10kHz to capture the transient changes of the battery pack during the charging and discharging process. The raw data obtained by sampling often contains various noises, such as electromagnetic interference, sensor errors, etc. In order to improve the data quality, digital filtering technology is used to remove noise from the sampled data. Common filtering methods include median filtering, wavelet denoising, etc. The filtered data is smoother and can more accurately reflect the actual state of the battery pack.

[0045] The filtered sampled data is divided into time windows. The size of the time window depends on the characteristics of the battery pack and the application scenario, usually between 1 second and 60 seconds. The purpose of window division is to divide the continuous time series data into multiple relatively independent data segments to facilitate subsequent feature extraction and analysis. Feature extraction is performed on each time window data segment, and the extracted features include statistical features (such as mean, variance, peak, etc.) and frequency domain features (such as Fourier transform coefficients). These features can describe the working state of the battery pack in each time window from different angles. In order to eliminate the dimensional differences between different features, the preliminary feature data is normalized. Commonly used normalization methods include minimum-maximum normalization and Z-score normalization. The normalized feature data are mapped to the same scale range, which is conducive to subsequent correlation analysis and feature selection. Correlation analysis is performed on the standardized feature data, and the correlation coefficients between the features are calculated to form a feature correlation matrix. Correlation analysis helps to identify redundant features and key features, and provides a basis for feature selection.

[0046] Based on the feature correlation matrix, feature selection algorithms (such as principal component analysis, recursive feature elimination, etc.) are used to screen the standardized feature data to obtain a candidate feature set. The purpose of feature selection is to reduce the data dimension, remove redundant and irrelevant features, and improve the training efficiency and generalization ability of subsequent models. Since different features may come from different sensors or sampling time points, it is necessary to perform time series alignment on the candidate feature set to ensure the consistency of all features in the time dimension. The aligned feature data is interpolated to obtain equally spaced sampling data. Interpolation methods include linear interpolation, spline interpolation, etc. Selecting a suitable interpolation method can generate uniformly distributed time series data while maintaining the original characteristics of the data. Equally spaced sampling data facilitates time series analysis and prediction. Subsequently, the equally spaced sampling data is reconstructed in time series, such as using the embedding dimension and time delay parameter to convert the one-dimensional time series into a multidimensional phase space to capture the dynamic characteristics of the system.

[0047] Finally, the reconstructed time series data is enhanced to obtain a multidimensional time series data set. Data enhancement techniques include adding Gaussian noise, time window sliding, random scaling, etc., with the aim of expanding the data set and improving the robustness and generalization ability of the model.

[0048] Take a high-power lithium-ion battery pack as an example. The battery pack consists of 200 single cells connected in series, with a rated voltage of 740V and a capacity of 300kWh. During the charging process, the sampling frequency is set to 5kHz, and the voltage, current, temperature and internal resistance are sampled synchronously. The original sampled data contains 50Hz grid interference and random noise, which are effectively removed by combining the band-stop filter and wavelet denoising. Subsequently, the filtered data is divided into 30-second time windows, and 20 statistical features and 10 frequency domain features are extracted in each window. All features are mapped to the [0, 1] interval through minimum-maximum normalization, and the Pearson correlation coefficient between features is calculated to form a 30×30 correlation matrix. Based on the correlation analysis results, the one with the lower correlation among the feature pairs with an absolute value of the correlation coefficient greater than 0.8 is selected, and finally 15 candidate features are obtained. The 15 features are time-series aligned and linearly interpolated to generate equally spaced data with a sampling interval of 1 second. The one-dimensional time series is reconstructed into a 5-dimensional phase space through the time delay embedding method. Finally, the original data was expanded 3 times by adding 5% Gaussian noise and ±10% random scaling to form a multidimensional time series dataset with 45 features, which not only retains the key information of the battery pack charging process but also enhances the diversity of the data.

[0049] In a specific embodiment, the process of executing step S2 may specifically include the following steps:

[0050] (1) Perform principal component analysis on the multidimensional time series data set to obtain the principal component feature matrix, and calculate the contribution rate of the principal component feature matrix to obtain the feature contribution rate sequence;

[0051] (2) According to the feature contribution rate sequence, the principal component feature matrix is ​​reduced in dimension to obtain preliminary reduced-dimensional data, and the preliminary reduced-dimensional data is subjected to wavelet transform to obtain multi-scale decomposition coefficients;

[0052] (3) Performing energy distribution analysis on the multi-scale decomposition coefficients to obtain energy distribution characteristics, and reconstructing the multi-scale decomposition coefficients according to the energy distribution characteristics to obtain reconstructed feature data;

[0053] (4) Perform nonlinear mapping on the reconstructed feature data to obtain a low-dimensional feature vector.

