Energy storage battery health state evaluation method based on neural network model

Abstract

The invention relates to the technical field of battery monitoring, in particular to an energy storage battery health state evaluation method based on a neural network model, and the method comprises the following steps: 1, obtaining the circulation characteristics of an energy storage battery; 2, performing regression classification on the cycle characteristics based on a support vector machine to generate an initial life interval; 3, collecting the original relaxation characteristics of the energy storage battery for a plurality of adjacent times, and obtaining a continuity factor based on the original relaxation characteristics and the historical relaxation characteristics; the sparse feature set comprises a pulse factor, a waveform factor, a peak factor, a wavelet energy ratio, a wavelet energy entropy and a root-mean-square root; 4, performing feature expansion on the sparse feature set to obtain three-dimensional information; and 5, inputting the key feature matrix and the initial life interval into a neural network model to generate accurate life information of the energy storage battery. According to the technical scheme provided by the invention, the approximate life interval of the energy storage battery is firstly obtained by adopting relatively simple cycle characteristics.

Description

Technical Field

[0001] The present application relates to the field of battery monitoring technology, and in particular to a method for evaluating the health status of energy storage batteries based on a neural network model. Background Art

[0002] Energy storage batteries play a vital role in renewable energy grid integration and grid peak regulation. Accurately predicting the lifespan of energy storage batteries is crucial. Existing lifespan prediction methods typically extract battery life-related features, such as cycle count, voltage / current data during charge and discharge, and internal resistance data, and analyze the correlations between these features to make predictions. However, these methods have limited accuracy and can only provide a rough estimate of lifespan. To improve accuracy, some solutions further utilize battery relaxation characteristics to extract more detailed information. However, this increase in feature dimensionality places higher demands on the prediction model: not only must the model process more complex information, but its output lifetime prediction labels (or the segmentation of fine features) must also be more precise. This leads to a significant expansion of the model's information processing dimensionality, which in turn leads to model structural redundancy and the risk of overfitting during training.

[0003] In summary, the current energy storage battery life prediction system is difficult to achieve accurate prediction of service life. Summary of the Invention

[0004] The content of this application is used to briefly introduce concepts that will be described in detail in the detailed description section below. The content of this application is not intended to identify key features or essential features of the technical solution for which protection is sought, nor is it intended to limit the scope of the technical solution for which protection is sought.

[0005] Some embodiments of the present application propose a method for evaluating the health status of energy storage batteries based on a neural network model to solve the technical problems mentioned in the above background technology section.

[0006] As a first aspect of the present application, some embodiments of the present application provide a method for evaluating the health status of an energy storage battery based on a neural network model, comprising the following steps: Step 1: Obtain the cycle characteristics of the energy storage battery and pre-process the cycle characteristics to generate initial life characteristics; the cycle characteristics include: number of cycles, time when the charging voltage reaches the peak, time when the discharge voltage reaches the peak, and time when the charging current reaches the peak; Step 2: Perform regression classification on the cycle characteristics based on support vector machine to generate the initial life span; Step 3: Collect several adjacent original relaxation features of the energy storage battery. Based on the continuity factor between the original relaxation features and the historical relaxation features, select the original relaxation features with the highest correlation with the historical relaxation features from the original relaxation features, and extract a sparse feature set from the Vt curve corresponding to the original relaxation features. The sparse feature set includes pulse factor, shape factor, peak factor, wavelet energy ratio, wavelet energy entropy, and RMS value; Step 4: Expand the sparse feature set to obtain three-dimensional information; Step 5: Input the key feature matrix and the initial life span into the neural network model to generate accurate life information of the energy storage battery.

[0007] In the technical solution provided by the present application, a relatively simple cycle feature is used to first obtain the approximate life range of the energy storage battery, and then a higher-dimensional key feature matrix is ​​used to further refine the approximate life range of the energy storage battery. Therefore, the accuracy of the life prediction can be effectively increased. In practice, this life prediction method can effectively increase the accuracy of life prediction, and in order to avoid the explosion of model dimensions, the approximate life range is used to reduce the dimension of the neural network model output. In addition, the original relaxation feature analysis is used, and only the most recent Vt curve needs to be input, which reduces the dimension of the neural network model input. When the input dimension and the output dimension decrease synchronously, the internal gradient explosion of the neural network model can be reduced while ensuring the prediction accuracy of the neural network model, and the overfitting phenomenon of the model can be reduced.

[0008] Furthermore, step 1 includes the following steps: Step 11: Obtain the model of the energy storage battery, and obtain the charging voltage peak, discharge voltage peak, and charging current peak of the energy storage battery according to the model of the energy storage battery; Step 12: Obtain historical charge and discharge data of the energy storage battery, and obtain the time when the charging voltage reaches the peak value, the time when the discharge voltage reaches the peak value, and the time when the charging current reaches the peak value during each charge and discharge according to the charging voltage peak value, the discharge voltage peak value, and the charging current peak value; Step 13: Extract the cycle number from the historical charge and discharge data; Step 14: The time when the charging voltage reaches a peak value, the time when the discharge voltage reaches a peak value, the time when the charging current reaches a peak value, and the number of cycles during each charge and discharge are taken as cycle characteristics.

[0009] In the technical solution provided in the present application, the predefined charging voltage peak, discharging voltage peak, and charging current peak of the energy storage battery are obtained according to the signal of the energy storage battery, so that the cycle characteristics of the energy storage battery can be accurately obtained, avoiding the inaccurate collection of cycle characteristics of the same model of energy storage battery due to the lack of standard charging voltage peak, discharge voltage peak, and charging current peak of the energy storage battery.

