Battery performance attenuation test method, device and equipment for energy storage battery

Abstract

The invention relates to the technical field of battery performance testing, and discloses a battery performance attenuation testing method, device and equipment for an energy storage battery. The method comprises: constructing a battery performance measurement data matrix of an energy storage battery during a charging and discharging process, and calculating an optimal value of a noise variance through a zero-mean censored normal distribution model to obtain a battery performance noise characteristic matrix; performing mutual information calculation on performance parameters to obtain a key feature set of battery performance, and constructing a battery attenuation characteristic vector of an operating condition parameter and a performance attenuation rate; extracting a weighted adaptive distance of a time interval, an operating condition difference and a performance difference according to the battery attenuation characteristic vector, and creating a battery attenuation characteristic association network; performing performance attenuation mode classification based on the battery attenuation characteristic association network and the battery attenuation characteristic vector to obtain a classification result of the performance attenuation mode of the energy storage battery. The invention comprehensively extracts the battery performance attenuation characteristics and realizes accurate classification of the performance attenuation mode.

Description

Technical Field

[0001] The present invention relates to the technical field of battery performance testing, and in particular to a method, device and equipment for testing battery performance attenuation of an energy storage battery. Background Art

[0002] With the rapid development of the new energy industry, energy storage batteries are increasingly used in power systems, and accurate evaluation of their performance degradation characteristics is of great significance to ensure the safe operation of energy storage systems. At present, energy storage batteries face complex and changeable working conditions during actual operation, which causes battery performance degradation to show nonlinear and time-varying characteristics, making it difficult for traditional performance degradation evaluation methods to accurately capture the dynamic process of battery performance degradation.

[0003] During the performance degradation test of energy storage batteries, the measured data often contains a lot of noise information, which comes from the accuracy error of the measuring equipment, environmental interference, operating condition fluctuations, etc. At the same time, due to the complex coupling relationship between the performance parameters of energy storage batteries, it is difficult to extract effective performance degradation characteristics from the original test data, which seriously affects the accuracy of judging the performance degradation trend. Summary of the invention

[0004] The present invention provides a battery performance attenuation test method, device and equipment for an energy storage battery. The present invention comprehensively extracts battery performance attenuation characteristics and realizes accurate classification of performance attenuation modes.

[0005] In a first aspect, the present invention provides a method for testing battery performance attenuation of an energy storage battery, the method comprising: Construct the battery performance measurement data matrix of the energy storage battery during the charging and discharging process, and calculate the optimal value of the noise variance through the zero-mean censored normal distribution model to obtain the battery performance noise characteristic matrix; Based on the battery performance noise characteristic matrix, the mutual information of performance parameters is calculated to obtain a key characteristic set of battery performance, and a battery attenuation characteristic vector of operating condition parameters and performance attenuation rate is constructed; Extracting the weighted adaptive distance of the time interval, the operating condition difference and the performance difference according to the battery attenuation feature vector, and creating a battery attenuation feature association network; Based on the battery attenuation feature association network and the battery attenuation feature vector, performance attenuation mode classification is performed to obtain a classification result of the energy storage battery performance attenuation mode.

[0006] In a second aspect, the present invention provides a battery performance degradation test device for an energy storage battery, the battery performance degradation test device for an energy storage battery comprising: A construction module is used to construct a battery performance measurement data matrix of the energy storage battery during the charging and discharging process, and calculate the optimal value of the noise variance through a zero-mean censored normal distribution model to obtain a battery performance noise characteristic matrix; A calculation module, used to calculate the mutual information of performance parameters based on the battery performance noise characteristic matrix, obtain a key characteristic set of battery performance, and construct a battery attenuation characteristic vector of operating condition parameters and performance attenuation rate; Creating a module for extracting a weighted adaptive distance of a time interval, a working condition difference, and a performance difference according to the battery attenuation feature vector, and creating a battery attenuation feature association network; A classification module is used to classify the performance attenuation mode based on the battery attenuation feature association network and the battery attenuation feature vector to obtain a classification result of the energy storage battery performance attenuation mode.

[0007] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned battery performance degradation test method for the energy storage battery.

[0008] In the technical solution provided by the present invention, the present invention models the measurement noise by adopting a zero-mean censored normal distribution model, and calculates the optimal value of the noise variance in combination with the maximum entropy principle, thereby achieving effective elimination of noise in the measurement data, and at the same time adopts Butterworth low-pass filtering and outlier processing to ensure data quality; through a feature dimensionality reduction method based on mutual information theory and a sequential forward selection algorithm, as well as an integrated feature measurement function of operating parameters and performance decay rate, the battery performance decay characteristics are comprehensively extracted; a weighted adaptive distance calculation and a community structure adjacency matrix construction method are introduced, combined with the Louvain community discovery algorithm, to reveal the intrinsic connection between the performance decay characteristics; a multi-scale network feature extraction method is adopted to characterize the performance decay characteristics from three levels: global, node, and community, and the radial basis kernel function support vector machine classifier is used to achieve accurate classification of performance decay patterns. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0010] Figure 1 A schematic diagram of the steps of a method for testing battery performance attenuation of an energy storage battery in an embodiment of the present invention; Figure 2It is a structural schematic diagram of a battery performance attenuation testing device for an energy storage battery in an embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0011] Embodiments of the present invention provide a method, device and apparatus for testing battery performance attenuation of an energy storage battery. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0012] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of a battery performance attenuation test method of an energy storage battery in an embodiment of the present invention includes: Step S1, constructing a battery performance measurement data matrix of the energy storage battery during the charging and discharging process, and calculating the optimal value of the noise variance through a zero-mean censored normal distribution model to obtain a battery performance noise characteristic matrix; It is understandable that the execution subject of the present invention may be a battery performance degradation test device for an energy storage battery, or may be a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0013] Specifically, the voltage, current, temperature, internal resistance and state of charge of the energy storage battery during the charging and discharging process are collected, and these data are recorded at a certain time interval to form an original data sequence. The original data sequence is segmented, and the entire data set is divided into multiple subsequences according to the time axis to form a time series data set. Signal processing is performed on the time series data set to filter out high-frequency noise and random interference and improve the smoothness and stability of the data. Signal processing methods include wavelet transform, median filtering, Kalman filtering, etc., among which wavelet transform effectively removes burst noise, and Kalman filtering is suitable for processing random noise in dynamic systems. In the filtering process, the appropriate filter window and parameters are selected according to the data characteristics to ensure that the noise suppression effect does not lead to the loss of useful signals, and the filtered data set is obtained. The outliers in the filtered data set are identified and corrected. Outliers are manifested as mutation points, drift points or outliers. Statistical methods, such as Z score detection, box plot analysis, etc., or combined with machine learning methods, such as isolation forest or LOF (local outlier factor) algorithm, are used to more accurately judge outliers. Adopt appropriate correction strategies, such as interpolation using data from adjacent moments, or regression prediction based on historical trends, to fill in abnormal data points and obtain a corrected data set. All performance parameters in the corrected data set are normalized to eliminate dimensional differences so that different features have the same scale in subsequent calculations. Through normalization, some parameters are prevented from having too much impact on the analysis results due to their large numerical range, thereby improving the stability and computational efficiency of the model and obtaining a normalized data set. The normalized data set is reconstructed into a matrix using the sampling time points as row vectors and the performance parameters as column vectors to obtain a battery performance measurement data matrix that reflects the trend of battery performance over time. A zero-mean censored normal distribution model is constructed based on the battery performance measurement data matrix to characterize the noise characteristics of battery performance data. The zero-mean censored normal distribution model is a statistical model suitable for processing noisy data. It assumes that the noise in the data follows a normal distribution, and returns the data mean to zero by censoring outliers, thereby improving the accuracy of noise estimation. In the process of building the model, the mean and variance of the battery performance measurement data matrix are calculated, and the extreme values ​​are removed by deleting abnormal data points to make the data more consistent with the assumption of normal distribution. On this basis, the maximum entropy principle is used to calculate the optimal value of the noise variance. Among all possible probability distributions, the distribution that satisfies the known constraints and has the largest entropy is selected to ensure that the model minimizes the assumptions about unknown information. In the calculation process, the sample entropy is calculated based on the battery performance measurement data matrix, and the maximum entropy distribution is solved under the constraints to obtain the optimal variance estimate of the noise. The battery performance noise characteristic matrix is ​​obtained by this method.