[0054] Specifically, the multidimensional time series data set is represented as a matrix , where rows represent samples and columns represent features. Calculate The covariance matrix of :

[0055]

[0056] in is the number of samples, for The mean vector of . Then solve the covariance matrix The eigenvalues ​​and eigenvectors of the principal component are obtained P . Eigenvalue and the corresponding eigenvector Satisfies the equation:

[0057]

[0058] Then, the contribution rate of the principal component feature matrix is ​​calculated to obtain the feature contribution rate sequence. The contribution rate of the principal components The calculation is as follows:

[0059]

[0060] in is the total number of features. The contribution rate reflects the degree to which each principal component explains the variance of the original data. According to the feature contribution rate sequence, select the top principal components whose cumulative contribution rate reaches the preset threshold (such as 95%). k principal components, and perform dimensionality reduction processing on the principal component feature matrix to obtain preliminary dimensionality reduction data

[0061] :

[0062] in For the front k The eigenvector matrix corresponding to the principal components.

[0063] Perform wavelet transform on the preliminary dimension reduction data to obtain multi-scale decomposition coefficients. Wavelet transform decomposes the signal into approximate coefficients and detail coefficients of different scales. Use discrete wavelet transform (DWT) to decompose each dimension-reduced feature:

[0064]

[0065] in For the dimensionality reduction features, for The approximation coefficient of level, , ,..., are the detail coefficients at each level. Perform energy distribution analysis on the multi-scale decomposition coefficients to obtain energy distribution characteristics. Calculate the energy of each scale coefficient:

[0066]

[0067] According to the energy distribution characteristics, the scale coefficient with a larger energy proportion is selected for reconstruction to obtain the reconstructed feature data. The reconstruction process uses the inverse discrete wavelet transform (IDWT):

[0068]

[0069] in, is a coefficient selected or adjusted according to the energy distribution characteristics.

[0070] Finally, nonlinear mapping is performed on the reconstructed feature data to obtain a low-dimensional feature vector. Nonlinear mapping can be achieved using methods such as kernel principal component analysis (KPCA) or autoencoders. Taking KPCA as an example, first select the kernel function , such as the Gaussian kernel:

[0071]

[0072] Then construct the kernel matrix , and solve the eigenvalue problem:

[0073]

[0074] in is the feature vector, is the number of samples, is the eigenvalue. Finally, the reconstructed feature data is projected into the feature space to obtain a low-dimensional feature vector :

[0075]

[0076] Take a high-power lithium battery pack as an example. The battery pack consists of 500 single cells connected in series and parallel, with a rated voltage of 1000V and a capacity of 500kWh. During a charging process, a multidimensional time series data set containing 50 features was collected, with a total of 10,000 time points. First, principal component analysis was performed to calculate 50 principal components and their corresponding contribution rates. The first 15 principal components with a cumulative contribution rate of 98% were selected to reduce the original 50-dimensional data to 15 dimensions. The 15-dimensional data was subjected to a 5-layer wavelet decomposition using the db4 wavelet. Through energy analysis, it was found that about 85% of the energy was concentrated in the first 3 layers, so only the coefficients of these 3 layers were retained during reconstruction. Finally, the KPCA method with a Gaussian kernel of 0.1 was used to further map the reconstructed 15-dimensional data to an 8-dimensional feature space. This series of dimensionality reduction and feature extraction processes not only greatly reduced the data dimension, but also retained the key dynamic characteristics of the battery charging process.

[0077] In a specific embodiment, the process of executing step S3 may specifically include the following steps:

[0078] (1) Decompose the low-dimensional feature vector into time series to obtain trend component, periodic component and random component, and perform trend extrapolation on the trend component to obtain trend prediction results;

[0079] (2) Performing periodic analysis on the periodic component to obtain periodic characteristic parameters, and performing periodic prediction on the periodic component based on the periodic characteristic parameters to obtain periodic prediction results;

[0080] (3) Performing autoregressive analysis on the random component to obtain autoregressive model parameters, and performing random prediction on the random component based on the autoregressive model parameters to obtain random prediction results;

[0081] (4) The trend prediction results, period prediction results, and random prediction results are weightedly fused to obtain the real-time state estimation and future state prediction information of the high-power battery pack.

[0082] Specifically, the seasonal and trend decomposition (STL) method is used to decompose the low-dimensional feature vector to obtain the trend component, the periodic component and the random component. The STL method splits the time series data into these three components through iterative weighted least squares, where the trend component reflects the long-term trend of the data, the periodic component captures the periodic change pattern of the data, and the random component contains irregular fluctuations and noise. When the trend component is extrapolated, the polynomial regression method is used. First, determine the appropriate polynomial order, and usually select the best order through cross-validation. Then use the least squares method to fit the polynomial function to obtain the mathematical expression of the trend component. Using this expression, the trend in the future can be extrapolated to obtain the trend prediction result. This method can effectively capture the long-term change trend of parameters such as battery pack capacity and internal resistance.

