Battery liquid leakage identification method, apparatus and device, and storage medium

Abstract

The invention discloses a battery liquid leakage identification method, device and equipment and a storage medium, which are applied to the technical field of battery liquid leakage identification, and are characterized in that after electrochemical impedance spectroscopy data of a plurality of to-be-identified batteries during working are acquired, channel separation is carried out on the electrochemical impedance spectroscopy data, and then logarithms of frequency channels are taken to obtain logarithmic frequencies; calculating a complex real part channel and a complex imaginary part channel by using the determined target logarithmic frequency to obtain impedance data, restoring the target logarithmic frequency to a linear frequency to obtain a final frequency, and standardizing the final frequency and the impedance data to obtain standardized frequency data and standardized impedance data. The standardized frequency data and the standardized impedance data are input into a target liquid leakage recognition model obtained through training based on an improved HFC-CNN model for recognition, a recognition result is obtained, whether the battery leaks liquid or not is judged according to the recognition result, and the accuracy of battery liquid leakage recognition is effectively improved through the method.

Description

Technical Field

[0001] The present invention relates to the technical field of battery leakage identification, and in particular to a battery leakage identification method, device, equipment and storage medium. Background Art

[0002] With the widespread application of lithium-ion batteries in electric vehicles, energy storage systems, portable electronic devices, and other fields, the safety and reliability of lithium batteries are receiving increasing attention. Battery cell leakage can easily cause short circuits, leading to more serious safety issues. Therefore, how to effectively identify and predict battery cell leakage events and promptly control them has become a key issue that the lithium battery testing industry urgently needs to address.

[0003] Traditionally, the detection of battery cell leakage mainly relies on methods such as manual visual inspection, weight change measurement or chemical analysis. However, these methods have many shortcomings. Manual visual inspection is not only time-consuming and labor-intensive, but also easily affected by human factors, resulting in low detection accuracy; weight change measurement needs to be performed after the battery cell has been used for a period of time, and early warning cannot be achieved; although chemical analysis methods are accurate, they are complex and costly to operate, making them unsuitable for large-scale application. In recent years, with the development of electrochemical impedance spectroscopy (EIS) technology, people have begun to try to apply EIS technology to the monitoring of battery cell health status. EIS technology can reflect the internal structure and kinetic information of the battery by measuring the impedance response of the battery under AC small signal perturbations, thereby indirectly evaluating the performance status of the battery. However, when using EIS data to identify battery cell leakage, the accuracy of leakage identification is low due to the limitations of feature extraction and analysis methods. Summary of the Invention

[0004] In order to solve the above technical problems, embodiments of the present invention provide a battery leakage identification method, device, equipment and storage medium to solve the technical problem of low accuracy of battery leakage identification in the prior art.

[0005] A first aspect of an embodiment of the present invention provides a battery leakage identification method, the method comprising:

[0006] Obtaining voltage signals and current signals of several batteries to be identified when they are in operation, extracting the voltage signals and current signals to obtain electrochemical impedance spectroscopy data, wherein the electrochemical impedance spectroscopy data includes frequency, real impedance part, and imaginary impedance part;

[0007] Channel separation is performed on the electrochemical impedance spectroscopy data to obtain a frequency channel, a complex real part channel, and a complex imaginary part channel. The logarithm of the frequency channel is taken to obtain a logarithmic frequency. A target logarithmic frequency is determined based on the logarithmic frequency. Based on the target logarithmic frequency, the complex real part channel and the complex imaginary part channel are used to perform calculations to obtain impedance data.

[0008] The target logarithmic frequency is restored to a linear frequency to obtain a final frequency, and the final frequency and impedance data are respectively standardized to obtain standardized frequency data and standardized impedance data;

[0009] The standardized frequency data and the standardized impedance data are input into the target leakage recognition model for recognition to obtain a recognition result, and whether the battery is leaking is determined based on the recognition result. The target leakage recognition model is trained based on the improved HFC-CNN model.