[0014] The battery performance measurement data matrix is ​​processed by data block processing, and is divided according to fixed time intervals or based on the changes in the battery working state, so that each data block represents a relatively stable working condition. Each data block is centralized, that is, the mean of each performance parameter in the data block is calculated, and the corresponding mean is subtracted from the original data, so that the mean of the processed data block is zero, and a zero-mean data block is obtained. The zero-mean data block is divided into censored intervals to obtain censored intervals. Statistical analysis methods are used, such as setting the censoring range based on the standard deviation threshold, or using an adaptive method to dynamically determine the censoring interval according to the change trend within the data block. Through the censoring interval division, the extremely deviated noise points are effectively eliminated while maintaining the representativeness of the data. Based on the censoring interval division, a zero-mean censored normal distribution model is constructed. The basic assumption of the model is that the true distribution of the data is approximately described by the normal distribution, and the noise part is adjusted through the censoring operation, so that the model more accurately reflects the actual statistical characteristics of the battery performance data. The maximum entropy principle is used to construct the noise variance optimization objective function and numerically optimize the zero mean censored normal distribution model to obtain the optimal value of the noise variance. The maximum entropy principle is to select the distribution with the highest entropy value as the optimal distribution among all possible probability distributions, because maximizing entropy means minimizing the assumption of unknown information, thereby avoiding artificial bias. In the specific implementation process, the entropy value of the battery performance measurement data matrix is ​​calculated, and constraints such as the mean and variance range of the data are established, and then the probability distribution that maximizes the entropy value is solved under these constraints. Through numerical optimization methods, such as the Lagrange multiplier method or the gradient descent algorithm, the optimal value of the noise variance is efficiently calculated and used for noise analysis. A white noise channel model is constructed based on the zero mean censored normal distribution model and the optimal value of the noise variance. The white noise channel model can simulate the noise propagation process of battery performance data and estimate the impact of different noise sources on the data. By constructing this model, the systematic changes and random noise components in the battery performance data are effectively separated, thereby more accurately describing the evolution trend of battery performance over time. After completing the construction of the white noise channel model, the noise distribution characteristics are calculated to extract the noise entropy value of each performance parameter. The noise entropy value measures the degree to which different performance parameters are affected by noise. The higher the entropy value, the greater the random fluctuation of the parameter and the higher the possibility of being interfered by noise. In order to calculate the noise entropy value, the information entropy of the noise distribution characteristics is calculated based on the Shannon entropy formula to obtain the noise entropy value of each performance parameter. The noise entropy value is constructed into a noise entropy feature set, and the noise entropy feature set is associated and mapped with the corresponding performance parameters to establish the relationship between the noise characteristics and the battery performance. On this basis, a matrix transformation operation is performed to integrate the noise entropy characteristics of multiple performance parameters, and finally the battery performance noise feature matrix is ​​obtained.

[0015] Step S2, calculating the mutual information of performance parameters based on the battery performance noise characteristic matrix, obtaining a key characteristic set of battery performance, and constructing a battery attenuation characteristic vector of operating condition parameters and performance attenuation rate; Specifically, edge entropy is calculated for each performance parameter in the battery performance noise characteristic matrix. Edge entropy is an important indicator for measuring the amount of information of a single parameter, reflecting the distribution uncertainty of the parameter itself. Joint entropy is constructed for edge entropy to measure the coupling relationship between different parameters and form a performance parameter entropy value set. The calculation of joint entropy is based on the joint probability distribution of multiple parameters to ensure that the dependency between multiple parameters is considered. A mutual information calculation formula is constructed based on the performance parameter entropy value set. Mutual information measures the degree of information sharing between two performance parameters, that is, whether the amount of information of one parameter can be predicted by another parameter. The calculation of mutual information uses edge entropy and joint entropy. Specifically, the mutual information matrix is ​​obtained by calculating the joint entropy of each performance parameter pair and combining the edge entropy of a single parameter. Each element of the mutual information matrix represents the size of the mutual information between the two parameters. The mutual information matrix is ​​optimized, and the corresponding noise entropy value is subtracted from the mutual information matrix to eliminate the interference of noise on the information relationship between parameters, and the net information characteristics of the performance parameters are obtained. Based on the net information characteristics of the performance parameters, correlation metrics and redundancy are calculated to analyze the information relationship between different performance parameters. The correlation measure is used to measure the dependency between parameters, while the redundancy is used to eliminate parameters with high information overlap to ensure that the final selected features are the most representative. Therefore, a feature evaluation function is constructed, which comprehensively considers the net information feature, correlation measure and redundancy to evaluate the importance of each performance parameter in the battery performance degradation. The sequential forward selection algorithm is executed on the feature evaluation function, and the feature with the largest F value increment is selected in each iteration, and the feature set is continuously expanded. The F value increment refers to the contribution of the newly added feature to the overall information amount. The larger the increment, the more obvious the role of the feature in describing the battery performance degradation mode. In each iteration, the F value is calculated, and the features with F values ​​greater than the first target value are marked as candidate features to form a feature candidate set. The feature candidate set contains a set of performance parameters that may be the most representative. The feature candidate set is subjected to correlation analysis to screen out the most representative feature combination. The correlation coefficient between the candidate features is calculated, and the feature combination with the absolute value of the correlation coefficient greater than the second target value is selected to ensure that the final selected features are complementary in terms of information amount, rather than highly correlated redundant features. Through this process, the key feature set of battery performance is obtained. Based on the key feature set of battery performance, an integrated feature measurement function of operating parameters and performance attenuation rate is constructed to comprehensively measure the impact of different features on battery attenuation. The construction of the integrated feature measurement function adopts a weighted method, in which the weight allocation is based on the feature importance measurement, so that the key features occupy a larger weight ratio in the calculation process. After the integrated feature measurement function is constructed, the local extreme points of the function are processed by the density clustering method to identify the key patterns of battery attenuation characteristics. The density clustering method identifies the cluster center based on the distribution density of the data, so as to more effectively extract the representative patterns of battery attenuation characteristics.Through this step, the battery attenuation characteristic vector is obtained to describe the attenuation characteristics of the battery under different working conditions.