[0083] When performing periodic analysis on the periodic component, first use the fast Fourier transform (FFT) to convert the time domain signal into a frequency domain signal. By analyzing the spectrum, the main periodic components are identified, and their periodic characteristic parameters such as frequency, amplitude and phase are extracted. These parameters reflect the periodic changes of the battery pack on different time scales, such as daily charge and discharge cycles, temperature changes, etc. According to the extracted periodic characteristic parameters, the periodic signal is reconstructed using the inverse Fourier transform, and it is extrapolated to the future time point to obtain the periodic prediction result. When performing autoregressive analysis on the random component, the autoregressive integrated moving average (ARIMA) model is used. First, the stationarity of the random component is tested. If it is not stationary, it is stabilized by differential processing. Then, the order of the ARIMA model is determined by analyzing the autocorrelation function (ACF) and partial autocorrelation function (PACF). The model parameters are estimated using the maximum likelihood estimation method to obtain the autoregressive model parameters. Based on the estimated model parameters, the Monte Carlo simulation method is used to generate multiple possible random component prediction paths, and the average value is taken as the random prediction result.

[0084] Finally, the trend prediction results, periodic prediction results, and random prediction results are weighted and fused to obtain the real-time state estimation and future state prediction information of the high-power battery pack. The weight coefficient is determined by the adaptive weight method, and the weight is dynamically adjusted according to the historical accuracy of each component prediction. This fusion method can comprehensively consider the long-term trend, periodic changes, and random fluctuations of the battery pack state, and improve the accuracy and robustness of the prediction.

[0085] Take a lithium-ion battery pack for a large electric vehicle as an example. The rated capacity of the battery pack is 100kWh, and it consists of 100 modules connected in series, each of which contains 10 battery cells connected in parallel. During the one-month usage cycle, data is collected at 48 time points every day to form a low-dimensional feature vector time series containing 8 features. After STL decomposition of these data, it is found that the trend component of the capacity feature shows a slow downward trend, reflecting the aging process of the battery; the periodic component shows an obvious 24-hour cycle, corresponding to the daily charging and discharging mode; the random component reflects the fluctuations caused by the usage conditions and ambient temperature changes.

[0086] A polynomial regression was performed on the trend component, and the third-order polynomial was selected as the best fitting model. The cycle analysis found two main cycles of 24 hours and 7 days through FFT, corresponding to daily use and weekend long-distance driving modes, respectively. The random component was fitted using the ARIMA (2,1,1) model, with a difference order of 1, an autoregressive order of 2, and a sliding average order of 1. The prediction results of these three components were weighted fused, with the initial weights of the trend, cycle, and random components being 0.5, 0.3, and 0.2, respectively, and then dynamically adjusted according to the prediction error. Through this decomposition-prediction-fusion method, not only the current state of the battery pack is accurately estimated, but also the capacity change trend in the next week is successfully predicted. The prediction results show that under normal use, the battery pack capacity will drop by about 0.2% within a week, but there may be a short-term sharp drop of about 1.5% on weekends.

[0087] In a specific embodiment, the process of executing step S4 may specifically include the following steps:

[0088] (1) Extract parameters of real-time state estimation and future state prediction information to obtain a key state parameter set, and quantize the key state parameter set to obtain a quantized parameter vector;

[0089] (2) Perform multi-dimensional mapping on the quantization parameter vector to obtain the initial decision space, and then grid the initial decision space to obtain the decision space matrix;

[0090] (3) The state transfer calculation of the decision space matrix is ​​performed through the dynamic programming algorithm to obtain the optimal state transfer path, and a real-time charging and discharging plan is generated based on the optimal state transfer path.

[0091] Specifically, in the AI-based high-power battery pack charge and discharge control method, parameter extraction of real-time state estimation and future state prediction information is a key step in generating an optimized charge and discharge plan. First, key state parameters are extracted from state estimation and prediction information, including remaining capacity (SOC), state of health (SOH), temperature distribution, and internal resistance change.

[0092] These parameters are determined by combining feature importance analysis and expert knowledge to form a set of key state parameters. ,in is the number of parameters. Then, the key state parameter set is quantized to unify the parameters of different dimensions and scales into the same numerical range to obtain the quantized parameter vector The quantization process uses the normalization method:

[0093]

[0094] in and Respectively The minimum and maximum values ​​of the parameters.

[0095] Perform multi-dimensional mapping on the quantization parameter vector to construct the initial decision space. The mapping process uses a nonlinear projection method, such as the t-SNE (t-distributed stochastic neighbor embedding) algorithm, to map the high-dimensional parameter space to the low-dimensional decision space. Suppose the mapping function is , then the initial decision space It can be expressed as:

[0096]

[0097] in Usually it is a two-dimensional or three-dimensional space, which is convenient for subsequent meshing and visualization. The initial decision space is meshed to obtain the decision space matrix The grid division adopts the adaptive grid method, and a finer grid is used in the area where the parameters change drastically to improve the decision accuracy. Suppose the grid division function is ,but:

[0098]

[0099] Decision Space Matrix Each element of represents a possible charge and discharge decision state, including charge and discharge power, duration and other information. Finally, the decision space matrix is ​​calculated through the dynamic programming algorithm to obtain the optimal state transition path. The dynamic programming process is defined as follows:

[0100] Assume the state transfer function is , indicating that at time state Take action The next state after . Define the value function Indicates from the state The maximum cumulative reward from the start to the end state.

