Battery classification method and device, storage medium and electronic device

Abstract

The application provides a battery classification method and device, a storage medium and an electronic device, and relates to the technical field of electrics.The method comprises the following steps: processing test data of a battery to be classified to obtain standardized data; using a trained LSTM network model to extract features of the standardized data to obtain a first feature vector; obtaining a second feature vector based on the test data of the battery to be classified; and taking the first feature vector and the second feature vector as inputs of a trained SVM model to obtain a classification result of the battery to be classified.The technical scheme provided by the application can effectively classify batteries with different performances in a simple and accurate manner.

Description

Technical Field

[0001] This invention relates to the field of electrical engineering, and in particular to a battery sorting method, apparatus, storage medium, and electronic device. Background Technology

[0002] In the development of new battery processes or the research and use of commercial batteries, it is necessary to verify the performance of the batteries. Existing technologies verify battery performance by predicting the remaining lifespan of the battery after multiple charge-discharge cycles.

[0003] Currently, most methods for predicting battery remaining life use pre-trained neural network models. The main drawbacks of existing neural network-based battery remaining life prediction technologies are: either they fail to extract effective feature values, leading to inaccurate assessments of battery performance; or they require long charge-discharge cycles to obtain sufficient prediction data, making the entire assessment process cumbersome and computationally intensive. Summary of the Invention

[0004] To address the problems in the prior art, this application proposes a battery classification method, apparatus, storage medium, and electronic device that can easily and accurately classify batteries with different performance characteristics.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0006] In a first aspect, embodiments of the present invention provide a battery classification method, the method comprising:

[0007] The test data of the batteries to be classified are processed to obtain standardized data;

[0008] The trained LSTM network model is used to extract features from the standardized data to obtain the first feature vector;

[0009] A second feature vector is obtained based on the test data of the batteries to be classified;

[0010] The first feature vector and the second feature vector are used as input to a trained SVM model to obtain the classification result of the battery to be classified.

[0011] Preferably, there are multiple test data sets for the batteries to be classified; the process of processing the test data to obtain standardized data includes:

[0012] Each of the test data is processed as follows to obtain multiple sets of standardized data:

[0013] The test data was preprocessed to obtain smooth, noise-free data.

[0014] The smoothed data is converted into a sequence signal within a preset charge-discharge test cycle based on a preset sampling frequency.

[0015] Instantaneous frequency analysis and spectral entropy analysis are performed on the sequence signal to obtain the instantaneous frequency signal and spectral entropy signal of the sequence signal;

[0016] The instantaneous frequency signal and the spectral entropy signal are combined to obtain a combined signal;

[0017] The merged signal is standardized to obtain the standardized data.

[0018] Preferably, the step of preprocessing the test data to obtain noise-free, smooth data includes:

[0019] The test data is then normalized to obtain normalized data.

[0020] The standardized data is filtered to obtain filtered data;

[0021] The filtered data is subjected to a preset denoising process to obtain the smoothed data.

[0022] Furthermore, the step of preprocessing the test data to obtain noise-free, smooth data also includes:

[0023] Before performing preset denoising processing on the filtered data to obtain the smoothed data, it is determined whether there are outliers in the filtered data;

[0024] When outliers exist in the filtered data, linear interpolation is used to fit the outliers based on the two sampling points before and after the outliers to obtain the fitted filtered data.

[0025] The step of performing preset denoising processing on the filtered data to obtain the smoothed data includes:

[0026] The fitted filtered data is subjected to a preset denoising process to obtain the smoothed data.

[0027] Preferably, the Savitzky-Golay filtering method is used to filter the per-unit data; an outlier in the filtered data is determined using the following method:

[0028] Calculate the difference between two adjacent data points in the filtered data;

[0029] When the difference satisfies the following relationship, it is determined that there is an outlier among the two adjacent data points:

[0030]

[0031] Wherein, ΔL is the difference; T is the sampling time interval of the test data of the battery to be classified; C Rate The discharge rate C of the batteries to be classified; k outlier The outlier detection coefficient is a preset value.

[0032] Provided that the aforementioned relation is satisfied, among two adjacent data points, if a data point satisfies the following condition, that data point is determined to be an outlier:

[0033] In the interval centered on this data Within a certain range, the difference between the filtered data and the corresponding standardized data is greater than a first preset threshold; within the interval centered on this data... In addition, the difference between the filtered data and the corresponding standardized data is less than a second preset threshold; wherein, L window This refers to the window length used in the Savitzky-Golay filtering method.

[0034] Furthermore, the step of preprocessing the test data to obtain noise-free, smooth data also includes:

[0035] Determine whether there is noise in the filtered data that meets preset conditions;

[0036] When the filtered data contains noise that meets the preset conditions, the filtered data is subjected to the preset denoising process to obtain the smoothed data.

[0037] Preferably, the Savitzky-Golay filtering method is used to filter the per-unit data; the following method is used to determine whether there is noise in the filtered data that meets the preset conditions:

[0038] The parameter L is calculated using the following expression. noise and parameter L noiseLimit :

[0039]

[0040]

[0041] Among them, L window The window length used in the Savitzky-Golay filtering method; i is the sequence number of the per-unit data within the window length; D i The value of the i-th standardized data within a given window length; median(L) window ) represents the interval [1+(i-1)×L window i×L window The median of all the standardized data within the range; knoise The noise judgment coefficient is the preset value; T is the sampling time interval of the test data of the battery to be classified; C Rate The discharge rate C of the battery to be classified;

[0042] When L noise ≥L noiseLimit At that time, it is confirmed that there is noise in the filtered data that meets the preset conditions.