[0010] This solution uses cycle features and support vector machines to obtain initial lifespan intervals. While the accuracy of the lifespan collection is not critical, it is important to ensure that the initial lifespan intervals cover the actual battery lifespan. However, support vector machines essentially classify the collected cycle features within a feature space. If the cycle features contain noise, the initial lifespan intervals obtained can be inaccurate.

[0011] Furthermore, step 14 includes the following steps: Step 141: Arrange the time when the charging voltage reaches the peak value in each cycle in sequence to obtain a charging voltage peak sequence; The time when the discharge voltage reaches the peak value in each cycle is arranged in sequence to obtain a discharge voltage peak sequence; The time when the charging current reaches the peak value in each cycle is arranged in sequence to obtain the charging current peak sequence; Step 142: performing noise reduction processing on the charging voltage peak sequence, the discharging voltage peak sequence, and the charging current peak sequence in sequence; Step 143: The noise-reduced charging voltage peak sequence, discharging voltage peak sequence, charging current peak sequence, and cycle number are used as cycle features.

[0012] This solution pre-derives standardized peak charging voltage, peak discharge voltage, and peak charging current values ​​for each energy storage battery model, using these values ​​as a benchmark to calculate the time it takes for the voltage and current to reach their peak values ​​during each charge and discharge cycle. This approach ensures consistent and accurate extraction of cycle characteristics (particularly the time to peak value) for batteries of the same model, effectively eliminating data bias caused by the lack of a unified standard reference value. This provides a more reliable and accurate initial feature input foundation for subsequent preliminary estimation and refined prediction of the lifespan range.

[0013] In the preprocessing of energy storage battery charge and discharge data (especially peak time series), traditional noise reduction methods often rely on fixed thresholds or global statistics, making them incapable of adapting to local fluctuations in the data. This results in inaccurate identification of outliers: either they are overly sensitive and mistakenly delete valid data points, or they are too insensitive and leave significant noise points. Furthermore, static processing methods are prone to the accumulation of outliers, further distorting the data distribution and affecting the input quality of subsequent life prediction models.

[0014] Furthermore, the noise reduction method in step 142 is as follows: Step 1421: pre-set the window length; Step 1422: For each data point x in the window, calculate whether x∈(u-2σ, u+2σ), where u represents the mean of the window and σ represents the standard deviation of the window. If x satisfies x∈(u-2σ,u+2σ), then the data point x is normal data. (u-2σ, u+2σ), then the data point x is noise data; Step 1423: Slide the window sequentially to remove all noise data to complete the noise reduction processing of all peak series.

[0015] This solution uses a dynamic threshold denoising method (steps 1421-1423) based on a sliding window and local statistics (mean u, standard deviation σ) to adaptively follow data trends. Using the ±2σ criterion (covering approximately 95% of normally distributed data), it accurately identifies and removes significantly deviating noise points (outliers) within the local window, effectively retaining data points that reflect the battery's true state. Furthermore, the sliding window mechanism prevents the accumulation of outliers within a fixed region, ensuring the smoothness and reliability of the peak time series after denoising, providing a cleaner and more accurate data foundation for subsequent feature extraction and lifespan prediction.

[0016] Step 2 includes the following steps: Step 21: Pre-acquire several initial life span labels, where the initial life spans are continuous and do not overlap; for each initial life span label, collect several corresponding cycle features, and use the collected cycle features and the corresponding life span label as one sample; collect N samples to generate a training data set S; ;x i represents the cyclic feature of the i-th sample, i represents the index of the sample, y i represents the life span label of the i-th sample, extract all labels to obtain the label set Kt; ;The elements in the tag set are different life span tags; Step 22: Construct the kernel function ; ; j represents the index of the sample, x j represents the cyclic feature of the j-th sample, Represents the kernel function bandwidth parameter; Step 23: Train a binary SVM classifier for each category, define the original optimization problem of SVM classification for each SVM classifier and set the optimization conditions; The original optimization problem is: ; w represents the weight vector, b represents the offset, and C represents the penalty factor. represents the slack variable; The constraints are: ; Represents a high-dimensional mapping; Step 24: Use the Lagrange multiplier method to transform the original problem into the dual problem; ; The constraints are: ; represents the Lagrange multiplier; Step 25: Construct a decision function based on the solution result; ; m represents the index of the SVM classifier; represents the temporary binary label of the m-th classifier; represents the Lagrange multiplier of the m-th classifier; Represents the offset of the mth classifier; Represents the classification result of sample x by the mth classifier; Step 26: Input the collected cycle features into the trained SVM classifier to obtain the classification results of all SVM classifiers, and take the label with the largest decision value as the output.

[0017] This approach uses a support vector machine (SVM) instead of a neural network to classify and predict the initial life span (steps 21-26). The SVM model offers greater transparency and interpretability, clearly revealing the weight of influence of different cycle features (such as peak times and number of cycles) on the classification decision boundary. This interpretability not only facilitates a deeper understanding of the relationship between features and life span but also guides feature optimization and model parameter adjustment, effectively improving the accuracy of the initial life span classification (i.e., the overall classification effect). Compared to black-box neural network models, SVM significantly enhances the analyzability of the prediction process and the credibility of the results while maintaining classification accuracy.

[0018] When performing refined lifespan prediction based on battery discharge curves (Vt curves), the battery state is not only affected by lifespan decay but also by other transient factors (such as environmental fluctuations, operating condition changes, and measurement noise). Consequently, the relaxation features (such as VC, VD, ReVD, tVD, and tReVD) directly extracted from the current or adjacent charge / discharge cycles, along with their corresponding Vt curves, contain significant noise and non-lifespan-related fluctuations. This noise can severely contaminate the feature extraction process, causing subsequent clustering and analysis results based on these features to deviate from the true battery aging state, making it difficult to accurately characterize the remaining lifespan of the current battery.