[0101] Recursion equation:

[0102]

[0103] in is the immediate reward function, is the discount factor. Through reverse recursion, the optimal state transfer path is obtained, and a real-time charging and discharging plan is generated based on this path.

[0104] Take a large lithium battery pack for electric buses as an example. The rated capacity of the battery pack is 350kWh and it consists of 1,400 battery cells connected in series. During a charging process, six key state parameters are extracted: average SOC, SOC consistency, average temperature, temperature uniformity, average internal resistance and internal resistance growth rate. After quantization, these parameters form a 6-dimensional quantized parameter vector. The t-SNE algorithm is used to map the 6-dimensional vector to a 2-dimensional plane to obtain the initial decision space. The 2-dimensional plane is divided into a 100x100 grid to form a decision space matrix. In the dynamic programming process, the state is defined as the current SOC and temperature, and the action is the charging power (range 0-250kW, step size 10kW). The immediate reward function considers three factors: charging speed, battery life and charging efficiency:

[0105]

[0106] in , , is the weight coefficient, is the current temperature, The optimal operating temperature is obtained by reverse recursion, and the optimal charging strategy from 20% of the current SOC to 90% of the target SOC is obtained: use 200kW fast charging when the SOC is lower than 60%, gradually reduce to 150kW when the SOC is 60%-80%, and finally reduce to 100kW slow charging in the final stage. This charging scheme not only ensures charging efficiency, but also avoids battery overheating and excessive aging.

[0107] In a specific embodiment, the process of executing step S5 may specifically include the following steps:

[0108] (1) Control the charging and discharging of the high-power battery pack according to the real-time charging and discharging plan to obtain real-time charging and discharging data, and collect and pre-process the real-time charging and discharging data to obtain the battery pack operation data;

[0109] (2) Extract multi-dimensional features from the battery pack operation data to obtain multi-dimensional charge and discharge index data, and perform statistical analysis on the multi-dimensional charge and discharge index data to obtain index distribution characteristics;

[0110] (3) Calculate the dynamic safety threshold through the adaptive threshold algorithm according to the indicator distribution characteristics and obtain the threshold adjustment parameter;

[0111] (4) The multi-dimensional charging and discharging indicator data are calibrated according to the threshold adjustment parameters to obtain real-time safety status indicators and potential risk warning signals.

[0112] Specifically, the battery pack is monitored in real time through high-precision current sensors and voltage acquisition modules, and parameters such as total voltage, single cell voltage, charge and discharge current, and temperature are collected to form real-time charge and discharge data. These raw data usually contain noise and outliers and need to be preprocessed. The preprocessing process includes data filtering, outlier detection, and compensation. The Kalman filter algorithm is used for filtering, which can effectively remove high-frequency noise; the local outlier factor (LOF) algorithm is used for outlier detection to identify and remove data points that are obviously deviated from the normal range; and the data compensation uses the interpolation algorithm to fill in missing values ​​to ensure the continuity and integrity of the data. The preprocessed data is the battery pack operation data, which provides a reliable basis for subsequent analysis. Multi-dimensional feature extraction of battery pack operation data is an important means to deeply understand the battery status. The feature extraction process involves multiple angles such as time domain, frequency domain, and statistics. In the time domain, statistical features such as mean, standard deviation, and peak are extracted; in the frequency domain, the spectrum characteristics of voltage and current fluctuations are analyzed by fast Fourier transform (FFT); at the statistical level, high-order statistics such as skewness and kurtosis are calculated to capture the subtle characteristics of data distribution. In addition, complexity indicators such as sample entropy are introduced to evaluate the complexity of the battery's working state. These multi-dimensional features together constitute the charge and discharge index data, which fully reflects the working state of the battery pack.

[0113] Statistical analysis of multi-dimensional charging and discharging indicator data is a key step in obtaining the distribution characteristics of indicators. Statistical analysis includes two aspects: descriptive statistics and inferential statistics. Descriptive statistics calculates the mean, median, quartile, etc. of each indicator, draws box plots and histograms, and intuitively displays the data distribution; inferential statistics evaluates the statistical significance and change trend of indicators through hypothesis testing and confidence interval estimation. In particular, the kernel density estimation method is used to obtain the probability density function of each indicator. These functions together constitute the indicator distribution characteristics and provide a basis for subsequent safety threshold calculation. The adaptive threshold algorithm is based on machine learning methods and can dynamically adjust the safety threshold according to historical data and current working conditions. The algorithm first establishes a mapping relationship between indicators and safety states, and uses classifiers such as support vector machines (SVM) or random forests to train a model that can distinguish between safe and dangerous states. Then, through the sliding time window method, the model parameters are continuously updated to enable it to adapt to changes in battery performance over time. The calculation of the dynamic safety threshold takes into account multiple factors, including the historical distribution of indicators, current working status, ambient temperature, etc., and finally obtains a set of threshold adjustment parameters for real-time adjustment of the safety thresholds of various indicators.