[0043] Preferably, the LSTM network model includes an input layer, an LSTM layer, a fully connected layer, a softmax layer, and a classification layer connected in sequence;

[0044] The input layer is used to receive the standardized data, process the standardized data, and output time series data.

[0045] The LSTM layer is used to learn the long-term dependency between the time step and the corresponding data value in the time series data, and to obtain the result of the instantaneous frequency signal after LSTM training and the result of the spectral entropy signal after LSTM training.

[0046] The fully connected layer is used to connect the result of the instantaneous frequency signal trained by LSTM and the result of the spectral entropy signal trained by LSTM into two-dimensional sequence data;

[0047] The softmax layer is used to limit the maximum value of the two-dimensional sequence data;

[0048] The classification layer is used to classify the two-dimensional sequence data after the limit is applied in order to obtain the first feature vector.

[0049] Preferably, obtaining the second feature vector based on the test data of the battery to be classified includes:

[0050] The test data of the batteries to be classified are preprocessed to obtain smooth data without noise.

[0051] The smoothed data is converted into a sequence signal within a preset charge-discharge test cycle based on a preset sampling frequency.

[0052] The peak-to-peak value, average value, standard deviation, kurtosis, and skewness of the sequence signal within the preset charge-discharge test cycle are calculated and used as the second feature vector.

[0053] Preferably, the test data of the batteries to be classified includes at least one of the following items:

[0054] Data on the changes in charging voltage over time, data on the changes in discharging voltage over time, data on the changes in charging current over time, data on the changes in discharging current over time, data on the changes in battery capacity over time, data on the changes in battery capacity with discharging voltage, data on the changes in battery capacity with charging voltage, data on the changes in battery capacity increment with discharging voltage, and data on the changes in battery capacity increment with charging voltage.

[0055] Secondly, embodiments of the present invention provide a battery sorting device, the device comprising:

[0056] The data processing module is used to process the test data of the batteries to be classified and obtain standardized data;

[0057] The first feature extraction module is used to extract features from the standardized data using a trained LSTM network model to obtain a first feature vector.

[0058] The second feature extraction module is used to obtain a second feature vector based on the test data of the battery to be classified;

[0059] The classification module is used to take the first feature vector and the second feature vector as input to a trained SVM model to obtain the classification result of the battery to be classified.

[0060] Thirdly, embodiments of the present invention provide a storage medium storing program code, which, when executed by a processor, implements the battery sorting method as described in any of the above embodiments.

[0061] Fourthly, embodiments of the present invention provide an electronic device, the electronic device including a memory and a processor, the memory storing program code executable on the processor, the program code being executed by the processor to implement the battery sorting method as described in any of the above embodiments.

[0062] This invention provides a battery classification method, apparatus, storage medium, and electronic device. By processing test data of batteries to be classified to obtain standardized data, a trained LSTM network model is used to extract features from this standardized data to obtain a first feature vector. A second feature vector obtained based on the test data of the batteries to be classified, along with the first feature vector, is used as input to a trained SVM model to obtain the classification result of the batteries. In other words, this invention combines an LSTM network model and an SVM model to classify batteries with different performance characteristics. The LSTM network model can effectively extract most of the main features of the test data without requiring a large amount of battery test data, thus eliminating the need for long charge-discharge cycle tests. Using the first and second feature vectors together as input to the SVM model comprehensively characterizes the battery features, enabling the SVM model to classify batteries more comprehensively and effectively based on the input feature vectors. Therefore, compared with existing technologies, the technical solution provided by this invention requires less test data and obtains more comprehensive battery feature values, thus enabling simpler and more accurate effective classification of batteries with different performance characteristics. Attached Figure Description

[0063] The scope of this invention can be better understood by reading the following detailed description of exemplary embodiments in conjunction with the accompanying drawings. The accompanying drawings are:

[0064] Figure 1 The method flow of this invention embodiment Figure 1 ;

[0065] Figure 2 The method flow of this invention embodiment Figure 2 ;

[0066] Figure 3 This is a schematic diagram illustrating the merging of sequence signals within each charge-discharge test cycle in an embodiment of the present invention;

[0067] Figure 4 This is a per-unit curve of the discharge capacity of a certain battery cell in an embodiment of the present invention.

[0068] Figure 5 for Figure 4 The curve data in the image is a curve after Savitzky-Golay filtering;

[0069] Figure 6 This is a schematic diagram illustrating the outlier detection process in an embodiment of the present invention;

[0070] Figure 7 This is a schematic diagram illustrating the processing of outliers in an embodiment of the present invention;

[0071] Figure 8 This is a graph showing the discharge capacity of a single battery cell after outlier processing in an embodiment of the present invention.

[0072] Figure 9 This is a segment of noisy battery discharge capacity data in an embodiment of the present invention;

[0073] Figure 10 To Figure 9 A schematic diagram showing the data after noise calculation.