[0016] The operating parameters are extracted from the key characteristics of battery performance, and the corresponding performance decay rate is calculated. The operating parameters include the battery's charge and discharge rate, temperature change, SOC (state of charge) range, etc., while the performance decay rate involves key indicators such as capacity retention rate and internal resistance growth rate that characterize the battery's health status. After extracting these parameters, a balance factor is set to establish a trade-off relationship between the operating parameters and the performance decay rate to ensure that the constructed integrated feature metric function fully reflects the battery's operating status and decay characteristics. The balance factor is selected based on the importance of different parameters on the battery life, and all features are integrated into a metric function in a weighted manner to form a unified set of evaluation criteria. A sliding time window is set for the integrated feature metric function to dynamically capture the time characteristics of battery performance decay. The setting of the sliding time window helps to perform localized analysis of the data so as to more finely identify the performance change trend. In each time window, the local mean and standard deviation are calculated to obtain the fluctuation characteristics of the data on a short time scale. The local mean reflects the central value of the feature within the time period, while the standard deviation is used to measure the degree of fluctuation of the data. Through these local statistical features, the trend of battery performance evolution over time is determined. Based on the local statistical features, local threshold conditions are set to screen out the key time points representing the battery decay mode. The threshold range is set so that the time points where the local mean and standard deviation change beyond a certain range are marked as candidate feature points. These candidate feature points correspond to the key moments when the battery performance changes during operation. Through this process, the most representative time points in the battery decay process are preliminarily screened out. The density clustering algorithm is performed on the candidate feature point set with a preset time span as the clustering radius to identify the patterns existing in the battery decay process. Based on the density distribution of data points, high-density areas are divided into clusters, and isolated points are filtered out to ensure that the obtained feature points are highly representative. In order to perform density clustering, a time span is preset as the clustering radius, and cluster division is performed based on point density and distance thresholds. If the number of candidate feature points around a certain time point exceeds a certain threshold, the point is classified as a core point, and the nearby candidate points are attributed to the same cluster. Through this step, the feature points in a similar time range are aggregated together to form a series of stable decay patterns. Based on the characteristic point clustering results, the center point of each cluster is selected, and the timestamps, operating parameters, and performance parameters corresponding to these center points are recorded. The center point is selected by the point with the highest density in the cluster, or the point in the cluster that best represents the overall trend is selected to ensure that the selected characteristic point accurately reflects the battery's attenuation process. Each center point corresponds to a critical moment in the battery's performance change, and is associated with specific operating conditions and performance parameters to form a sequence of key characteristic points. These key characteristic points are reconstructed in chronological order and expressed in vector form to obtain the battery attenuation feature vector.

[0017] Step S3, extracting the weighted adaptive distance of the time interval, operating condition difference and performance difference according to the battery attenuation feature vector, and creating a battery attenuation feature association network; Specifically, adjacent feature points are identified in the battery attenuation feature vector, and the time intervals, operating condition differences, and performance differences between them are calculated. The time interval reflects the time characteristics of battery attenuation, the operating condition differences describe the changes in the battery under different operating conditions, and the performance differences measure the fluctuations in the health status of the battery at different time points. These three differences are weighted so that the contributions of different factors are dynamically adjusted according to the specific usage scenarios of the battery. The weighted adaptive distance calculation method uses normalized Euclidean distance or Mahalanobis distance to ensure that each factor is comparable on the same numerical scale, and forms a feature point distance matrix, each element of which represents the weighted distance between any two feature points. A connection threshold is set for the feature point distance matrix to determine which feature points are close enough to establish an association. The connection threshold is set by the statistical characteristics of the data or the empirical threshold, so that most feature points with similar attenuation patterns are connected. For all feature point pairs whose distance is less than the threshold, they are marked as connection point pairs to form an initial connection relationship set. Based on the initial connection relationship set, an adjacency matrix A is constructed, where the matrix element A ijWhen feature points i and j are connected, the value is 1, otherwise the value is 0, forming a binary adjacency matrix. In the process of constructing the adjacency matrix, sparsity control is performed to ensure that the network structure has a certain degree of connectivity and to avoid affecting the accuracy of community division due to too many isolated nodes. The degree centrality and betweenness centrality are calculated for each node in the binary adjacency matrix. Degree centrality indicates the number of neighbors directly connected to a node, and is an important indicator for measuring the direct influence of a node in the network, while betweenness centrality describes the degree to which the node acts as a bridge in the network, that is, how many shortest paths it plays a connecting role. The calculated degree centrality and betweenness centrality are normalized and weighted summed to obtain the node importance index, which measures the core degree of different feature points in the entire network. Based on the node importance index, the Louvain community discovery algorithm is executed on the binary adjacency matrix, and the community division is performed through modularity Q value optimization. The Louvain algorithm is an efficient modular optimization algorithm that automatically detects high-density connected node groups, i.e., communities, based on the network topology. The modularity Q value is a key indicator to measure the quality of community division. The higher the value, the higher the connection density within the divided community, and the sparser the connection between different communities. When executing the Louvain algorithm, the modularity Q value is continuously optimized to obtain the optimal community division result, thereby ensuring that battery feature points with different attenuation modes can belong to different communities. The topological characteristics of the network community structure are combined with the binary adjacency matrix to construct a battery attenuation feature association network. The network community structure provides clustering information of feature points, while the adjacency matrix describes the connection relationship between points. The combination of the two effectively characterizes the association pattern of battery attenuation characteristics in time and space.

[0018] Step S4: classify the performance attenuation mode based on the battery attenuation feature association network and the battery attenuation feature vector to obtain a classification result of the energy storage battery performance attenuation mode.