[0114] Finally, the multi-dimensional charge and discharge indicator data is calibrated according to the threshold adjustment parameters to obtain real-time safety status indicators and potential risk warning signals. The calibration process uses fuzzy logic control method to map the original indicator value to the [0,1] interval to indicate the degree of safety. At the same time, multi-level warning thresholds are set. When the indicator approaches or exceeds a certain level threshold, the corresponding warning signal is triggered. The warning signals are divided into three levels: mild, moderate and severe. Each level corresponds to a different processing strategy, such as reducing the charge and discharge power, starting forced heat dissipation or emergency disconnection.

[0115] Take the lithium-ion battery pack of a large energy storage power station as an example. The battery pack consists of 5,000 single cells connected in series and parallel, with a total capacity of 10MWh. During a charging process, data is collected once per second, including 5,000 single cell voltages, 100 temperature points, total current and total voltage. After preprocessing, 20 time domain features, 10 frequency domain features and 5 statistical features are extracted to form 35-dimensional charging and discharging index data. Statistical analysis of these indicators shows that the single cell voltage balance shows a skewed distribution with a kurtosis of 3.5, indicating a slight imbalance; the temperature distribution is approximately normal, with a mean of 35°C and a standard deviation of 2.5°C.

[0116] Through the adaptive threshold algorithm, considering that the battery has been used for two years and the capacity has decayed to 92% of the initial value, the upper limit of the single cell voltage is dynamically adjusted from 4.2V to 4.15V, and the upper limit of the temperature is adjusted from 55°C to 50°C. When charging to 80% SOC, it was detected that the voltage of 3 cells exceeded 4.13V, triggering a mild warning signal, and the system automatically reduced the charging power from 2MW to 1.5MW. At the same time, it was observed that the temperature point No. 2 rose abnormally, reaching 48°C, close to the adjusted upper temperature limit, and the system immediately initiated forced heat dissipation measures in this area. These real-time adjustments ensure that the battery pack always remains within a safe operating range while charging efficiently.

[0117] In a specific embodiment, the process of executing step S6 may specifically include the following steps:

[0118] (1) Weights are assigned to potential risk warning signals and real-time safety status indicators to obtain a weighted indicator set, and multi-objective optimization calculations are performed on the weighted indicator set to obtain alternative decision-making plans;

[0119] (2) Conduct feasibility assessment on alternative decision plans to obtain a feasible decision set, and construct a decision space for the feasible decision set to obtain the target decision space;

[0120] (3) The target decision space is traversed and searched through a heuristic search algorithm to obtain the optimal decision path, and a battery pack management control instruction sequence is generated based on the optimal decision path.

[0121] It should be noted that the analytic hierarchy process (AHP) is used to evaluate the importance of each indicator. Factors such as the urgency, impact range and duration of the indicator are taken into account during the evaluation process. For example, battery temperature overlimit is usually given a higher weight because excessive temperature will directly threaten battery safety; while slight voltage imbalance is given a relatively low weight. After the weight allocation is completed, the original value of each indicator is multiplied by its corresponding weight to obtain a weighted indicator set. Multi-objective optimization calculation of the weighted indicator set is the core link in generating alternative decision plans. Multi-objective optimization considers multiple objectives such as battery safety, charging and discharging efficiency, and life extension. The improved non-dominated sorting genetic algorithm (NSGA-III) is used for optimization calculation. The algorithm simulates the biological evolution process, continuously iterates to generate new solution sets, and maintains the diversity of solutions through non-dominated sorting and congestion calculation. In each generation of iteration, the algorithm evaluates the fitness of each solution in the current solution set. The fitness function comprehensively considers factors such as safety risk, energy efficiency and cycle life. After multiple iterations, a set of Pareto optimal solutions, i.e., the set of alternative decision plans, is finally obtained.

[0122] Feasibility evaluation of alternative decision plans is an important step to ensure that the plans are actually executable. The evaluation process uses the Monte Carlo simulation method to simulate the performance of each alternative plan under different working conditions. During the simulation process, various possible working states and external conditions, such as load changes, ambient temperature fluctuations, etc., are randomly generated, and then the performance indicators of the plan under these conditions are calculated. By statistically analyzing the simulation results, the plans that can maintain stable performance in most cases are screened out to form a feasible decision set. Constructing the decision space for the feasible decision set provides the basis for subsequent optimization search. The decision space construction uses the principal component analysis (PCA) method to compress high-dimensional decision parameters into a low-dimensional space. PCA calculates the eigenvalues ​​and eigenvectors of the covariance matrix to find the main direction of change in the data, thereby achieving dimensionality reduction. The decision space after dimensionality reduction is usually two-dimensional or three-dimensional, which is convenient for visualization and subsequent search. In this low-dimensional space, each point represents a feasible decision plan, and the distance between points reflects the similarity between plans.