[0074] Figure 11 To Figure 9 The result after the data in the image has undergone preset noise reduction processing;

[0075] Figure 12A This is a schematic diagram of the instantaneous frequency signal obtained by performing instantaneous frequency analysis on the sequence signal in an embodiment of the present invention;

[0076] Figure 12B This is a schematic diagram of the spectral entropy signal obtained by performing spectral entropy analysis on the sequence signal in an embodiment of the present invention;

[0077] Figure 13 This is a structural diagram of the device according to an embodiment of the present invention. Detailed Implementation

[0078] To make the objectives, technical solutions, and advantages of the present invention clearer, the implementation method of the present invention will be described in detail below with reference to the accompanying drawings and embodiments, so that the process of how the present invention uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0079] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0080] Example 1

[0081] According to embodiments of the present invention, a battery classification method is provided, such as... Figure 1 and Figure 2 As shown, the method described in this embodiment includes:

[0082] Step S101: Process the test data of the batteries to be classified to obtain standardized data;

[0083] In this embodiment, the test data of the battery to be classified may include multiple data, including but not limited to: data on the change of charging voltage over time, data on the change of discharging voltage over time, data on the change of charging current over time, data on the change of discharging current over time, data on the change of battery capacity over time, data on the change of battery capacity with discharging voltage, data on the change of battery capacity with charging voltage, data on the change of battery capacity increment with discharging voltage, and data on the change of battery capacity increment with charging voltage.

[0084] The battery test data in this embodiment includes at least one of the above items.

[0085] In this embodiment, step S101 specifically includes: performing the following processing on each of the test data to obtain multiple standardized data: performing data preprocessing on the test data to obtain noise-free smooth data; forming a sequence signal from the smooth data within a preset charge-discharge test cycle based on a preset sampling frequency; performing instantaneous frequency analysis and spectral entropy analysis on the sequence signal to obtain the instantaneous frequency signal and spectral entropy signal of the sequence signal; performing merging processing on the instantaneous frequency signal and the spectral entropy signal to obtain a merged signal; and performing standardization processing on the merged signal to obtain the standardized data.

[0086] In this embodiment, the step of preprocessing the test data to obtain noise-free smooth data includes: normalizing the test data to obtain normalized data; filtering the normalized data to obtain filtered data; and performing preset denoising processing on the filtered data to obtain the smooth data.

[0087] Specifically, the following expression is used to normalize a set of test data:

[0088]

[0089] Among them, D p.u. Let D be the standardized data, and D be the test data. Base This is the preset baseline data.

[0090] When the test data D is voltage data, the rated voltage of the battery is used as the reference data, and the above-mentioned per-unit processing is performed; when the test data D is current data, the median current during battery operation is used as the reference data, and the above-mentioned per-unit processing is performed; when the test data D is capacity data, the rated capacity of the battery is used as the reference data, and the above-mentioned per-unit processing is performed; when the test data D is temperature data, the median battery temperature during battery operation is used as the reference data, and the above-mentioned per-unit processing is performed.

[0091] To remove obvious outliers obtained during sampling and make the processed data smoother and more accurate, in this embodiment, before performing preset denoising processing on the filtered data to obtain the smoothed data, it is further possible to determine whether there are outliers in the filtered data. When outliers are found, a linear interpolation method is used to fit the outlier based on the two sampling points before and after it to obtain the fitted filtered data. Therefore, the step of performing preset denoising processing on the filtered data to obtain the smoothed data specifically includes: performing preset denoising processing on the fitted filtered data to obtain the smoothed data.

[0092] In this embodiment, the Savitzky-Golay filtering method is used to filter the per-unit data. The Savitzky-Golay filtering method is a filtering method based on local polynomial least squares fitting in the time domain. Its greatest advantage is that it can remove noise while ensuring that the shape and width of the signal remain unchanged. The main parameters that need to be determined for the Savitzky-Golay filtering method are the window length and order used during filtering.

[0093] Specifically, the following method is used to determine if a data point in the filtered data is an outlier:

[0094] Calculate the difference between two adjacent data points in the filtered data;

[0095] When the difference satisfies the following relationship (criterion 1), it is determined that there is an outlier among the two adjacent data points:

[0096]

[0097] Wherein, ΔL is the difference; T is the sampling time interval of the test data of the battery to be classified; C Rate The discharge rate C of the battery to be classified; k outlier This is the preset outlier detection coefficient.

[0098] Under the premise of satisfying the above equation (2), among the two adjacent data, if a data satisfies the following condition (criterion 2), the data is judged to be an outlier:

[0099] In the interval centered on this data Within a certain range, the difference between the filtered data and the corresponding standardized data is greater than a first preset threshold; within the interval centered on this data... In addition, the difference between the filtered data and the corresponding standardized data is less than a second preset threshold; wherein, L window This refers to the window length used in the Savitzky-Golay filtering method.

[0100] That is, the first criterion is used to determine that there is an outlier in two adjacent data points of the filtered data, and the second criterion is used to further determine which specific data point in the two adjacent data points is the outlier.

[0101] To reduce unnecessary filtering operations and computational overhead, in this embodiment, before performing preset denoising processing on the filtered data, it can be determined whether there is noise in the filtered data that meets preset conditions; when there is noise in the filtered data that meets the preset conditions, the preset denoising processing is performed on the filtered data to obtain the smoothed data.