[0019] Furthermore, step 3 includes the following steps: Step 31: Collecting several adjacent original relaxation characteristics of the energy storage battery, the original relaxation characteristics include VC, VD, ReVD, tVD, and tReVD; wherein VC represents the voltage at the start of charge relaxation; VD represents the voltage drop amplitude at the initial relaxation stage; ReVD represents the voltage value at the end of the relaxation process; tVD represents the time required for the voltage to drop to a preset amplitude; and tReVD represents the time required for the voltage to drop close to ReVD; Step 32: Performing continuity analysis on the collected original relaxation characteristics and the historical relaxation characteristics of the energy storage battery, and selecting the original relaxation characteristics with the highest continuity with the historical relaxation characteristics from a number of original relaxation characteristics; Step 33: Extract a sparse feature set from the Vt curve corresponding to the original relaxation feature.

[0020] This solution effectively addresses the noise interference issue by introducing a relaxation feature screening mechanism based on continuity analysis (steps 31-33). Specifically, a continuity analysis is performed on several currently acquired raw relaxation features and historical relaxation features to screen out the relaxation features with the highest continuity with the historical attenuation trajectory and their corresponding Vt curves. This method prioritizes data segments that best reflect the inherent aging trends of the battery and are least susceptible to transient interference. Sparse feature sets (such as pulse factors and waveform factors) are extracted based on the filtered, highly representative Vt curves, significantly improving the correlation and reliability of the extracted features with the actual remaining life of the battery. This provides more accurate and pure input data for the subsequent refined life prediction model, enabling a more accurate assessment of the remaining life of the current battery.

[0021] Furthermore, step 32 includes the following steps: Step 321: obtaining the initial life span of the energy storage battery calculated when collecting the original relaxation characteristics; Obtain the initial lifetime interval corresponding to the historical relaxation characteristics; Step 322: Arrange the VCs in the historical relaxation features and the original relaxation features in chronological order to generate a VC sequence; Arrange the historical relaxation features VD in chronological order to generate a VD sequence; Arrange the historical relaxation features ReVD in chronological order to generate a ReVD sequence; Arrange the historical relaxation features tVD in chronological order to generate a tVD sequence; Arrange the historical relaxation features tReVD in chronological order to generate a tReVD sequence; Arrange the initial life span intervals corresponding to the historical relaxation characteristics in chronological order to generate a life span sequence; Step 323: Add the VC corresponding to the original relaxation characteristics of several adjacent times in the VC sequence, add the initial life interval corresponding to the original relaxation characteristics of the time to the life sequence, and calculate the continuity coefficient of the VC sequence and the life sequence to obtain ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the VC in the kth original relaxation feature to the VC sequence and calculating the lifetime sequence; In the VD sequence, the VD corresponding to the original relaxation characteristics of several adjacent times is added in sequence, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the VD sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the VD in the kth original relaxation feature to the VD sequence and calculating the lifetime sequence; In the ReVD sequence, the ReVD corresponding to the original relaxation characteristics of several adjacent times is added in sequence, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the ReVD sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the ReVD in the kth original relaxation feature to the ReVD sequence and calculating the lifetime sequence; In the tVD sequence, the tVD corresponding to the original relaxation characteristics of several adjacent times is sequentially added, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the tVD sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the tVD in the kth original relaxation feature to the tVD sequence and calculating the lifetime sequence; In the tReVD sequence, the tReVD corresponding to the original relaxation characteristics of several adjacent times is sequentially added, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the VC sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the tReVD in the k-th original relaxation feature to the tReVD sequence and calculating the lifetime sequence; Step 324: Calculate the sum of the continuity coefficients R for each original relaxation feature k k ; ; Step 325: The sum of the continuity coefficients R k The highest original relaxation feature k is taken as the original relaxation feature with the highest continuity with the historical relaxation features.

[0022] This solution innovatively proposes an automated screening mechanism based on multi-parameter continuity coefficient calculation (steps 321-325). Newly acquired raw relaxation features (by type, such as VC, VD) are sequentially added to the corresponding historical feature sequence and simultaneously added to their corresponding initial lifetime intervals to form a lifetime sequence. The continuity coefficient between each newly acquired feature sequence and the lifetime sequence is calculated (quantifying the synchronization of the feature's changing trend with the lifetime decay trend). Finally, for each candidate raw relaxation feature (corresponding to a set of k acquisitions), the sum Rk of the continuity coefficients of all its relaxation feature types (VC, VD, ReVD, tVD, tReVD) is calculated, and the feature k with the highest Rk is selected. This method uses continuity coefficients (such as correlation metrics) to accurately measure the match between new features and historical decay trajectories, eliminating subjectivity. By accumulating the continuity coefficients of five key relaxation features (VC, VD, ReVD, tVD, tReVD), the consistency of the new data segment with the lifetime decay trend across multiple dimensions is comprehensively assessed, resulting in more representative screening results. Based on the criterion of maximizing Rk, the original relaxation features and their Vt curves that have the best continuity with the historical aging process and most accurately reflect the current battery's remaining life (i.e., the "most representative") are automatically and efficiently identified. This mechanism ensures that the Vt curves subsequently used to extract sparse feature sets (such as pulse factor and waveform factor) are more representative and reliable, providing a better input data foundation for refined life prediction models and significantly improving the accuracy of the final life prediction.