[0123] The improved ant colony algorithm (ACO) is used for search. The ant colony algorithm simulates the behavior of ants in searching for food and guides the search direction through the accumulation and volatilization of pheromones. During the search process, each "ant" represents a search agent, moving in the decision space and evaluating the path it passes through. The evaluation criteria include the safety, efficiency and reliability of the path. As the search proceeds, the pheromone concentration on the high-quality path gradually increases, attracting more search agents and finally converging to the optimal decision path. Generating the battery pack management control instruction sequence according to the optimal decision path is the last step to achieve intelligent control. The control instruction sequence includes specific operation instructions such as charge and discharge power regulation, temperature control, and balancing management. These instructions are converted from each node of the decision path through a pre-defined mapping relationship. The generation of the instruction sequence takes into account the execution order and time interval to ensure that each operation can transition smoothly and avoid sudden changes that cause impact on the battery.

[0124] Take a charging station for a large fleet of electric buses as an example. The charging station is equipped with 100 charging piles, each with a rated power of 120kW. On a busy weekday evening, 50 electric buses arrive at the charging station at the same time for charging. The system first evaluates the battery status of each vehicle, including the remaining power, health status, and temperature distribution. Through the AHP method, weights are assigned to each indicator. For example, battery temperature, remaining power, and charging urgency are assigned weights of 0.4, 0.3, and 0.3, respectively. Subsequently, the system uses the NSGA-III algorithm to generate 100 alternative charging schemes, each of which contains the charging power and charging sequence of 50 vehicles. Through Monte Carlo simulation, 1,000 different load conditions and ambient temperature changes are simulated, and finally 20 stable and feasible charging schemes are selected. These 20 schemes are reduced in dimension by PCA to form a three-dimensional decision space.

[0125] The improved ant colony algorithm is used to search in this three-dimensional space, with an initial setting of 50 "ants" and 100 iterations. In each iteration, the total charging time of the path, the balance of the grid load, and the impact on battery life are evaluated. Finally, the algorithm converges to an optimal charging strategy: the 50 vehicles are divided into three batches, the first batch of 20 vehicles with the lowest power are charged at 100kW, the second batch of 15 vehicles are charged at 80kW, and the last 15 vehicles are charged at 60kW. The entire charging process is expected to last 3 hours, during which the charging power is dynamically adjusted to respond to changes in the grid load. Based on this optimal strategy, the system generates a detailed sequence of control instructions, including charging power adjustment instructions every 15 minutes, battery balancing inspection instructions every 10 minutes, and cooling system control instructions adjusted in real time according to battery temperature changes. This intelligent charging solution not only meets the charging needs of all buses, but also effectively balances the grid load and maximizes battery life.

[0126] The embodiment of the present invention also provides a high-power battery pack charge and discharge control system based on artificial intelligence, such as Figure 2 As shown, the artificial intelligence-based high-power battery pack charge and discharge control system specifically includes:

[0127] The sampling module 201 is used to perform high-frequency sampling on the high-power battery pack to obtain sampling data, and to perform segmented processing on the sampling data to obtain a multi-dimensional time series data set, wherein the sampling data includes voltage, current, temperature and internal resistance;

[0128] A decomposition module 202, configured to perform dimensionality reduction and multi-scale decomposition on the multi-dimensional time series data set to obtain a low-dimensional feature vector capable of characterizing the dynamic characteristics of the high-power battery pack;

[0129] A fusion module 203 is used to perform time series analysis and data fusion on the low-dimensional feature vector to obtain real-time state estimation and future state prediction information of the high-power battery pack;

[0130] A mapping module 204 is used to perform multi-dimensional parameter mapping on the real-time state estimation and the future state prediction information to generate a decision space matrix, and traverse and optimize the decision space matrix through a dynamic programming algorithm to obtain a real-time charging and discharging plan;

[0131] The calibration module 205 is used to control the high-power battery pack to perform charging and discharging operations according to the real-time charging and discharging scheme, collect the battery pack operation data of the high-power battery pack in real time, and perform parallel analysis on the battery pack operation data to obtain multi-dimensional charging and discharging index data, and dynamically calibrate the multi-dimensional charging and discharging index data through an adaptive threshold algorithm to obtain real-time safety status indicators and potential risk warning signals;

[0132] The traversal module 206 is used to perform a multi-objective trade-off analysis on the potential risk warning signal and the real-time safety status indicator, generate a target decision space, and perform a traversal analysis on the target decision space through a heuristic search algorithm to obtain a battery pack management control instruction sequence.

[0133] Through the collaborative work of the above modules, high-frequency sampling and multi-dimensional time series data processing technology significantly improve the accuracy and real-time performance of data acquisition, can capture the transient changes of the battery pack during the charging and discharging process, and help to timely discover potential safety hazards. The use of dimensionality reduction and multi-scale decomposition technology greatly reduces the complexity of data processing and improves computational efficiency. By extracting low-dimensional feature vectors, not only the data scale is compressed, but also the key dynamic characteristics information of the battery pack is retained, making the subsequent state estimation and prediction more accurate and efficient. The combination of multi-dimensional parameter mapping and dynamic programming algorithm realizes the real-time optimization of the charging and discharging scheme. It can quickly generate the optimal charging and discharging strategy while considering multiple objectives (such as charging efficiency, battery life and safety), thereby improving the efficiency and life of the battery pack. The introduction of the adaptive threshold algorithm greatly enhances the safety and reliability of the system. By dynamically adjusting the safety threshold, the control parameters can be adjusted in time according to the actual state of the battery pack and environmental changes, effectively preventing safety issues such as overcharging, over-discharging and temperature abnormalities. The application of multi-objective trade-off analysis and heuristic search algorithms can quickly find the optimal solution in a complex decision space and generate a sophisticated control instruction sequence. It significantly improves the utilization efficiency of the battery pack, extends its service life, and also improves energy utilization efficiency.