[0102] Specifically, the following method is used to determine whether there is noise in the filtered data that meets the preset conditions:

[0103] The parameter L is calculated using the following expression. noise and parameter L noiseLimit :

[0104]

[0105]

[0106] Among them, L window The window length used in the Savitzky-Golay filtering method; i is the sequence number of the per-unit data within the window length; D i The value of the i-th standardized data within a given window length; median(L) window ) represents the interval [1+(i-1)×L window i×L window The median of all the standardized data within the range; k noise The noise judgment coefficient is the preset value; T is the sampling time interval of the test data of the battery to be classified; C Rate The discharge rate C of the battery to be classified;

[0107] When L noise ≥L noiseLimit At that time, it is confirmed that there is noise in the filtered data that meets the preset conditions.

[0108] That is, when the noise value L calculated by the above formula (3) is... noise Reaching or exceeding the noise threshold L calculated by equation (4) above noiseLimit If noise is found in the filtered data, further denoising processing is required.

[0109] In this embodiment, the merged signal is standardized using the following expression to obtain the standardized data:

[0110]

[0111] Where X' is the standardized data, X is the combined signal, mean(X) is the average value of the combined signal, and std(X) is the standard deviation of the combined signal.

[0112] Step S102: Use the trained LSTM network model to extract features from the standardized data to obtain the first feature vector;

[0113] In this embodiment, the LSTM (Long Short-Term Memory) network model includes an input layer, an LSTM layer, a fully connected layer, a softmax layer, and a classification layer connected in sequence. The input layer receives the standardized data obtained in step S101, processes the standardized data, and outputs time-series data. The LSTM layer learns the long-term dependency between the time step and the corresponding data value in the time-series data, obtaining the results of LSTM training on the instantaneous frequency signal and the LSTM training on the spectral entropy signal. The fully connected layer concatenates the results of LSTM training on the instantaneous frequency signal and the LSTM training on the spectral entropy signal into two-dimensional sequence data.

[0114] The softmax layer is used to limit the maximum value of the two-dimensional sequence data; the classification layer is used to classify the two-dimensional sequence data after the limit is applied to obtain the first feature vector.

[0115] When training the LSTM network model described above, the sample data is randomly divided into a training set and a test set, with the training set comprising 90% and the test set comprising 10%. The LSTM network model uses an LSTM layer with an input sequence size of 2 and an output size of 100, outputting the last element of the sequence. Classification training is performed using a fully connected layer of size 2, a softmax layer, and a classification layer.

[0116] Step S103: Obtain the second feature vector based on the test data of the battery to be classified;

[0117] In this embodiment, step S103 specifically includes:

[0118] The test data of the batteries to be classified are preprocessed to obtain noise-free smooth data; the smooth data is formed into a sequence signal within a preset charge-discharge test cycle based on a preset sampling frequency; the peak-to-peak value, average value, standard deviation, kurtosis and skewness of the sequence signal within the preset charge-discharge test cycle are calculated as the second feature vector.

[0119] In this embodiment, there are multiple test data for the batteries to be classified. The above process is as follows: each test data is preprocessed to obtain multiple noise-free smooth data; each smooth data is formed into a sequence signal within a preset charge-discharge test cycle based on a preset sampling frequency; the peak-to-peak value, average value, standard deviation, kurtosis and skewness of each sequence signal within the preset charge-discharge test cycle are calculated as the second feature vector.

[0120] The process involves preprocessing each set of test data to obtain multiple noise-free smooth data sets. This includes: normalizing the test data to obtain normalized data; filtering the normalized data to obtain filtered data; determining whether outliers exist in the filtered data; when outliers exist, fitting the outlier to the two sampling points before and after the outlier using linear interpolation to obtain fitted filtered data; determining whether noise satisfying preset conditions exists in the fitted filtered data; and when noise satisfying preset conditions exists in the fitted filtered data, performing preset denoising processing on the fitted filtered data to obtain the smooth data.

[0121] The above schematic diagram illustrates how each smoothed data point is formed into a sequence signal within a preset charge-discharge test cycle based on a preset sampling frequency. Figure 3 As shown. The preset charge / discharge test cycle refers to... Figure 3 The duration formed by N consecutive cyclic cycles is the time to merge the sequence signals within that duration, and each test data can form a sequence signal.

[0122] Calculate the peak-to-peak value, average value, standard deviation, kurtosis, and skewness of each sequence signal within the preset charge-discharge test cycle. Use these calculated peak-to-peak value, average value, standard deviation, kurtosis, and skewness as the second feature vector, and input them together into the subsequent SVM (Support Vector Machine) model for classification.

[0123] Step S104: Use the first feature vector and the second feature vector as input to the trained SVM model to obtain the classification result of the battery to be classified.

[0124] Among them, the SVM model is a type of generalized linear classifier that performs binary classification of data using supervised learning. It can effectively classify batteries based on the input feature vector.