[0019] Specifically, global topological features are extracted from the battery attenuation feature association network, and the network global structural feature vector is constructed based on these topological information. Global topological features include average shortest path length, network diameter, network density and global clustering coefficient, where the average shortest path length measures the average distance between nodes in the network, reflecting the connectivity of the network, the network diameter represents the maximum shortest path length, characterizing the overall expansion range of the network, and the network density is used to describe the ratio between the actual edges in the network and the maximum number of edges that may exist, reflecting the compactness of the network, while the global clustering coefficient measures the proportion of triangular relationships between nodes to describe the overall aggregation of the network. By calculating these global topological features and integrating them into the global structural feature vector of the network, the macroscopic structural characteristics of the battery attenuation feature association network are comprehensively characterized. The node-level topological features of the battery attenuation feature association network are calculated and spliced ​​with the global structural feature vector to construct a node feature matrix. Node-level topological features include node degree distribution, betweenness centrality, and eigenvector centrality. Node degree distribution describes the distribution of the number of connections of each node, reflecting the difference in the importance of nodes in the network. Betweenness centrality is used to measure the bridge role of the node in the shortest path and characterize the control ability of the node in the network, while eigenvector centrality is measured based on the centrality of neighboring nodes to measure the influence of the node in the entire network. By calculating these node-level topological features and splicing them with the global topological features into the same feature matrix, it is ensured that the model considers both the overall network structure and the characteristics of local nodes when classifying the decay mode. Community-level topological features are extracted from the battery decay feature association network, and the features are combined with the node feature matrix to obtain a multi-scale network feature vector. Community-level topological features mainly include the number of communities, community size distribution, and inter-community connection strength. The number of communities represents the number of communities identified in the network, reflecting the diversity of decay modes. The community size distribution is used to describe the number of nodes contained in different communities, thereby revealing the scale characteristics of decay modes, while the inter-community connection strength measures the degree of association between different communities, reflecting the transition characteristics of batteries between different decay modes. By combining these community-level topological features with the node feature matrix, a multi-scale network feature vector is formed to more comprehensively describe the structural characteristics of the battery attenuation pattern. The multi-scale network feature vector and the battery attenuation feature vector are input into the input layer of the radial basis kernel function support vector machine classifier to form standardized input features. The radial basis kernel function support vector machine constructs the best classification hyperplane in the high-dimensional feature space to achieve the distinction between different attenuation modes. When inputting features, all features are normalized to ensure that features of different dimensions are compared on the same scale and to avoid excessive impact of features with too large a numerical range on model training. The radial basis kernel function is used to perform nonlinear mapping on the standardized input features, converting the original features into a high-dimensional space to improve the separability of the classification.After completing the kernel function mapping of the features, the classification decision function needs to be executed based on the obtained kernel function mapping features to classify the performance attenuation mode of the energy storage battery. The classification decision function optimizes the classification hyperplane of the support vector machine and combines the distance information between different modes to maximize the interval between categories, thereby improving the robustness of the classification. Through this classification model, different energy storage battery performance attenuation modes can be accurately identified and the corresponding classification results can be output.

[0020] In an embodiment of the present invention, the present invention models the measurement noise by adopting a zero-mean censored normal distribution model, and calculates the optimal value of the noise variance in combination with the maximum entropy principle, thereby achieving effective elimination of noise in the measurement data, and at the same time adopts Butterworth low-pass filtering and outlier processing to ensure data quality; through a feature dimensionality reduction method based on mutual information theory and a sequential forward selection algorithm, as well as an integrated feature measurement function of operating parameters and performance decay rate, the battery performance decay characteristics are comprehensively extracted; a weighted adaptive distance calculation and a community structure adjacency matrix construction method are introduced, combined with the Louvain community discovery algorithm, to reveal the intrinsic connection between the performance decay characteristics; a multi-scale network feature extraction method is adopted to characterize the performance decay characteristics from three levels: global, node, and community, and the radial basis kernel function support vector machine classifier is used to achieve accurate classification of performance decay patterns.

[0021] In a specific embodiment, the process of executing step S1 may specifically include the following steps: Collect data on the voltage, current, temperature, internal resistance and state of charge of the energy storage battery during the charging and discharging process to obtain the original data sequence; The original data sequence is segmented to obtain a time series data set, and the time series data set is signal processed to obtain a filtered data set; Identifying and correcting outliers in the filtered data set to obtain a corrected data set, and normalizing all performance parameters in the corrected data set to obtain a normalized data set; Reconstruct the matrix of the normalized data set by taking the sampling time points as row vectors and the performance parameters as column vectors to obtain the battery performance measurement data matrix; A zero-mean censored normal distribution model is constructed based on the battery performance measurement data matrix, and the optimal value of the noise variance is calculated using the maximum entropy principle to obtain the battery performance noise characteristic matrix.

[0022] Specifically, the voltage is collected in real time during the battery operation. , Current ,temperature 、Internal resistance and state of charge As a basic performance parameter. Assume that at time The observation data of each sampling point is expressed as:

[0023] Then the original data sequence in the entire test cycle is expressed as:

[0024] in, Indicates the total number of sampling points. Segment the original data sequence. Based on a fixed time window Or the start and end points of the charge and discharge cycle. For example, if a fixed time window is used, the time series data set obtained after division is expressed as:

[0025] in, Indicates A data set for a period of time, is the number of sampling points contained in each segment. Perform signal processing on the time series data set to filter out high-frequency interference components. Use low-pass filtering to effectively smooth the data, such as voltage data. For sliding mean filtering, use the following formula:

[0026] in, is the smoothed voltage data, is the window width, which determines the smoothness of the filter. ,temperature Other parameters are filtered accordingly to obtain a filtered data set. Outliers in the filtered data set are identified and corrected. Outliers are detected using the Z score method, that is, the mean of each performance parameter is calculated. and standard deviation , and then determine whether a data point exceeds the set threshold (Usually take ), the outlier detection formula is as follows:

[0027] like , then It is considered as an abnormal point and needs to be corrected. The correction method uses linear interpolation, that is:

[0028] in, is the corrected data point. After this process, the corrected data set is obtained. The corrected data set is normalized. The minimum-maximum normalization is used to map all data to the [0,1] interval. The calculation formula is:

[0029] in, and are the minimum and maximum values ​​of the performance parameter, respectively. After normalization, all performance parameters are in the same numerical range to avoid parameters with a larger numerical range from having too much impact on subsequent calculations. All normalized data are reconstructed with the sampling time point as the row vector and the performance parameter as the column vector to obtain the battery performance measurement data matrix:

[0030] Based on the data matrix, a zero-mean censored normal distribution model is constructed, and the optimal value of the noise variance is calculated using the maximum entropy principle. Perform zero mean processing, that is, calculate the mean of each column And subtract the mean so that the mean of each column of the data matrix is ​​zero:

[0031] Set the deletion area for the missing data points, remove the abnormal data through statistical analysis methods, and assume that the noise follows a normal distribution In order to determine the optimal noise variance , using the maximum entropy principle, define the entropy function:

[0032] in, is the probability density function of the censored normal distribution:

[0033] maximize Find the best The value is set to maximize the entropy of the noise and obtain the noise distribution parameters with the best information balance. Based on the optimized noise parameters, a battery performance noise characteristic matrix is ​​constructed, which effectively describes the noise characteristics in the battery data.

[0034] In a specific embodiment, the execution step constructs a zero-mean censored normal distribution model based on the battery performance measurement data matrix, and calculates the optimal value of the noise variance by the maximum entropy principle to obtain the battery performance noise characteristic matrix. The process may specifically include the following steps: The battery performance measurement data matrix is ​​divided into blocks to obtain multiple data blocks, and each data block is centrally processed to obtain a zero-mean data block; The zero-mean data block is divided into censored intervals to obtain censored intervals, and a zero-mean censored normal distribution model is constructed based on the censored intervals; The maximum entropy principle is used to construct the noise variance optimization objective function and numerically optimize the zero-mean censored normal distribution model to obtain the optimal value of the noise variance. A white noise channel model is constructed based on the zero-mean censored normal distribution model and the optimal value of noise variance, and the noise distribution characteristics are obtained by performing noise analysis on the battery performance measurement data matrix; The noise entropy value of each performance parameter is calculated based on the noise distribution characteristics, and the noise entropy value is constructed into a noise entropy feature set. The noise entropy feature set is then associated with the corresponding performance parameters and matrix transformation is performed to obtain the battery performance noise characteristic matrix.