[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0135] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0136] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-power battery pack charge and discharge control method based on artificial intelligence, characterized in that: include: S1. Perform high-frequency sampling on a high-power battery pack to obtain sampled data, and perform segmented processing on the sampled data to obtain a multi-dimensional time series data set, wherein the sampled data includes voltage, current, temperature and internal resistance; S2, perform dimensionality reduction and multi-scale decomposition on the multi-dimensional time series data set to obtain a low-dimensional feature vector that can characterize the dynamic characteristics of the high-power battery pack; Step S2 includes: performing principal component analysis on the multi-dimensional time series data set to obtain a principal component feature matrix, and calculating the contribution rate of the principal component feature matrix to obtain a feature contribution rate sequence; performing dimensionality reduction processing on the principal component feature matrix according to the feature contribution rate sequence to obtain preliminary dimensionality reduction data, and performing wavelet transform on the preliminary dimensionality reduction data to obtain multi-scale decomposition coefficients; performing energy distribution analysis on the multi-scale decomposition coefficients to obtain energy distribution characteristics, and reconstructing the multi-scale decomposition coefficients according to the energy distribution characteristics to obtain reconstructed feature data; performing nonlinear mapping on the reconstructed feature data to obtain the low-dimensional feature vector; S3, performing time series analysis and data fusion on the low-dimensional feature vector to obtain real-time state estimation and future state prediction information of the high-power battery pack; S3 step includes: performing time series decomposition on the low-dimensional feature vector to obtain trend component, periodic component and random component, and performing trend extrapolation on the trend component to obtain trend prediction result; performing periodic analysis on the periodic component to obtain periodic characteristic parameters, and performing periodic prediction on the periodic component according to the periodic characteristic parameters to obtain periodic prediction result; performing autoregressive analysis on the random component to obtain autoregressive model parameters, and performing random prediction on the random component according to the autoregressive model parameters to obtain random prediction result; performing weighted fusion on the trend prediction result, the periodic prediction result and the random prediction result to obtain real-time state estimation and future state prediction information of the high-power battery pack; S4, performing multi-dimensional parameter mapping on the real-time state estimation and the future state prediction information to generate a decision space matrix, and traversing and optimizing the decision space matrix through a dynamic programming algorithm to obtain a real-time charging and discharging plan; Step S4 includes: performing parameter extraction on the real-time state estimation and the future state prediction information to obtain a key state parameter set, and quantizing the key state parameter set to obtain a quantized parameter vector; performing multi-dimensional mapping on the quantized parameter vector to obtain an initial decision space, and gridding the initial decision space to obtain the decision space matrix; performing state transfer calculation on the decision space matrix through a dynamic programming algorithm to obtain an optimal state transfer path, and generating the real-time charging and discharging plan according to the optimal state transfer path; S5. Controlling the high-power battery pack to perform charging and discharging operations according to the real-time charging and discharging scheme, collecting battery pack operation data of the high-power battery pack in real time, and performing parallel analysis on the battery pack operation data to obtain multi-dimensional charging and discharging index data, and dynamically calibrating the multi-dimensional charging and discharging index data through an adaptive threshold algorithm to obtain real-time safety status indicators and potential risk warning signals; S6. Perform a multi-objective trade-off analysis on the potential risk warning signal and the real-time safety status indicator to generate a target decision space, and perform a traversal analysis on the target decision space through a heuristic search algorithm to obtain a battery pack management control instruction sequence.

2. The artificial intelligence-based high-power battery pack charge and discharge control method according to claim 1, characterized in that: Step S1 includes: Performing high-frequency sampling on the voltage, current, temperature and internal resistance of the high-power battery pack to obtain the sampled data, and performing noise removal processing on the sampled data to obtain filtered sampled data; Dividing the filtered sampling data into time windows to obtain a plurality of time window data segments, and extracting features from the plurality of time window data segments to obtain preliminary feature data; Normalizing the preliminary feature data to obtain standardized feature data, and performing correlation analysis on the standardized feature data to obtain a feature correlation matrix; Performing feature selection on the standardized feature data according to the feature correlation matrix to obtain a candidate feature set, and performing time series alignment on the candidate feature set to obtain aligned feature data; Performing interpolation processing on the aligned feature data to obtain equally spaced sampling data, and performing time series reconstruction on the equally spaced sampling data to obtain reconstructed time series data; The reconstructed time series data is subjected to data enhancement processing to obtain the multidimensional time series data set.