[0125] This invention provides a battery classification method. It processes test data of batteries to be classified to obtain standardized data, uses a trained LSTM network model to extract features from this standardized data to obtain a first feature vector, and uses a second feature vector obtained from the test data of the batteries to be classified, along with the first feature vector, as input to a trained SVM model to obtain the classification result of the batteries. In other words, this invention combines an LSTM network model and an SVM model to classify batteries with different performance characteristics. The LSTM network model can effectively extract most of the main features of the test data without requiring a large amount of battery test data, thus eliminating the need for long charge-discharge cycle tests. Using the first and second feature vectors together as input to the SVM model comprehensively characterizes the battery features, enabling the SVM model to classify batteries more comprehensively and effectively based on the input feature vectors. Therefore, compared with existing technologies, the technical solution provided by this invention requires less test data and obtains more comprehensive battery feature values, thus enabling simpler and more accurate effective classification of batteries with different performance characteristics.

[0126] Example 2

[0127] This embodiment uses the actual process of classifying batteries as an example to describe in detail the battery classification method, which includes:

[0128] Step S201: Standardize the battery test data to obtain standardized data;

[0129] Taking the discharge capacity of the battery as an example, the rated capacity of the battery actually tested in this embodiment is 1.1Ah. Therefore, 1.1Ah is used as the baseline data for discharge capacity to normalize the actual test data:

[0130]

[0131] In this embodiment, 124 batteries are used. Taking the discharge capacity data of a single battery cell as an example, the normalized battery discharge capacity is as follows: Figure 4 As shown.

[0132] Of course, the test data in this embodiment may also include data such as voltage, current, and temperature, and there are no restrictions here.

[0133] Step S202: Filter the standardized data to obtain filtered data;

[0134] In this embodiment, the Savitzky-Golay filtering method is used to filter the per-unit data. The window length L used in the Savitzky-Golay filtering method is...window =21, order n=3.

[0135] Figure 5 This demonstrates the application of the Savitzky-Golay filtering method to... Figure 4 The curve of the battery discharge capacity after filtering the normalized data shows that the filtered data is smoother than the original data.

[0136] Step S203: Determine whether there are outliers in the filtered data. If there are outliers, process the outliers.

[0137] Specifically, the following method is used to determine if a data point in the filtered data is an outlier:

[0138] Calculate the difference between two adjacent data points in the filtered data;

[0139] When the difference satisfies the following relationship (criterion 1), it is determined that there is an outlier among the two adjacent data points:

[0140]

[0141] Wherein, ΔL is the difference; T is the sampling time interval of the test data of the battery to be classified; C Rate The discharge rate C of the batteries to be classified; k outlier This is the preset outlier detection coefficient.

[0142] Under the premise of satisfying the above equation (2), among the two adjacent data, if a data satisfies the following condition (criterion 2), the data is judged to be an outlier:

[0143] In the interval centered on this data Within a certain range, the difference between the filtered data and the corresponding standardized data is greater than a first preset threshold; within the interval centered on this data... In addition, the difference between the filtered data and the corresponding standardized data is less than a second preset threshold; wherein, L window This refers to the window length used in the Savitzky-Golay filtering method.

[0144] That is, the first criterion is used to determine that there is an outlier in two adjacent data points of the filtered data, and the second criterion is used to further determine which specific data point in the two adjacent data points is the outlier.

[0145] In this embodiment, ko utlier =10, C Rate =4C, T=5s, set L can then be calculated using this formula. outlier=0.056. When the difference between two adjacent data points in the filtered data reaches or exceeds the above limit of 0.056, it is determined that there is an outlier in the two adjacent data points.

[0146] Because of L window =21, therefore, among the two adjacent data points, a data point is considered an outlier if it meets the following condition:

[0147] Within the interval [-10, 10] centered on the data, the difference between the filtered data and the corresponding standardized data is greater than a first preset threshold; outside the interval [-10, 10] centered on the data, the difference between the filtered data and the corresponding standardized data is less than a second preset threshold. The first and second preset thresholds can be set according to actual needs.

[0148] A schematic diagram illustrating the outlier detection method described above is shown below. Figure 6 As shown in the diagram, after finding outliers, linear interpolation is used to fit the outlier based on the two sampling points before and after it. The fitted value is then used to replace the outlier, resulting in the fitted filtered data. Figure 7 The fitted values ​​are shown in the figure.

[0149] The discharge capacity curve of the battery cell after processing outliers is shown in the figure below. Figure 8 As shown, after the above processing, there are no obvious abnormal data in the discharge capacity curve, and the data is smoother.

[0150] Step S204: Determine whether there is noise in the filtered data that meets the preset conditions. If so, perform preset denoising processing on the filtered data to obtain smooth data.

[0151] The following method is used to determine whether noise satisfying the preset conditions exists in the filtered data:

[0152] The parameter L is calculated using the following expression. noise and parameter L noiseLimit :

[0153]

[0154]

[0155] Among them, L window The window length used in the Savitzky-Golay filtering method; i is the sequence number of the per-unit data within the window length; D i The value of the i-th standardized data within a given window length; median(L) window ) represents the interval [1+(i-1)×L windowi×L window The median of all the standardized data within the range; k noise The noise judgment coefficient is the preset value; T is the sampling time interval of the test data of the battery to be classified; C Rate The discharge rate C of the battery to be classified;

[0156] When L noise ≥L noiseLimit At that time, it is confirmed that there is noise in the filtered data that meets the preset conditions.