[0035] Specifically, the battery performance measurement data matrix is ​​divided into blocks. Assume that the battery performance measurement data matrix is ,in Represents at time The first performance parameters, such as voltage , Current ,temperature 、Internal resistance and state of charge . The matrix is ​​represented as:

[0036] in is the number of time sampling points, and the number of columns is different battery performance parameters. Divide into blocks according to the time axis, each data block contains consecutive time points, namely:

[0037] in Representative data blocks, each block is of size Each data block is centralized, that is, the mean of each parameter is calculated and subtracted to return the mean of the data block to zero. A performance parameter in a data block The mean of is:

[0038] Then the zero-mean data block after centering is expressed as:

[0039] The zero-mean data block is divided into censored intervals. The censored interval is used to identify outliers and high-noise areas in the data. It is set based on the standard deviation threshold, that is, if the absolute value of a data point exceeds its standard deviation, times, then the point is considered to belong to the censoring interval. Let the standard deviation be:

[0040] The censoring interval is defined as:

[0041] in, Usually 3 is taken to ensure that the censored data points are outliers. On this basis, a zero-mean censored normal distribution model is constructed. Assuming that the censored data follows a censored normal distribution, its probability density function is:

[0042] Based on this model, the maximum entropy principle is used to optimize the noise variance An estimate of . Entropy is defined as:

[0043] In order to obtain the maximum entropy , using numerical optimization methods, such as gradient descent or the method of Lagrange multipliers, to solve:

[0044] Get the optimal noise variance , which is the noise variance under the maximum entropy condition. Based on the optimal noise variance, a white noise channel model is constructed, and the white noise channel is expressed as:

[0045] in, is the observation data, It is a real signal. is zero-mean white noise. Perform noise analysis on the battery performance measurement data matrix, calculate the noise distribution characteristics, and define the variance of the noise as:

[0046] The noise entropy value of each performance parameter is calculated based on the noise distribution characteristics. The noise entropy value measures the randomness of the noise and is calculated using Shannon entropy:

[0047] in, is the probability density function of the observed data. All noise entropy values ​​are constructed into a noise entropy feature set:

[0048] The noise entropy feature set is associated and mapped with the corresponding performance parameters to construct the noise feature matrix:

[0049] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Calculate the edge entropy of the performance parameters in the battery performance noise feature matrix, and perform joint entropy construction on the edge entropy to obtain a performance parameter entropy value set; A mutual information calculation formula is constructed based on the performance parameter entropy value set, and mutual information is calculated for all performance parameter pairs to obtain a mutual information matrix; Subtract the corresponding noise entropy value from the mutual information matrix and construct the net information feature of the performance parameter, calculate the correlation measure and redundancy based on the net information feature of the performance parameter, and construct the feature evaluation function; The sequential forward selection algorithm is executed on the feature evaluation function. The feature with the largest F value increment is selected in each iteration, and the features with F value greater than the first target value are marked as candidate features to obtain a feature candidate set. Perform correlation analysis on the feature candidate set, select the feature combination whose absolute value of the correlation coefficient is greater than the second target value, and obtain the key feature set of battery performance; An integrated feature metric function of operating parameters and performance attenuation rate is constructed according to the key feature set of battery performance, and the local extreme points of the integrated feature metric function are processed by density clustering to obtain the battery attenuation feature vector.

[0050] Specifically, based on the battery performance noise characteristic matrix Estimation of probability distribution. Assume Include samples, each sample corresponds to performance parameters, namely:

[0051] in Indicates The sample performance parameters. For each performance parameter , based on its probability distribution Calculate the edge entropy, the formula is as follows:

[0052] in, Indicates The marginal entropy of performance parameters, represent Value The probability of . Calculate the joint entropy of multiple performance parameters. For two performance parameters and , the joint entropy is defined as:

[0053] in, represent Value and Value By calculating the marginal entropy and joint entropy of all performance parameters, the performance parameter entropy value set is constructed. Based on the performance parameter entropy value set, a mutual information calculation formula is constructed to measure the degree of information correlation between different performance parameters. The calculation formula is as follows:

[0054] Among them, mutual information Reflects performance parameters and The correlation between them is as follows: the larger the value, the higher the information overlap between them. Constructing the mutual information matrix :

[0055] In order to eliminate the influence of noise, the corresponding noise entropy value is subtracted from the mutual information matrix to construct the net information characteristics of the performance parameters. The calculation is as follows:

[0056] in, Represents the noise signal Probability distribution of performance parameters. Net information characteristics The calculation is as follows:

[0057] The relevance measure and redundancy are calculated based on the performance parameter net information characteristics in order to construct the feature evaluation function. The calculation formula is as follows:

[0058] The redundancy The calculation formula is as follows:

[0059] Constructing feature evaluation function :

[0060] Based on the feature evaluation function, the sequence forward selection algorithm is executed, and each round of iteration selects The feature with the largest value increment, and Greater than the first target value The features are marked as candidate features to form a feature candidate set. The feature candidate set is subjected to correlation analysis to select the features with an absolute value of correlation coefficient greater than the second target value. The key feature set of battery performance is obtained by combining the features of The calculation is as follows:

[0061] in, yes and The covariance of and They are and The standard deviation of Based on the key feature set of battery performance, an integrated feature measurement function of operating parameters and performance attenuation rate is constructed to comprehensively measure the impact of different features on battery attenuation. Integrated feature measurement function The calculation is as follows:

[0062] in, It is The weight of each feature. In order to identify the key attenuation features, the local extreme points of the function are analyzed. Through the density clustering method, the time span is set to , define the threshold of the number of points in the cluster , execute the density clustering algorithm, the clustering results are as follows:

[0063] Among them, each Represents a feature cluster. By extracting the cluster center point, a battery attenuation feature vector is formed. :

[0064] In a specific embodiment, the execution step constructs an integrated characteristic metric function of the operating condition parameters and the performance attenuation rate according to the key characteristic set of battery performance, and processes the local extreme points of the integrated characteristic metric function by density clustering to obtain the battery attenuation characteristic vector, which may specifically include the following steps: Extract operating parameters from the key battery performance feature set and calculate the performance decay rate, and set the balance factor to construct an integrated feature measurement function; Set a sliding time window for the integrated feature measurement function, and calculate the local mean and standard deviation within the time window to obtain local statistical features; Determine the local threshold condition according to the local statistical characteristics, mark the time points that meet the local threshold condition, and obtain a set of candidate feature points; The density clustering algorithm is executed on the candidate feature point set with the preset time span as the clustering radius, and clustering is performed based on the point density and distance threshold to obtain the feature point clustering result; Based on the characteristic point clustering results, the center point of each cluster is selected, and the timestamp, operating parameters and performance parameters corresponding to the center point are recorded to obtain a key characteristic point sequence. The key characteristic point sequence is reconstructed into a vector form in chronological order to obtain a battery attenuation characteristic vector.