3. The artificial intelligence-based high-power battery pack charge and discharge control method according to claim 1, characterized in that: Step S5 includes: Controlling the charging and discharging of the high-power battery pack according to the real-time charging and discharging scheme to obtain real-time charging and discharging data, and collecting and preprocessing the real-time charging and discharging data to obtain the battery pack operation data; Performing multi-dimensional feature extraction on the battery pack operation data to obtain the multi-dimensional charge and discharge index data, and performing statistical analysis on the multi-dimensional charge and discharge index data to obtain index distribution characteristics; According to the indicator distribution characteristics, a dynamic safety threshold is calculated by an adaptive threshold algorithm to obtain a threshold adjustment parameter; The multi-dimensional charge and discharge index data is calibrated according to the threshold adjustment parameter to obtain the real-time safety status index and the potential risk warning signal.

4. The artificial intelligence-based high-power battery pack charge and discharge control method according to claim 3 is characterized in that: Step S6 includes: Weights are assigned to the potential risk warning signal and the real-time safety status index to obtain a weighted index set, and multi-objective optimization calculation is performed on the weighted index set to obtain an alternative decision-making plan; Conducting feasibility assessment on the alternative decision plans to obtain a feasible decision set, and constructing a decision space for the feasible decision set to obtain the target decision space; The target decision space is traversed and searched by a heuristic search algorithm to obtain an optimal decision path, and the battery pack management control instruction sequence is generated according to the optimal decision path.

5. A high-power battery pack charge and discharge control system based on artificial intelligence, used to execute the high-power battery pack charge and discharge control method based on artificial intelligence as claimed in any one of claims 1 to 4, characterized in that: include: A sampling module, used to perform high-frequency sampling on a high-power battery pack to obtain sampling data, and to perform segmented processing on the sampling data to obtain a multi-dimensional time series data set, wherein the sampling data includes voltage, current, temperature and internal resistance; A decomposition module, used for performing dimensionality reduction and multi-scale decomposition on the multi-dimensional time series data set to obtain a low-dimensional feature vector capable of characterizing the dynamic characteristics of a high-power battery pack, including: performing principal component analysis on the multi-dimensional time series data set to obtain a principal component feature matrix, and performing contribution rate calculation on the principal component feature matrix to obtain a feature contribution rate sequence; performing dimensionality reduction processing on the principal component feature matrix according to the feature contribution rate sequence to obtain preliminary dimensionality reduction data, and performing wavelet transform on the preliminary dimensionality reduction data to obtain multi-scale decomposition coefficients; performing energy distribution analysis on the multi-scale decomposition coefficients to obtain energy distribution characteristics, and reconstructing the multi-scale decomposition coefficients according to the energy distribution characteristics to obtain reconstructed feature data; performing nonlinear mapping on the reconstructed feature data to obtain the low-dimensional feature vector; A fusion module, for performing time series analysis and data fusion on the low-dimensional feature vector to obtain real-time state estimation and future state prediction information of the high-power battery pack, including: performing time series decomposition on the low-dimensional feature vector to obtain trend components, periodic components and random components, and performing trend extrapolation on the trend components to obtain trend prediction results; performing periodic analysis on the periodic components to obtain periodic characteristic parameters, and performing periodic prediction on the periodic components according to the periodic characteristic parameters to obtain periodic prediction results; performing autoregressive analysis on the random components to obtain autoregressive model parameters, and performing random prediction on the random components according to the autoregressive model parameters to obtain random prediction results; performing weighted fusion on the trend prediction results, the periodic prediction results and the random prediction results to obtain real-time state estimation and future state prediction information of the high-power battery pack; A mapping module, used for performing multi-dimensional parameter mapping on the real-time state estimation and the future state prediction information to generate a decision space matrix, and traversing and optimizing the decision space matrix through a dynamic programming algorithm to obtain a real-time charging and discharging scheme, including: performing parameter extraction on the real-time state estimation and the future state prediction information to obtain a key state parameter set, and quantizing the key state parameter set to obtain a quantized parameter vector; performing multi-dimensional mapping on the quantized parameter vector to obtain an initial decision space, and gridding the initial decision space to obtain the decision space matrix; performing state transfer calculation on the decision space matrix through a dynamic programming algorithm to obtain an optimal state transfer path, and generating the real-time charging and discharging scheme according to the optimal state transfer path; A calibration module, used to control the high-power battery pack to perform charging and discharging operations according to the real-time charging and discharging scheme, collect battery pack operation data of the high-power battery pack in real time, and perform parallel analysis on the battery pack operation data to obtain multi-dimensional charging and discharging index data, and dynamically calibrate the multi-dimensional charging and discharging index data through an adaptive threshold algorithm to obtain real-time safety status indicators and potential risk warning signals; The traversal module is used to perform a multi-objective trade-off analysis on the potential risk warning signal and the real-time safety status indicator, generate a target decision space, and perform a traversal analysis on the target decision space through a heuristic search algorithm to obtain a battery pack management control instruction sequence.

Images