[0157] Figure 9 The example shows a segment of battery discharge capacity data containing noise. In this embodiment, L window =21,k noise =4, L calculated according to the above formula (4) noiseLimit =0.0003, and the parameters in equation (3) can be determined based on the Savitzky-Golay filter order n, thereby further calculating L. noise The value of .

[0158] Figure 10 It shows the Figure 9 A schematic diagram showing the data after noise calculation.

[0159] When the above method determines that further denoising processing of the filtered data is needed, further Savitzky-Golay filtering is performed on the filtered data. Figure 11 It shows the Figure 9 The result of performing preset noise reduction on the data is the obtained smooth data.

[0160] Step S205: Based on a preset sampling frequency, the smoothed data is converted into a sequence signal within a preset charge-discharge test cycle.

[0161] A schematic diagram illustrating the process of forming a sequence signal from smoothed data within a preset charge-discharge test cycle based on a preset sampling frequency is shown below. Figure 3 As shown. The preset charge / discharge test cycle refers to... Figure 3 The duration formed by N consecutive cyclic cycles is the time to merge the sequence signals within that duration, and each test data can form a sequence signal.

[0162] In this embodiment, the preset charge-discharge test cycle is the duration of 5 charge-discharge cycles, and the battery test data includes, but is not limited to: data on the change of charging voltage over time, data on the change of discharging voltage over time, data on the change of charging current over time, data on the change of discharging current over time, data on the change of battery capacity over time, data on the change of battery capacity with discharging voltage, data on the change of battery capacity with charging voltage, data on the change of battery capacity increment with discharging voltage, and data on the change of battery capacity increment with charging voltage.

[0163] Step S206: Perform instantaneous frequency analysis and spectral entropy analysis on the sequence signal to obtain the instantaneous frequency signal and spectral entropy signal of the sequence signal;

[0164] In this embodiment of the invention, Figure 12A This diagram illustrates the instantaneous frequency signal obtained by performing instantaneous frequency analysis on a sequence signal. Figure 12B A schematic diagram of the spectral entropy signal obtained by performing spectral entropy analysis on a sequence signal is shown.

[0165] Step S207: Combine the instantaneous frequency signal and the spectral entropy signal to obtain a combined signal;

[0166] Step S208: Standardize the merged signal to obtain standardized data;

[0167] In this embodiment, the merged signal is standardized using the following expression to obtain the standardized data:

[0168]

[0169] Where X' is the standardized data, X is the combined signal, mean(X) is the average value of the combined signal, and std(X) is the standard deviation of the combined signal.

[0170] Step S209: Use the trained LSTM network model to extract features from the standardized data to obtain the first feature vector;

[0171] In this embodiment, the LSTM network model includes an input layer, an LSTM layer, a fully connected layer, a softmax layer, and a classification layer connected in sequence. The input layer receives the standardized data obtained in step S101, processes the standardized data, and outputs time-series data. The LSTM layer learns the long-term dependency between the time step and the corresponding data value in the time-series data, obtaining the results of LSTM training on the instantaneous frequency signal and the LSTM training on the spectral entropy signal. The fully connected layer concatenates the results of LSTM training on the instantaneous frequency signal and the LSTM training on the spectral entropy signal into two-dimensional sequence data. The softmax layer limits the maximum value of the two-dimensional sequence data. The classification layer classifies the limited two-dimensional sequence data to obtain the first feature vector.

[0172] Step S210: Obtain the second feature vector based on the test data of the batteries to be classified;

[0173] In this embodiment, the second feature vector is obtained in the following way: the test data of the batteries to be classified is preprocessed to obtain noise-free smooth data; the smooth data is formed into a sequence signal within a preset charge-discharge test cycle based on a preset sampling frequency; the peak-to-peak value, average value, standard deviation, kurtosis and skewness of the sequence signal within the preset charge-discharge test cycle are calculated as the second feature vector.

[0174] For detailed implementation processes of acquiring noise-free smooth data and forming sequence signals, please refer to Example 1, which will not be repeated here.

[0175] Step S211: Use the first feature vector and the second feature vector as input to the trained SVM model to obtain the classification result of the battery to be classified.

[0176] Practice shows that the prediction accuracy of the SVM model in this embodiment is 93%, which meets the needs of practical applications.

[0177] This invention provides a battery classification method. It processes test data of batteries to be classified to obtain standardized data, uses a trained LSTM network model to extract features from this standardized data to obtain a first feature vector, and uses a second feature vector obtained from the test data of the batteries to be classified, along with the first feature vector, as input to a trained SVM model to obtain the classification result of the batteries. In other words, this invention combines an LSTM network model and an SVM model to classify batteries with different performance characteristics. The LSTM network model can effectively extract most of the main features of the test data without requiring a large amount of battery test data, thus eliminating the need for long charge-discharge cycle tests. Using the first and second feature vectors together as input to the SVM model comprehensively characterizes the battery features, enabling the SVM model to classify batteries more comprehensively and effectively based on the input feature vectors. Therefore, compared with existing technologies, the technical solution provided by this invention requires less test data and obtains more comprehensive battery feature values, thus enabling simpler and more accurate effective classification of batteries with different performance characteristics.

[0178] Example 3

[0179] Corresponding to the above method embodiments, the present invention also provides a battery sorting device, such as... Figure 13 As shown, the device includes:

[0180] The data processing module 301 is used to process the test data of the batteries to be classified and obtain standardized data.