[0065] Specifically, parameters related to operating conditions are selected from the key characteristics of battery performance, such as charge and discharge rate, battery temperature, SOC variation range, etc., and the battery performance decay rate is calculated at the same time. The decay rate is expressed as the capacity retention rate. and internal resistance growth rate Assume that at time The nominal capacity of the battery is , then its relative capacity is defined as:

[0066] in, It's time The actual measured battery capacity at . Similarly, the relative rate of change of the battery internal resistance is expressed as:

[0067] in, It's time The actual internal resistance at is the initial internal resistance. On this basis, the balance factor is set An integrated feature metric function is constructed to weigh the relationship between operating parameters and performance degradation rate. The integrated feature metric function is defined as:

[0068] in, is the weight coefficient of the corresponding feature, which is used to control the contribution of different features to the comprehensive metric and is adjusted through optimization methods. A sliding time window is set for the integrated feature metric function to analyze the local statistical characteristics of the battery attenuation feature. The window size is set to , calculate the local mean in each window and standard deviation To obtain the statistical characteristics within the window:

[0069]

[0070] in, Reflects the feature mean in the current window, Indicates the characteristic fluctuation within the time window. Determine the local threshold condition based on the local statistical characteristics to screen out potential key time points. The local threshold is set as:

[0071] in, is the threshold factor, which is set to 2 or 3 to ensure that the selected time point reflects the significant characteristics of battery degradation. or , then mark the time point as a candidate feature point and obtain a set of candidate feature points. Perform a density clustering algorithm on the set of candidate feature points to identify the attenuation mode under different working conditions. Density clustering divides high-density areas into clusters based on the density distribution of data points and filters out isolated points. Set the time span As the cluster radius and define the minimum number of cluster points To perform the density clustering algorithm. If a certain time point The number of candidate feature points around exceeds , then the point is considered as a core point, and the neighboring points around it are classified into the same cluster. The process is calculated using the following distance metric:

[0072] if ,but and Belong to the same cluster and get the feature point clustering result :

[0073] After obtaining the characteristic point clustering results, extract the center point of each cluster and record the timestamps, operating parameters and performance parameters corresponding to these center points. The center point is selected by using the point with the highest density in the cluster, or selecting the point that best represents the overall trend in the cluster, that is, calculating the mean center in each cluster:

[0074] The working condition parameters corresponding to this time point and performance parameters Record it to form a sequence of key feature points:

[0075] The key feature point sequence is reconstructed in time order to form the battery attenuation feature vector. Assume that the time sequence of the key feature point is , then the battery attenuation feature vector is expressed as:

[0076] This vector contains the time evolution information of battery degradation and integrates key performance parameters under different operating conditions.

[0077] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Calculate the time interval, working condition difference and performance difference of adjacent feature points in the battery attenuation feature vector, and perform weighted adaptive distance calculation to obtain the feature point distance matrix; A connection threshold is set for the feature point distance matrix, and feature point pairs whose distance is less than the connection threshold are marked as connection point pairs to obtain an initial connection relationship set; Construct an adjacency matrix A based on the initial connection relationship set. The matrix element Aij takes 1 when the feature points i and j are connected, otherwise it takes 0, and a binary adjacency matrix is ​​obtained. The degree centrality and betweenness centrality of each node in the binary adjacency matrix are calculated, and the degree centrality and betweenness centrality are normalized and weighted summed to obtain the node importance index; Based on the node importance index, Louvain community discovery algorithm is performed on the binary adjacency matrix, and the community is divided by modularity Q value optimization to obtain the network community structure. The topological characteristics of the network community structure are combined with the binary adjacency matrix to construct a battery attenuation feature association network.

[0078] Specifically, define the battery attenuation feature vector. Assume that the battery attenuation feature vector is Each feature point consists of Including time , working parameters and performance parameters , then the feature vector is expressed as:

[0079] In order to measure the adjacent feature points and The difference between the two is calculated separately. , Differences in working conditions and performance differences The time interval is calculated as follows:

[0080] Differences in working conditions The Euclidean distance is used for calculation:

[0081] Performance Differences The calculation method is similar, also using Euclidean distance:

[0082] In order to comprehensively measure the overall distance between feature points, a weighted adaptive distance calculation is performed. Define the weight parameter Corresponding to the time interval, working condition difference and performance difference respectively, the weighted adaptive distance The calculation formula is as follows:

[0083] in, The value of is optimized based on experimental data, so that different features have reasonable influence weights when calculating distance. Calculate the weighted adaptive distance between all feature point pairs to form a feature point distance matrix :

[0084] After obtaining the feature point distance matrix, set the connection threshold To filter out the feature point pairs with strong correlation. , then and Considered as a pair of connection points, forming an initial connection relationship set :

[0085] Based on the initial connection relationship set, build an adjacency matrix ,in exist and When connected, the value is 1, otherwise it is 0, forming a binary adjacency matrix:

[0086] in:

[0087] After constructing the binary adjacency matrix, calculate the degree centrality of each node and betweenness centrality . Degree centrality is defined as the number of neighbors a node is directly connected to:

[0088] Betweenness centrality measures the role of a node as a bridge in the shortest path and is defined as:

[0089] in, It is a slave node To Node The number of shortest paths, These paths pass through the nodes The number of and Normalize and calculate node importance index :

[0090] in, and To normalize the weights, adjust the contribution of degree centrality and betweenness centrality in calculating importance. Based on the node importance index, the Louvain community discovery algorithm is performed on the binary adjacency matrix, and the modularity is used to find the community. The value is optimized to divide the community. The modularity is defined as follows:

[0091] in, and The nodes are and nodes The degree, and represents the community to which the node belongs, when It takes 1 when , otherwise it takes 0. By maximizing The optimal community division result is obtained by averaging the values ​​and the network community structure is obtained. The topological characteristics of the network community structure are combined with the binary adjacency matrix to construct a battery attenuation feature association network, which effectively describes the association relationship between different attenuation modes.

[0092] In a specific embodiment, the process of executing step S4 may specifically include the following steps: Extract the global topological features of the battery attenuation feature association network and construct the network global structural feature vector. The global topological features include the average shortest path length, network diameter, network density and global clustering coefficient. The node-level topological features are calculated for the battery attenuation feature-related network, and the node-level topological features are concatenated with the network global structure feature vector to obtain the node feature matrix. The node-level topological features include node degree distribution, betweenness centrality, and feature vector centrality. Community-level topological features are extracted from the battery attenuation feature association network and combined with the node feature matrix to obtain a multi-scale network feature vector. The community-level topological features include the number of communities, community size distribution, and inter-community connection strength. Input the multi-scale network feature vector and the battery attenuation feature vector into the input layer of the radial basis kernel function support vector machine classifier to obtain standardized input features, and perform nonlinear mapping on the standardized input features through the radial basis kernel function to obtain kernel function mapping features; Based on the kernel function mapping features, a classification decision function is executed to classify the performance degradation mode of the energy storage battery, and a classification result of the performance degradation mode of the energy storage battery is obtained.