[0023] Furthermore, the number of collected historical relaxation features is c, and the number of collected original relaxation features is d; 5d<c<10d.

[0024] In the technical solution provided in the present application, by controlling the number of historical relaxation features and the number of original relaxation features, an excessive number of historical relaxation features is avoided, and the changes newly added to the original relaxation features are not obvious, and thus the original relaxation features with low continuity cannot be accurately discovered.

[0025] Furthermore, the continuity coefficient The calculation formula is as follows: ; Among them, X i Represents the value of the i-th element in the VC sequence, represents the average value of the VC sequence; represents the mean value of the life series, Represents the value of the i-th element of the lifespan sequence; Continuity coefficient 、 、 、 、 is calculated in the same way.

[0026] In the technical solution provided in this application, the continuity coefficient is the correlation between two series. It can be foreseen that if the correlation between the two series is high, it means that the inertia of the data is relatively strong and the continuity is high, and the corresponding original relaxation characteristics inserted into the series are highly credible.

[0027] Furthermore, step 4 includes the following steps: Step 41: Obtain a sparse feature set, align each feature in the sparse feature set with each other according to time, and expand the sparse feature set into a two-dimensional matrix, where the two-dimensional matrix represents the type of the sparse set of behavioral features, and the columns of the two-dimensional matrix represent time; Step 42: Adding data values ​​of the sparse feature set to the two-dimensional matrix to form a third dimension, thereby converting the two-dimensional matrix into three-dimensional information; Step 43: Perform data enhancement on the three-dimensional information to obtain a key feature matrix.

[0028] In the technical solution provided in the present application, isolated sequences are converted into corresponding elements by aligning sparse feature sets in time.

[0029] Furthermore, the neural network model includes: an input layer for receiving the key feature matrix R;

[0030] Convolutional layer, used to extract the key feature matrix R for convolution operation to generate feature maps; BN layer, used to standardize the feature map; Activation layer, used to activate the standardized features; Pooling layer, which performs maximum pooling on the output of the activation layer; Connection layer, connects the output of the pooling layer;

[0031] The output layer outputs the precise lifespan information corresponding to the key features.

[0032] In the technical solution provided in this application, the constructed neural network model can capture the key information in the key feature matrix R through convolution, thereby accurately identifying the correspondence between the key feature matrix R and the feature map.

[0033] The technical solutions of the embodiments of the present application have at least the following advantages and beneficial effects: Accuracy Improvement: First, a preliminary lifetime interval estimate is made using basic cycle characteristics (steps 1-2). Then, a key feature matrix formed by introducing high-dimensional relaxation characteristics (steps 3-4) is used to refine this interval (step 5). This two-stage "coarse screening + fine tuning" strategy effectively integrates information of different granularities, overcomes the limitations of insufficient prediction accuracy based on a single feature dimension, and significantly improves the accuracy of the final lifetime prediction.

[0034] Model complexity and overfitting control: Output dimension compression: The neural network model receives preliminary life span intervals instead of the original, large and fine feature segmentation labels, which greatly reduces the complexity of the model output layer.

[0035] Input dimension compression: Through relaxation feature screening based on continuity factors (step 3) and data enhancement (step 4), only the key feature matrix extracted from the recent, highly correlated Vt curve needs to be input, rather than the lengthy historical relaxation data sequence, which effectively reduces the complexity of the model input layer.

[0036] Enhanced model stability: The simultaneous reduction of input and output dimensions significantly alleviates the information processing pressure within the neural network model. This not only reduces model structural redundancy but also effectively mitigates the risk of gradient instability (exploding / vanishing gradients) and overfitting during training caused by high-dimensional features and complex output spaces, thereby improving the model's generalization and training stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The flowchart of the energy storage battery health status assessment method based on the neural network model.

[0038] Figure 2 It is a time curve diagram of the charging voltage reaching the peak value.

[0039] Figure 3 It is a time curve diagram for the discharge voltage to reach the peak value.

[0040] Figure 4 A graph showing the number of cycles for denoising the spike sequence.

[0041] Figure 5 It is the frequency diagram of the pulse factor, the horizontal axis is the frequency, and the vertical axis is the pulse factor value.

[0042] Figure 6 This is a frequency diagram of the waveform factor, with the horizontal axis being the frequency and the vertical axis being the waveform factor value.

[0043] Figure 7 This is the frequency diagram of the peak factor, the horizontal axis is the frequency, and the vertical axis is the peak factor value.

[0044] Figure 8 It is the frequency diagram of wavelet energy ratio, the horizontal axis is frequency, and the vertical axis is the wavelet energy ratio.

[0045] Figure 9 It is the frequency diagram of wavelet energy entropy, the horizontal axis is the frequency, and the vertical axis is the wavelet energy entropy value. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific implementation methods. The same figure marks in the accompanying drawings represent the same components. It should be noted that the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the described embodiments of this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0047] Compared to the embodiments shown in the drawings, feasible embodiments within the scope of protection of the present application may have fewer components, other components not shown in the drawings, different components, differently arranged components, or differently connected components, etc. In addition, two or more components in the drawings may be implemented in a single component, or a single component shown in the drawings may be implemented as multiple separate components.

[0048] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning understood by persons of ordinary skill in the field to which this application belongs. The words "first", "second" and similar terms used in the specification and claims of this application do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "a" or "an" do not necessarily indicate a quantitative limitation. "Up", "down" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0049] refer to Figure 1 Embodiment 1: A method for evaluating the health status of an energy storage battery based on a neural network model, characterized by comprising: Step 1: Obtain the cycle characteristics of the energy storage battery and pre-process the cycle characteristics to generate initial life characteristics; the cycle characteristics include: number of cycles, time when the charging voltage reaches the peak, time when the discharge voltage reaches the peak, and time when the charging current reaches the peak.