[0181] The first feature extraction module 302 is used to extract features from the standardized data using a trained LSTM network model to obtain a first feature vector;

[0182] The second feature extraction module 303 is used to obtain a second feature vector based on the test data of the battery to be classified;

[0183] The classification module 304 is used to take the first feature vector and the second feature vector as input to a trained SVM model to obtain the classification result of the battery to be classified.

[0184] For details regarding the working principle, workflow, and specific implementation methods of the aforementioned device, please refer to the specific implementation methods of the battery classification method provided by this invention. The same technical content will not be described in detail here.

[0185] Example 4

[0186] According to an embodiment of the present invention, a storage medium is also provided, wherein program code is stored on the storage medium, and when the program code is executed by a processor, the battery sorting method as described in any of the above embodiments is implemented.

[0187] Example 5

[0188] According to an embodiment of the present invention, an electronic device is also provided, the electronic device including a memory and a processor, wherein the memory stores program code that can run on the processor, and when the program code is executed by the processor, it implements the battery sorting method as described in any of the above embodiments.

[0189] This invention provides a battery classification method, apparatus, storage medium, and electronic device. By processing test data of batteries to be classified to obtain standardized data, a trained LSTM network model is used to extract features from this standardized data to obtain a first feature vector. A second feature vector obtained based on the test data of the batteries to be classified, along with the first feature vector, is used as input to a trained SVM model to obtain the classification result of the batteries. In other words, this invention combines an LSTM network model and an SVM model to classify batteries with different performance characteristics. The LSTM network model can effectively extract most of the main features of the test data without requiring a large amount of battery test data, thus eliminating the need for long charge-discharge cycle tests. Using the first and second feature vectors together as input to the SVM model comprehensively characterizes the battery features, enabling the SVM model to classify batteries more comprehensively and effectively based on the input feature vectors. Therefore, compared with existing technologies, the technical solution provided by this invention requires less test data and obtains more comprehensive battery feature values, thus enabling simpler and more accurate effective classification of batteries with different performance characteristics.

[0190] This invention achieves battery classification through noise reduction processing, feature value extraction, and a classification method based on LSTM+SVM for battery test data. Practice shows that this invention can classify batteries based on their quality by measuring ultra-short cycle times (≤10 cycles), significantly improving the speed of developing and verifying new battery processes and greatly increasing the sorting speed of commercial batteries, thus possessing significant practical value.

[0191] The technical solution provided by this invention has a clear principle, a simple method, is easy to implement, and is convenient to operate, and has good application prospects.

[0192] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0193] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0194] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0195] If the integrated unit is implemented as 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 this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0196] While the embodiments disclosed in this invention are as described above, the content is merely for the purpose of facilitating understanding of the invention and is not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and changes in form and detail of the implementation without departing from the spirit and scope disclosed herein; however, the scope of protection of this invention shall still be determined by the scope defined in the appended claims.

Claims

1. A battery classification method, characterized in that, The method includes: The test data of the batteries to be classified are processed to obtain standardized data; The trained LSTM network model is used to extract features from the standardized data to obtain the first feature vector; A second feature vector is obtained based on the test data of the batteries to be classified; The first feature vector and the second feature vector are used as input to a trained SVM model to obtain the classification result of the battery to be classified. There are multiple test data sets for the batteries to be classified; the process of processing the test data to obtain standardized data includes: Each of the test data is processed as follows to obtain multiple sets of standardized data: The test data was preprocessed to obtain smooth, noise-free data. The smoothed data is converted into a sequence signal within a preset charge-discharge test cycle based on a preset sampling frequency. Instantaneous frequency analysis and spectral entropy analysis are performed on the sequence signal to obtain the instantaneous frequency signal and spectral entropy signal of the sequence signal; The instantaneous frequency signal and the spectral entropy signal are combined to obtain a combined signal; The merged signal is standardized to obtain the standardized data; The step of preprocessing the test data to obtain noise-free, smooth data includes: The test data is then normalized to obtain normalized data. The standardized data is filtered to obtain filtered data; The filtered data is subjected to a preset denoising process to obtain the smoothed data; Before performing preset denoising processing on the filtered data, the method further includes: Determine whether there is noise in the filtered data that meets preset conditions; When the filtered data contains noise that meets the preset conditions, the filtered data is subjected to the preset denoising process to obtain the smoothed data; The per-unit data is filtered using the Savitzky-Golay filtering method; the presence of noise satisfying the preset conditions in the filtered data is determined using the following method: The parameters are calculated using the following expressions. and parameters : ， in, The window length used in the Savitzky-Golay filtering method; i is the sequence number of the per-unit data within the window length; The value of the i-th standardized data within a given window length; For interval The median of all the aforementioned per-unit data; The preset noise judgment coefficient; The sampling time interval for the test data of the batteries to be classified; The discharge rate C of the battery to be classified; when At that time, it is confirmed that there is noise in the filtered data that meets the preset conditions.

2. The battery classification method according to claim 1, characterized in that, The step of preprocessing the test data to obtain noise-free, smooth data further includes: Before performing preset denoising processing on the filtered data to obtain the smoothed data, it is determined whether there are outliers in the filtered data; When outliers exist in the filtered data, linear interpolation is used to fit the outliers based on the two sampling points before and after the outliers to obtain the fitted filtered data. The step of performing preset denoising processing on the filtered data to obtain the smoothed data includes: The fitted filtered data is subjected to a preset denoising process to obtain the smoothed data.