[0010] In a possible implementation of the first aspect, based on the target logarithmic frequency, performing calculations using a complex real channel and a complex imaginary channel to obtain impedance data includes:

[0011] Based on the complex real part channel and the complex imaginary part channel, the corresponding real part interpolation function and imaginary part interpolation function are constructed respectively;

[0012] The target logarithmic frequency is determined according to the logarithmic frequency, and the target logarithmic frequency is input into the real part interpolation function and the imaginary part interpolation function for processing to obtain impedance data, wherein the impedance data includes the real part of impedance and the imaginary part of impedance.

[0013] In a possible implementation of the first aspect, constructing a real part interpolation function and an imaginary part interpolation function based on the complex real part channel and the complex imaginary part channel respectively to obtain corresponding real part interpolation functions includes:

[0014] Based on the complex real part channel, the real part interpolation function is constructed, where the real part interpolation function is:

[0015] S real (F) = a i (FF i ) 3 +b i (FF i ) 2 +c i (FF i )+d i

[0016] Where, F i is the logarithmic frequency value of the left endpoint of the i-th interval, a i is the curvature change rate of the real part interpolation function, b i is the curvature of the real part interpolation function, ci is the slope of the real interpolation function, d i is a constant term, F is the logarithmic frequency value, and represents the independent variable of the function;

[0017] Based on the complex imaginary part channel, the imaginary part interpolation function is constructed, where the imaginary part interpolation function is:

[0018] S imag (F) = a j (FF j ) 3 +b j (FF j ) 2 +c j (FF j )+d j

[0019] Where, F j is the logarithmic frequency value of the left endpoint of the jth interval, a j is the curvature change rate of the imaginary interpolation function, b j is the curvature of the imaginary interpolation function, c j is the slope of the imaginary interpolation function, d j is a constant term, F is the logarithmic frequency value, and represents the independent variable of the function.

[0020] In a possible implementation of the first aspect, inputting the standardized frequency data and the standardized impedance data into a leakage identification model for identification to obtain an identification result includes:

[0021] The standardized frequency data and the standardized impedance data are convolved using a custom complex convolution layer of the leakage recognition model to obtain a first convolution result, a second convolution result, and a third convolution result;

[0022] The pooling layer of the liquid leakage recognition model is used to fuse the first convolution result, the second convolution result, and the third convolution result to obtain a fusion result;

[0023] The fusion result is processed using the fully connected layer of the leakage recognition model to obtain the recognition result.

[0024] In a possible implementation of the first aspect, a custom complex convolution layer of the leakage recognition model is used to convolve the standardized frequency data and the standardized impedance data to obtain a first convolution result, a second convolution result, and a third convolution result, including:

[0025] Divide the input normalized frequency data and normalized impedance data into frequency data channel and complex impedance channel data according to the channel dimension;

[0026] Convolve the frequency data channel to obtain the first convolution result;

[0027] The complex impedance channel data is split to obtain a real component and an imaginary component, and a cross convolution operation is performed on the real component and the imaginary component to obtain a second convolution result and a third convolution result.

[0028] In a possible implementation of the first aspect, the target leakage recognition model is trained based on an improved HFC-CNN model, including:

[0029] Obtaining electrochemical impedance spectroscopy sample data of a plurality of batteries to be identified, and dividing the electrochemical impedance spectroscopy sample data into a training set and a test set according to a preset ratio;

[0030] Preprocess the electrochemical impedance spectroscopy sample data in the training set to obtain standardized sample data;

[0031] An initial leakage recognition model is constructed based on the improved HFC-CNN model. The standardized sample data is input into the initial leakage recognition model for training to obtain a trained initial leakage recognition model.

[0032] The initial leakage recognition model is verified using the electrochemical impedance spectroscopy sample data in the test set. If the verification conditions are met, the target leakage recognition model is obtained.