[0050] Step 1 includes the following steps: Step 11: Obtain the model of the energy storage battery, and obtain the charging voltage peak, discharge voltage peak, and charging current peak of the energy storage battery according to the model of the energy storage battery.

[0051] The charging voltage peak, discharge voltage peak, and charging current peak values ​​are different for each battery model, so they need to be pre-set according to the battery model.

[0052] Step 12: Obtain historical charge and discharge data of the energy storage battery, and obtain the time when the charging voltage reaches the peak, the time when the discharge voltage reaches the peak, and the time when the charging current reaches the peak during each charge and discharge according to the charging voltage peak, the discharge voltage peak, and the charging current peak.

[0053] like Figure 2 As shown, Figure 2 The time when the charging voltage reaches the peak value (F1) is shown. Figure 3 The time when the discharge voltage reaches the peak value is shown. Figure 3 The peak value in is the first defined voltage minimum).

[0054] The battery's historical charge and discharge data is the current and voltage information during the battery's discharge and charge processes. Obtaining historical charge and discharge data can reveal peak charge voltage, peak discharge voltage, and peak charge current.

[0055] Step 13: Extract the cycle number from the historical charge and discharge data; Step 14: The time when the charging voltage reaches a peak value, the time when the discharge voltage reaches a peak value, the time when the charging current reaches a peak value, and the number of cycles during each charge and discharge are taken as cycle characteristics.

[0056] Step 14 includes the following steps: Step 141: Arrange the time when the charging voltage reaches the peak value in each cycle in sequence to obtain a charging voltage peak sequence; The time when the discharge voltage reaches the peak value in each cycle is arranged in sequence to obtain a discharge voltage peak sequence; The time when the charging current reaches the peak value in each cycle is arranged in sequence to obtain the charging current peak sequence; Step 142: performing noise reduction processing on the charging voltage peak sequence, the discharging voltage peak sequence, and the charging current peak sequence in sequence.

[0057] The noise reduction method in step 142 is as follows: Step 1421: pre-set the window length; Step 1422: For each data point x in the window, calculate whether x∈(u-2σ, u+2σ), where u represents the mean of the window and σ represents the standard deviation of the window. If x satisfies x∈(u-2σ,u+2σ), then the data point x is normal data. (u-2σ, u+2σ), then the data point x is noise data; Step 1423: Slide the window sequentially to remove all noise data to complete the noise reduction processing of all peak series.

[0058] Step 143: The noise-reduced charging voltage peak sequence, discharging voltage peak sequence, charging current peak sequence, and cycle number are used as cycle features.

[0059] like Figure 4 As shown, Figure 4 After preprocessing, some outliers can be removed.

[0060] Step 2: Perform regression classification on the cycle features based on support vector machine to generate the initial life span.

[0061] The initial life span is a larger pre-set life span. For example, the precise life span information is set to 0-1000 weeks. The precise life span information needs to be accurate to the week, with an error of no more than two weeks. To this end, the span of the initial life span can be set to 50, so that 0-1000 is divided into 20 initial life span intervals. Each initial life span interval is 20 weeks long.

[0062] Step 2 includes the following steps: Step 21: Pre-acquire several initial life span labels, where the initial life spans are continuous and do not overlap; for each initial life span label, collect several corresponding cycle features, and use the collected cycle features and the corresponding life span label as one sample; collect N samples to generate a training data set S; ;x i represents the cyclic feature of the i-th sample, i represents the index of the sample, y i represents the life span label of the i-th sample, extract all labels to obtain the label set Kt; ;The elements in the tag set are different life span tags; Step 22: Construct the kernel function ; ; j represents the index of the sample, x jrepresents the cyclic feature of the j-th sample, Represents the kernel function bandwidth parameter; Step 23: Train a binary SVM classifier for each category, define the original optimization problem of SVM classification for each SVM classifier and set the optimization conditions; The original optimization problem is: ; w represents the weight vector, b represents the offset, and C represents the penalty factor. represents the slack variable; The constraints are: ; Represents a high-dimensional mapping; Step 24: Use the Lagrange multiplier method to transform the original problem into the dual problem; ; The constraints are: ; represents the Lagrange multiplier; Step 25: Construct a decision function based on the solution result; ; m represents the index of the SVM classifier; represents the temporary binary label of the m-th classifier; represents the Lagrange multiplier of the m-th classifier; Represents the offset of the mth classifier; Represents the classification result of sample x by the mth classifier; Step 26: Input the collected cycle features into the trained SVM classifier to obtain the classification results of all SVM classifiers, and take the label with the largest decision value as the output.

[0063] In this way, by using the solution provided in step 2 and classifying the SVM classifier, a relatively accurate initial life span interval can be obtained.

[0064] Step 3: Collect the original relaxation features of the energy storage battery for several adjacent times. Based on the continuity factor of the original relaxation features and the historical relaxation features, select the original relaxation features with the highest correlation with the historical relaxation features from the several original relaxation features, and extract the sparse feature set from the Vt curve corresponding to the original relaxation features; the sparse feature set includes pulse factor, waveform factor, peak factor, wavelet energy ratio, wavelet energy entropy, and root mean square value.