3. The battery classification method according to claim 2, characterized in that, The Savitzky-Golay filtering method is used to filter the per-unit data; The following method is used to determine if a data point in the filtered data is an outlier: Calculate the difference between two adjacent data points in the filtered data; When the difference satisfies the following relationship, it is determined that there is an outlier among the two adjacent data points: in, The difference; The sampling time interval for the test data of the batteries to be classified; The discharge rate C of the battery to be classified; The outlier detection coefficient is a preset value. Provided that the aforementioned relation is satisfied, among two adjacent data points, if a data point satisfies the following condition, that data point is determined to be an outlier: In the interval centered on this data Within a certain range, the difference between the filtered data and the corresponding standardized data is greater than a first preset threshold; within the interval centered on this data... In addition, the difference between the filtered data and the corresponding standardized data is less than a second preset threshold; wherein, This refers to the window length used in the Savitzky-Golay filtering method.

4. The battery classification method according to claim 1, characterized in that, The LSTM network model includes an input layer, an LSTM layer, a fully connected layer, a softmax layer, and a classification layer connected in sequence. The input layer is used to receive the standardized data, process the standardized data, and output time series data. The LSTM layer is used to learn the long-term dependency between the time step and the corresponding data value in the time series data, and to obtain the result of the instantaneous frequency signal after LSTM training and the result of the spectral entropy signal after LSTM training. The fully connected layer is used to connect the result of the instantaneous frequency signal trained by LSTM and the result of the spectral entropy signal trained by LSTM into two-dimensional sequence data; The softmax layer is used to limit the maximum value of the two-dimensional sequence data; The classification layer is used to classify the two-dimensional sequence data after the limit is applied in order to obtain the first feature vector.

5. The battery classification method according to claim 1, characterized in that, The process of obtaining the second feature vector based on the test data of the battery to be classified includes: The test data of the batteries to be classified are preprocessed to obtain smooth data without noise. The smoothed data is converted into a sequence signal within a preset charge-discharge test cycle based on a preset sampling frequency. The peak-to-peak value, average value, standard deviation, kurtosis, and skewness of the sequence signal within the preset charge-discharge test cycle are calculated and used as the second feature vector.

6. The battery classification method according to claim 1, characterized in that, The test data for the batteries to be classified includes at least one of the following items: Data on the changes in charging voltage over time, data on the changes in discharging voltage over time, data on the changes in charging current over time, data on the changes in discharging current over time, data on the changes in battery capacity over time, data on the changes in battery capacity with discharging voltage, data on the changes in battery capacity with charging voltage, data on the changes in battery capacity increment with discharging voltage, and data on the changes in battery capacity increment with charging voltage.

7. A battery sorting device, characterized in that, The device includes: The data processing module is used to process the test data of the batteries to be classified and obtain standardized data; The first feature extraction module is used to extract features from the standardized data using a trained LSTM network model to obtain a first feature vector. There are multiple test data sets for the batteries to be classified; the process of processing the test data to obtain standardized data includes: Each of the test data is processed as follows to obtain multiple sets of standardized data: The test data was preprocessed to obtain smooth, noise-free data. The smoothed data is converted into a sequence signal within a preset charge-discharge test cycle based on a preset sampling frequency. Instantaneous frequency analysis and spectral entropy analysis are performed on the sequence signal to obtain the instantaneous frequency signal and spectral entropy signal of the sequence signal; The instantaneous frequency signal and the spectral entropy signal are combined to obtain a combined signal; The merged signal is standardized to obtain the standardized data; The step of preprocessing the test data to obtain noise-free, smooth data includes: The test data is then normalized to obtain normalized data. The standardized data is filtered to obtain filtered data; The filtered data is subjected to a preset denoising process to obtain the smoothed data; Before performing preset denoising processing on the filtered data, the device further includes: Determine whether there is noise in the filtered data that meets preset conditions; When the filtered data contains noise that meets the preset conditions, the filtered data is subjected to the preset denoising process to obtain the smoothed data; The per-unit data is filtered using the Savitzky-Golay filtering method; the presence of noise satisfying the preset conditions in the filtered data is determined using the following method: The parameters are calculated using the following expressions. and parameters : ， in, The window length used in the Savitzky-Golay filtering method; i is the sequence number of the per-unit data within the window length; The value of the i-th standardized data within a given window length; For interval The median of all the aforementioned per-unit data; The preset noise judgment coefficient; The sampling time interval for the test data of the batteries to be classified; The discharge rate C of the battery to be classified; when At that time, it is confirmed that there is noise in the filtered data that meets the preset conditions; The second feature extraction module is used to obtain a second feature vector based on the test data of the battery to be classified; The classification module is used to take the first feature vector and the second feature vector as input to a trained SVM model to obtain the classification result of the battery to be classified.

8. A storage medium storing program code, characterized in that, When the program code is executed by the processor, it implements the battery classification method as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores program code that can run on the processor. When the program code is executed by the processor, it implements the battery sorting method as described in any one of claims 1 to 6.

Images