[0093] Specifically, the global topological features of the battery degradation feature association network are extracted. Assume that the battery degradation feature association network is ,in is a node set, representing the battery attenuation feature points, is an edge set, which represents the association relationship between different decay feature points. In order to describe the overall structure of the network, the global topological features are calculated, including the average shortest path length, network diameter, network density and global clustering coefficient. The calculation is as follows:

[0094] in, is the number of nodes in the network, Representation Node and The shortest path length between them. This indicator is used to measure the overall connectivity of the network. Network diameter The calculation is as follows:

[0095] It indicates the shortest path length between the two farthest nodes in the network, reflecting the expansion range of the network. Network density The calculation is as follows:

[0096] in, is the number of edges in the network, which is used to measure the closeness of the node connections in the network. Global clustering coefficient The calculation is as follows:

[0097] in, Is a node The local clustering coefficient, which indicates the connectivity between the neighbors of the node, is calculated as follows:

[0098] in, Is a node The actual number of connections between neighbors, Is a node After calculating the above global topological features, they are combined into the network global structure feature vector :

[0099] The node-level topological features of the battery degradation feature association network are calculated and concatenated with the global structure feature vector to form a node feature matrix. The node-level topological features include node degree distribution, betweenness centrality, and eigenvector centrality. The calculation is as follows:

[0100] in, is an element of the adjacency matrix. If the node and If there is a connection between ,otherwise Betweenness Centrality The calculation is as follows:

[0101] in, It is a slave node To Node The number of shortest paths, These paths pass through the nodes The number of eigenvector centrality The calculation is as follows:

[0102] in, is the eigenvector, is the adjacency matrix, is the maximum eigenvalue. This centrality is used to measure the influence of a node in the entire network. The above node-level topological features are combined to form a node feature matrix :

[0103] Community-level topological features are extracted and combined with node feature matrix to form multi-scale network feature vectors. Community-level topological features include the number of communities, community size distribution, and inter-community connection strength. It is Communities, number of communities Defined as:

[0104] Community size distribution The calculation is as follows:

[0105] Strength of connections between communities The calculation is as follows:

[0106] Combine community-level topological features to obtain multi-scale network feature vectors :

[0107] The multi-scale network feature vector and the battery degradation characteristic vector As input, it is input into the input layer of the radial basis kernel support vector machine. Standardization is performed to ensure that different features have the same scale. The standardization formula is as follows:

[0108] in, is the mean, is the standard deviation. Using the radial basis kernel function Perform nonlinear mapping:

[0109] in, is a hyperparameter used to control the smoothness of high-dimensional mapping. After obtaining the kernel function mapping features, the classification decision function is executed:

[0110] in, are the coefficients of the support vectors, is the category label, is a bias term. Based on the classification decision function, the performance degradation mode of the energy storage battery is classified and the final classification result is obtained.

[0111] The above describes the battery performance degradation test method of the energy storage battery in the embodiment of the present invention. The following describes the battery performance degradation test device of the energy storage battery in the embodiment of the present invention. Figure 2 , an embodiment of a battery performance attenuation testing device for an energy storage battery in an embodiment of the present invention comprises: A construction module is used to construct a battery performance measurement data matrix of the energy storage battery during the charging and discharging process, and calculate the optimal value of the noise variance through a zero-mean censored normal distribution model to obtain a battery performance noise characteristic matrix; A calculation module is used to calculate the mutual information of performance parameters based on the battery performance noise characteristic matrix, obtain the key characteristic set of battery performance, and construct the battery attenuation characteristic vector of operating condition parameters and performance attenuation rate; Creating a module for extracting weighted adaptive distances of time intervals, operating condition differences, and performance differences based on a battery attenuation feature vector, and creating a battery attenuation feature association network; The classification module is used to classify the performance attenuation mode based on the battery attenuation feature association network and the battery attenuation feature vector to obtain the classification result of the energy storage battery performance attenuation mode.

[0112] Through the collaborative cooperation of the above-mentioned components, the present invention models the measurement noise by adopting a zero-mean censored normal distribution model, and calculates the optimal value of the noise variance in combination with the maximum entropy principle, thereby achieving effective elimination of noise in the measurement data, and at the same time adopts Butterworth low-pass filtering and outlier processing to ensure data quality; through a feature dimensionality reduction method based on mutual information theory and a sequential forward selection algorithm, as well as an integrated feature measurement function of operating parameters and performance decay rate, the battery performance decay characteristics are comprehensively extracted; a weighted adaptive distance calculation and a community structure adjacency matrix construction method are introduced, combined with the Louvain community discovery algorithm, to reveal the intrinsic connection between performance decay characteristics; a multi-scale network feature extraction method is adopted to characterize the performance decay characteristics from three levels: global, node, and community, and the radial basis kernel function support vector machine classifier is used to achieve accurate classification of performance decay patterns.

[0113] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0114] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0115] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0116] 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.

[0117] 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 method for testing battery performance attenuation of an energy storage battery, characterized in that: The method comprises: Construct the battery performance measurement data matrix of the energy storage battery during the charging and discharging process, and calculate the optimal value of the noise variance through the zero-mean censored normal distribution model to obtain the battery performance noise characteristic matrix; Based on the battery performance noise characteristic matrix, the mutual information of performance parameters is calculated to obtain a key characteristic set of battery performance, and a battery attenuation characteristic vector of operating condition parameters and performance attenuation rate is constructed; Extracting the weighted adaptive distance of the time interval, the operating condition difference and the performance difference according to the battery attenuation feature vector, and creating a battery attenuation feature association network; Based on the battery attenuation feature association network and the battery attenuation feature vector, performance attenuation mode classification is performed to obtain a classification result of the energy storage battery performance attenuation mode.

2. The battery performance attenuation testing method of the energy storage battery according to claim 1, characterized in that: The battery performance measurement data matrix of the energy storage battery during the charging and discharging process is constructed, and the optimal value of the noise variance is calculated through a zero-mean censored normal distribution model to obtain a battery performance noise characteristic matrix, including: Collect data on the voltage, current, temperature, internal resistance and state of charge of the energy storage battery during the charging and discharging process to obtain the original data sequence; Performing segment processing on the original data sequence to obtain a time series data set, and performing signal processing on the time series data set to obtain a filtered data set; Identifying and correcting outliers in the filtered data set to obtain a corrected data set, and normalizing all performance parameters in the corrected data set to obtain a normalized data set; Reconstructing the matrix of the normalized data set by taking the sampling time points as row vectors and the performance parameters as column vectors to obtain a battery performance measurement data matrix; A zero-mean censored normal distribution model is constructed based on the battery performance measurement data matrix, and the optimal value of the noise variance is calculated by the maximum entropy principle to obtain the battery performance noise characteristic matrix.