[0065] "Relaxation characteristics" refers to the process in which the terminal voltage (or current) of an energy storage battery slowly returns to an equilibrium state (open circuit voltage) after a rapid charge and discharge pulse. This process reflects physical and chemical phenomena such as charge redistribution, ion diffusion, and interface reaction within the device. During the operation of the energy storage battery, multiple relaxation characteristics can be collected, but these relaxation characteristics may not accurately reflect the current state of the energy storage battery. For this reason, the essence of step 3 is to further filter out single-time life information that can reflect the current state of the energy storage battery. Specifically: Step 3 includes the following steps: Step 31: Collecting several adjacent original relaxation characteristics of the energy storage battery, the original relaxation characteristics include VC, VD, ReVD, tVD, and tReVD; wherein VC represents the voltage at the start of charge relaxation; VD represents the voltage drop amplitude at the initial relaxation stage; ReVD represents the voltage value at the end of the relaxation process; tVD represents the time required for the voltage to drop to a preset amplitude; and tReVD represents the time required for the voltage to drop close to ReVD; Step 32: Performing continuity analysis on the collected original relaxation characteristics and the historical relaxation characteristics of the energy storage battery, and selecting the original relaxation characteristics with the highest continuity with the historical relaxation characteristics from a number of original relaxation characteristics; The number of collected historical relaxation features is c, and the number of collected original relaxation features is d; 5d<c<10d.

[0066] Step 32 includes the following steps: Step 321: obtaining the initial life span of the energy storage battery calculated when collecting the original relaxation characteristics; Obtain the initial lifetime interval corresponding to the historical relaxation characteristics; Step 322: Arrange the VCs in the historical relaxation features and the original relaxation features in chronological order to generate a VC sequence; Arrange the historical relaxation features VD in chronological order to generate a VD sequence; Arrange the historical relaxation features ReVD in chronological order to generate a ReVD sequence; Arrange the historical relaxation features tVD in chronological order to generate a tVD sequence; Arrange the historical relaxation features tReVD in chronological order to generate a tReVD sequence; Arrange the initial life span intervals corresponding to the historical relaxation characteristics in chronological order to generate a life span sequence; Step 323: Add the VC corresponding to the original relaxation characteristics of several adjacent times in the VC sequence, add the initial life interval corresponding to the original relaxation characteristics of the time to the life sequence, and calculate the continuity coefficient of the VC sequence and the life sequence to obtain ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the VC in the kth original relaxation feature to the VC sequence and calculating the lifetime sequence; In the VD sequence, the VD corresponding to the original relaxation characteristics of several adjacent times is added in sequence, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the VD sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the VD in the kth original relaxation feature to the VD sequence and calculating the lifetime sequence; In the ReVD sequence, the ReVD corresponding to the original relaxation characteristics of several adjacent times is added in sequence, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the ReVD sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the ReVD in the kth original relaxation feature to the ReVD sequence and calculating the lifetime sequence; In the tVD sequence, the tVD corresponding to the original relaxation characteristics of several adjacent times is sequentially added, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the tVD sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the tVD in the kth original relaxation feature to the tVD sequence and calculating the lifetime sequence; In the tReVD sequence, the tReVD corresponding to the original relaxation characteristics of several adjacent times is sequentially added, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the VC sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the tReVD in the k-th original relaxation feature to the tReVD sequence and calculating the lifetime sequence.

[0067] For example, in the historical relaxation feature, the VD sequence is: 1, 2, 3, 4. The lifetime sequence (number of lifetime intervals) is 4, 5, 6, 7. The VD that generated the current original relaxation feature is 5, and the corresponding lifetime interval calculated in step 2 is 6.

[0068] Then put the VD sequence: 1, 2, 3, 4, 5 and the life sequence 4, 5, 6, 7, 6 together for continuity calculation. The calculation method is as follows: Continuity coefficient The calculation formula is as follows: ; Among them, X i Represents the value of the i-th element in the VC sequence, represents the average value of the VC sequence; represents the mean value of the life series, Represents the value of the i-th element of the lifespan sequence; Continuity coefficient 、 、 、 、 is calculated in the same way.

[0069] Step 324: Calculate the sum of the continuity coefficients R for each original relaxation feature k k ; ; Step 325: The sum of the continuity coefficients R k The highest original relaxation feature k is taken as the original relaxation feature with the highest continuity with the historical relaxation features.

[0070] In combination with the above content, each of the several most recently generated original relaxation features can be added to the historical relaxation features for continuity analysis, so as to select the closest original relaxation feature.

[0071] Step 33: Extract a sparse feature set from the Vt curve corresponding to the original relaxation feature. The sparse feature set includes pulse factor, shape factor, peak factor, wavelet energy ratio, wavelet energy entropy, and root mean square value.

[0072] This Vt curve is a graph of the relationship between current and voltage during the relaxation process. From this graph, wavelet transform, fuzzy entropy calculation and other methods can be used to obtain the elements in the required sparse feature set. The pulse factor, waveform factor, peak factor, wavelet energy ratio, wavelet energy entropy, and root mean square value listed in this scheme are several important features extracted from the Vt curve. In practice, the energy entropy and fuzzy entropy can be further increased, and some features can also be reduced. The key to this application is the need to find an accurate Vt curve.

[0073] like Figures 5 to 9 As shown, we can see the feature maps of each element in the sparse feature set.

[0074] Step 4: Expand the sparse feature set to obtain three-dimensional information;

[0075] Step 4 includes the following steps: Step 41: Obtain a sparse feature set, align each feature in the sparse feature set with each other according to time, and expand the sparse feature set into a two-dimensional matrix, where the two-dimensional matrix represents the type of the sparse set of behavioral features, and the columns of the two-dimensional matrix represent time; Step 42: Add the data values ​​of the sparse feature set to the two-dimensional matrix to form a third dimension, so as to convert the two-dimensional matrix into three-dimensional information.

[0076] Step 4 actually involves expanding multiple curves onto a two-dimensional plane to form a two-dimensional image information matrix with channel information.

[0077] Step 5: Input the key feature matrix and the initial life span into the neural network model to generate accurate life information of the energy storage battery.

[0078] After processing in steps 3 and 4, the three-dimensional information most closely related to the battery's precise lifespan is extracted. Using this three-dimensional information as input and precise lifespan information as output, sufficient data samples are created. This data sample is then used to train a neural network model, which can then be used to predict precise lifespan information.

[0079] The neural network model can be selected arbitrarily based on computing resources. The neural network models provided in this application are as follows: Furthermore, the neural network model includes: The input layer is used to receive the key feature matrix R; Convolutional layer, used to extract the key feature matrix R for convolution operation to generate feature maps; BN layer, used to standardize the feature map; Activation layer, used to activate the standardized features; Pooling layer, which performs maximum pooling on the output of the activation layer; Connection layer, connects the output of the pooling layer; The output layer outputs the precise lifespan information corresponding to the key features.

[0080] The neural network model in this solution mainly performs feature standardization, and the standardization process mainly standardizes the data. The standardization formula is: ; Q represents the feature map of the i-th data sample, is the mean of each channel, is the variance of each channel.

[0081] In practice, other neural network model structures can also be selected for training to obtain accurate and precise life information.

[0082] The above are merely preferred embodiments of the present application and are not intended to limit the present application. Those skilled in the art will readily appreciate that various modifications and variations are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A method for evaluating the health status of energy storage batteries based on a neural network model, characterized in that: include: Step 1: Obtain the cycle characteristics of the energy storage battery and preprocess the cycle characteristics to generate initial life characteristics; Cycle characteristics include: number of cycles, time when charging voltage reaches peak value, time when discharge voltage reaches peak value, time when charging current reaches peak value; Step 2: Perform regression classification on the cycle characteristics based on support vector machine to generate the initial life span; Step 3: Collect several adjacent original relaxation features of the energy storage battery. Based on the continuity factor between the original relaxation features and the historical relaxation features, select the original relaxation features with the highest correlation with the historical relaxation features from the original relaxation features, and extract a sparse feature set from the Vt curve corresponding to the original relaxation features. The sparse feature set includes pulse factor, shape factor, peak factor, wavelet energy ratio, wavelet energy entropy, and RMS value; Step 4: Expand the sparse feature set to obtain three-dimensional information; Step 5: Input the key feature matrix and the initial life span into the neural network model to generate accurate life information of the energy storage battery.

2. The method for evaluating the health status of energy storage batteries based on a neural network model according to claim 1, characterized in that: Step 1 includes the following steps: Step 11: Obtain the model of the energy storage battery, and obtain the charging voltage peak, discharge voltage peak, and charging current peak of the energy storage battery according to the model of the energy storage battery; Step 12: Obtain historical charge and discharge data of the energy storage battery, and obtain the time when the charging voltage reaches the peak value, the time when the discharge voltage reaches the peak value, and the time when the charging current reaches the peak value during each charge and discharge according to the charging voltage peak value, the discharge voltage peak value, and the charging current peak value; Step 13: Extract the cycle number from the historical charge and discharge data; Step 14: The time when the charging voltage reaches a peak value, the time when the discharge voltage reaches a peak value, the time when the charging current reaches a peak value, and the number of cycles during each charge and discharge are taken as cycle characteristics.

3. The method for evaluating the health status of energy storage batteries based on a neural network model according to claim 1, characterized in that: Step 14 includes the following steps: Step 141: Arrange the time when the charging voltage reaches the peak value in each cycle in sequence to obtain a charging voltage peak sequence; The time when the discharge voltage reaches the peak value in each cycle is arranged in sequence to obtain a discharge voltage peak sequence; The time when the charging current reaches the peak value in each cycle is arranged in sequence to obtain the charging current peak sequence; Step 142: performing noise reduction processing on the charging voltage peak sequence, the discharging voltage peak sequence, and the charging current peak sequence in sequence; Step 143: The noise-reduced charging voltage peak sequence, discharging voltage peak sequence, charging current peak sequence, and cycle number are used as cycle features.

4. The method for evaluating the health status of energy storage batteries based on a neural network model according to claim 1, wherein: The noise reduction method in step 142 is as follows: Step 1421: pre-set the window length; Step 1422: For each data point x in the window, calculate whether x∈(u-2σ, u+2σ), where u represents the mean of the window and σ represents the standard deviation of the window. If x satisfies x∈(u-2σ,u+2σ), then the data point x is normal data. (u-2σ, u+2σ), then the data point x is noise data; Step 1423: Slide the window sequentially to remove all noise data to complete the noise reduction processing of all peak series.

5. The method for evaluating the health status of energy storage batteries based on a neural network model according to claim 3, characterized in that: Step 2 includes the following steps: Step 21: Pre-acquire several initial life span labels, where the initial life spans are continuous and do not overlap; for each initial life span label, collect several corresponding cycle features, and use the collected cycle features and the corresponding life span label as one sample; collect N samples to generate a training data set S; ;x i represents the cyclic feature of the i-th sample, i represents the index of the sample, y i represents the life span label of the i-th sample, extract all labels to obtain the label set Kt; ;The elements in the tag set are different life span tags; Step 22: Construct the kernel function ; ; j represents the index of the sample, x j represents the cyclic feature of the j-th sample, Represents the kernel function bandwidth parameter; Step 23: Train a binary SVM classifier for each category, define the original optimization problem of SVM classification for each SVM classifier and set the optimization conditions; The original optimization problem is: ； w represents the weight vector, b represents the offset, and C represents the penalty factor. represents the slack variable; The constraints are: ； Represents a high-dimensional mapping; Step 24: Use the Lagrange multiplier method to transform the original problem into the dual problem; ； The constraints are: ； represents the Lagrange multiplier; Step 25: Construct a decision function based on the solution result; ; m represents the index of the SVM classifier; represents the temporary binary label of the m-th classifier; represents the Lagrange multiplier of the m-th classifier; Represents the offset of the mth classifier; Represents the classification result of sample x by the mth classifier; Step 26: Input the collected cycle features into the trained SVM classifier to obtain the classification results of all SVM classifiers, and take the label with the largest decision value as the output.