3. The battery performance attenuation testing method of the energy storage battery according to claim 2, characterized in that: The method constructs a zero-mean censored normal distribution model based on the battery performance measurement data matrix, and calculates the optimal value of the noise variance by the maximum entropy principle to obtain a battery performance noise characteristic matrix, including: Dividing the battery performance measurement data matrix into blocks to obtain a plurality of data blocks, and performing centralization processing on each data block to obtain a zero-mean data block; Dividing the zero-mean data block into censored intervals to obtain censored intervals, and constructing a zero-mean censored normal distribution model based on the censored intervals; The maximum entropy principle is used to construct a noise variance optimization objective function and perform numerical optimization on the zero-mean censored normal distribution model to obtain the optimal value of the noise variance; Constructing a white noise channel model according to the zero-mean censored normal distribution model and the optimal value of the noise variance, and performing noise analysis on the battery performance measurement data matrix to obtain noise distribution characteristics; The noise entropy value of each performance parameter is calculated for the noise distribution characteristics, the noise entropy value is constructed into a noise entropy feature set, and the noise entropy feature set is associated with the corresponding performance parameters and matrix transformation operations are performed to obtain a battery performance noise feature matrix.

4. The battery performance attenuation testing method of the energy storage battery according to claim 1, characterized in that: The method of calculating the mutual information of performance parameters based on the battery performance noise characteristic matrix to obtain a key characteristic set of battery performance and constructing a battery attenuation characteristic vector of operating condition parameters and performance attenuation rate includes: Calculating the marginal entropy of the performance parameters in the battery performance noise characteristic matrix, and performing joint entropy construction on the marginal entropy to obtain a performance parameter entropy value set; Constructing a mutual information calculation formula based on the performance parameter entropy value set, and performing mutual information calculation on all performance parameter pairs to obtain a mutual information matrix; Subtracting the corresponding noise entropy value from the mutual information matrix, and constructing a performance parameter net information feature, calculating a correlation measure and a redundancy according to the performance parameter net information feature, and constructing a feature evaluation function; Execute a sequential forward selection algorithm on the feature evaluation function, select the feature with the largest F value increment in each iteration, and mark the features with F values ​​greater than the first target value as candidate features to obtain a feature candidate set; Performing correlation analysis on the feature candidate set, selecting a feature combination whose absolute value of the correlation coefficient is greater than a second target value, and obtaining a battery performance key feature set; An integrated feature metric function of operating parameters and performance attenuation rate is constructed according to the key feature set of battery performance, and local extreme points of the integrated feature metric function are processed by density clustering to obtain a battery attenuation feature vector.

5. The battery performance attenuation testing method of the energy storage battery according to claim 4, characterized in that: The method of constructing an integrated characteristic metric function of operating condition parameters and performance attenuation rate according to the key characteristic set of battery performance, and processing the local extreme points of the integrated characteristic metric function by density clustering to obtain a battery attenuation characteristic vector includes: Extracting operating parameters from the battery performance key feature set and calculating the performance decay rate, and setting a balance factor to construct an integrated feature measurement function; Setting a sliding time window for the integrated feature metric function, and calculating the local mean and standard deviation within the time window to obtain local statistical features; Determine a local threshold condition according to the local statistical feature, mark the time points that meet the local threshold condition, and obtain a set of candidate feature points; Execute a density clustering algorithm on the candidate feature point set with a preset time span as the clustering radius, perform cluster division based on point density and distance threshold, and obtain a feature point clustering result; Based on the characteristic point clustering result, the center point of each cluster is selected, and the timestamp, operating parameters and performance parameters corresponding to the center point are recorded to obtain a key characteristic point sequence, and the key characteristic point sequence is reconstructed into a vector form in chronological order to obtain a battery attenuation characteristic vector.

6. The battery performance attenuation testing method of the energy storage battery according to claim 1, characterized in that: The step of extracting the weighted adaptive distance of the time interval, the operating condition difference and the performance difference according to the battery attenuation feature vector and creating a battery attenuation feature association network includes: Calculating the time interval, operating condition difference and performance difference of adjacent characteristic points in the battery attenuation characteristic vector, and performing weighted adaptive distance calculation to obtain a characteristic point distance matrix; Setting a connection threshold for the feature point distance matrix, marking feature point pairs whose distance is less than the connection threshold as connection point pairs, and obtaining an initial connection relationship set; Based on the initial connection relationship set, an adjacency matrix A is constructed. The matrix element A ij When feature points i and j are connected, take 1, otherwise take 0, and get a binary adjacency matrix; Calculating the degree centrality and betweenness centrality of each node in the binary adjacency matrix, and normalizing and weighted summing the calculated degree centrality and the betweenness centrality to obtain a node importance index; Based on the node importance index, the Louvain community discovery algorithm is executed on the binary adjacency matrix, and the community is divided by optimizing the modularity Q value to obtain the network community structure; The topological features of the network community structure are combined with the binary adjacency matrix to construct a battery attenuation feature association network.

7. The battery performance attenuation testing method of the energy storage battery according to claim 1, characterized in that: The performance attenuation mode classification is performed based on the battery attenuation feature association network and the battery attenuation feature vector to obtain a classification result of the energy storage battery performance attenuation mode, including: Extracting the global topological features of the battery attenuation feature association network and constructing a global structural feature vector of the network, wherein the global topological features include an average shortest path length, a network diameter, a network density, and a global clustering coefficient; Calculating node-level topological features for the battery attenuation feature association network, and concatenating the node-level topological features with the network global structure feature vector to obtain a node feature matrix, wherein the node-level topological features include node degree distribution, betweenness centrality, and feature vector centrality; Extracting community-level topological features from the battery attenuation feature association network, and combining them with the node feature matrix to obtain a multi-scale network feature vector, wherein the community-level topological features include the number of communities, community size distribution, and connection strength between communities; Inputting the multi-scale network feature vector and the battery attenuation feature vector into the input layer of the radial basis kernel function support vector machine classifier to obtain standardized input features, and performing nonlinear mapping on the standardized input features through the radial basis kernel function to obtain kernel function mapping features; A classification decision function is executed based on the kernel function mapping feature to classify the performance attenuation mode of the energy storage battery to obtain a classification result of the performance attenuation mode of the energy storage battery.

8. A battery performance attenuation test device for an energy storage battery, characterized in that: The method for testing the battery performance attenuation of an energy storage battery according to any one of claims 1 to 7, wherein the battery performance attenuation testing device of the energy storage battery comprises: A construction module is used to construct a battery performance measurement data matrix of the energy storage battery during the charging and discharging process, and calculate the optimal value of the noise variance through a zero-mean censored normal distribution model to obtain a battery performance noise characteristic matrix; A calculation module, used to calculate the mutual information of performance parameters based on the battery performance noise characteristic matrix, obtain a key characteristic set of battery performance, and construct a battery attenuation characteristic vector of operating condition parameters and performance attenuation rate; Creating a module for extracting a weighted adaptive distance of a time interval, a working condition difference, and a performance difference according to the battery attenuation feature vector, and creating a battery attenuation feature association network; A classification module is used to classify the performance attenuation mode based on the battery attenuation feature association network and the battery attenuation feature vector to obtain a classification result of the energy storage battery performance attenuation mode.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor implements the battery performance attenuation test method of the energy storage battery described in any one of claims 1 to 7 when executing the computer program.

Images