6. The method for evaluating the health status of energy storage batteries based on a neural network model according to claim 1, characterized in that: Step 3 includes the following steps: Step 31: Collecting several adjacent original relaxation characteristics of the energy storage battery, the original relaxation characteristics include VC, VD, ReVD, tVD, and tReVD; wherein VC represents the voltage at the start of charge relaxation; VD represents the voltage drop amplitude at the initial relaxation stage; ReVD represents the voltage value at the end of the relaxation process; tVD represents the time required for the voltage to drop to a preset amplitude; and tReVD represents the time required for the voltage to drop close to ReVD; Step 32: Performing continuity analysis on the collected original relaxation characteristics and the historical relaxation characteristics of the energy storage battery, and selecting the original relaxation characteristics with the highest continuity with the historical relaxation characteristics from a number of original relaxation characteristics; Step 33: Extract a sparse feature set from the Vt curve corresponding to the original relaxation feature.

7. The method for evaluating the health status of energy storage batteries based on a neural network model according to claim 1, characterized in that: Step 32 includes the following steps: Step 321: obtaining the initial life span of the energy storage battery calculated when collecting the original relaxation characteristics; Obtain the initial lifetime interval corresponding to the historical relaxation characteristics; Step 322: Arrange the VCs in the historical relaxation features and the original relaxation features in chronological order to generate a VC sequence; Arrange the historical relaxation features VD in chronological order to generate a VD sequence; Arrange the historical relaxation features ReVD in chronological order to generate a ReVD sequence; Arrange the historical relaxation features tVD in chronological order to generate a tVD sequence; Arrange the historical relaxation features tReVD in chronological order to generate a tReVD sequence; Arrange the initial life span intervals corresponding to the historical relaxation characteristics in chronological order to generate a life span sequence; Step 323: In the VC sequence, the VC corresponding to the original relaxation characteristics of several adjacent times is sequentially added, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics of the time, and the calculated continuity coefficient of the VC sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the VC in the kth original relaxation feature to the VC sequence and calculating the lifetime sequence; In the VD sequence, the VD corresponding to the original relaxation characteristics of several adjacent times is added in sequence, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the VD sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the VD in the kth original relaxation feature to the VD sequence and calculating the lifetime sequence; In the ReVD sequence, the ReVD corresponding to the original relaxation characteristics of several adjacent times is added in sequence, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the ReVD sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the ReVD in the kth original relaxation feature to the ReVD sequence and calculating the lifetime sequence; In the tVD sequence, the tVD corresponding to the original relaxation characteristics of several adjacent times is sequentially added, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the tVD sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the tVD in the kth original relaxation feature to the tVD sequence and calculating the lifetime sequence; In the tReVD sequence, the tReVD corresponding to the original relaxation characteristics of several adjacent times is sequentially added, and the life sequence is added with the initial life interval corresponding to the original relaxation characteristics. The calculated continuity coefficient of the VC sequence and the life sequence is obtained. ; k represents the index of the original relaxation feature, It represents the continuity coefficient obtained by adding the tReVD in the k-th original relaxation feature to the tReVD sequence and calculating the lifetime sequence; Step 324: Calculate the sum of the continuity coefficients R for each original relaxation feature k k ; ； Step 325: The sum of the continuity coefficients R k The highest original relaxation feature k is taken as the original relaxation feature with the highest continuity with the historical relaxation features.

8. The method for evaluating the health status of energy storage batteries based on a neural network model according to claim 6, wherein: Continuity coefficient The calculation formula is as follows: ; Among them, X i Represents the value of the i-th element in the VC sequence, represents the average value of the VC sequence; represents the mean value of the life series, Represents the value of the i-th element of the lifespan sequence; Continuity coefficient 、 、 、 、 is calculated in the same way.

9. The method for evaluating the health status of energy storage batteries based on a neural network model according to claim 1, characterized in that: Step 4 includes the following steps: Step 41: Obtain a sparse feature set, align each feature in the sparse feature set with each other according to time, and expand the sparse feature set into a two-dimensional matrix, where the two-dimensional matrix represents the type of the sparse set of behavioral features, and the columns of the two-dimensional matrix represent time; Step 42: Adding data values ​​of the sparse feature set to the two-dimensional matrix to form a third dimension, thereby converting the two-dimensional matrix into three-dimensional information; Step 43: Perform data enhancement on the three-dimensional information to obtain a key feature matrix.

10. The method for evaluating the health status of energy storage batteries based on a neural network model according to claim 9, characterized in that: Neural network models include: The input layer is used to receive the key feature matrix R; Convolutional layer, used to extract the key feature matrix R for convolution operation to generate feature maps; BN layer, used to standardize the feature map; Activation layer, used to activate the standardized features; Pooling layer, which performs maximum pooling on the output of the activation layer; Connection layer, connects the output of the pooling layer; The output layer outputs the precise lifespan information corresponding to the key features.

Images