[0033] In a possible implementation of the first aspect, the initial leakage identification model is verified using electrochemical impedance spectroscopy sample data in a test set. If the verification conditions are met, a target leakage identification model is obtained, including:

[0034] The trained initial leakage recognition model is used to identify the electrochemical impedance spectroscopy sample data in the test set to obtain the initial recognition results. Based on the initial recognition results and the actual results, the recall rate and false positive rate are obtained.

[0035] Based on the recall rate and false positive rate, the Youden index is calculated using the Youden index calculation formula. The Youden index calculation formula is:

[0036] J(τ)=α*TPR+(1-α)*(1-FPR)

[0037] Where TPR is the recall rate, FPR is the false positive rate, α is the weight, J(τ) is the Youden index, and τ is the decision threshold;

[0038] According to the Youden index, a final decision threshold is obtained. If the final decision threshold is greater than or equal to the preset threshold, the trained initial leakage recognition model is determined to be the target leakage recognition model. If the final decision threshold is less than the preset threshold, the initial leakage recognition model is continued to be trained using the electrochemical impedance spectroscopy sample data in the training set until the target leakage recognition model is obtained.

[0039] In order to solve the same technical problem, a second aspect of an embodiment of the present invention provides a battery leakage identification device, including an acquisition module, a separation module, a standardization processing module and an identification module, wherein:

[0040] The acquisition module is used to obtain the voltage signals and current signals of several batteries to be identified when they are working, extract the voltage signals and current signals, and obtain electrochemical impedance spectroscopy data, wherein the electrochemical impedance spectroscopy data includes frequency, real impedance part, and imaginary impedance part;

[0041] The separation module is used to perform channel separation on the electrochemical impedance spectroscopy data to obtain a frequency channel, a complex real channel, and a complex imaginary channel, take the logarithm of the frequency channel to obtain a logarithmic frequency, determine a target logarithmic frequency based on the logarithmic frequency, and calculate based on the target logarithmic frequency using the complex real channel and the complex imaginary channel to obtain impedance data;

[0042] The standardization processing module is used to restore the target logarithmic frequency to a linear frequency to obtain a final frequency, and to standardize the final frequency and impedance data to obtain standardized frequency data and standardized impedance data respectively;

[0043] The recognition module is used to input the standardized frequency data and the standardized impedance data into the target leakage recognition model for recognition, obtain a recognition result, and determine whether the battery is leaking based on the recognition result. The target leakage recognition model is trained based on the improved HFC-CNN model.

[0044] A third aspect of an embodiment of the present invention provides a computer device, including:

[0045] memory for storing computer programs;

[0046] The processor is configured to implement the steps of the battery leakage identification method of the first aspect when executing the computer program.

[0047] A fourth aspect of the embodiments of the present invention provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the battery leakage identification method according to the first aspect are implemented.

[0048] The technical solution of the present invention has the following advantages:

[0049] The battery leakage identification method provided by the embodiment of the present invention obtains electrochemical impedance spectroscopy data of several batteries to be identified during operation, performs channel separation on the electrochemical impedance spectroscopy data to obtain a frequency channel, a complex real channel, and a complex imaginary channel, then takes the logarithm of the frequency channel to obtain a logarithmic frequency, uses the logarithmic frequency to determine a target logarithmic frequency, and then calculates based on the target logarithmic frequency using the complex real channel and the complex imaginary channel to obtain impedance data, restores the target logarithmic frequency to a linear frequency to obtain a final frequency, and standardizes the final frequency and impedance data to obtain standardized frequency data and standardized impedance data, respectively. The electrochemical impedance spectroscopy data is resampled and standardized to make the electrochemical impedance spectroscopy data have the same data length, and then the standardized frequency data and the standardized impedance data are input into a target leakage identification model trained based on the improved HFC-CNN model for identification to obtain an identification result, and determines whether the battery is leaking based on the identification result. The above method effectively improves the accuracy of battery leakage identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific implementation methods or the description of the prior art. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0051] Figure 1 Flowchart of a battery leakage identification method according to an embodiment of the present invention;

[0052] Figure 2 2. This is a structural diagram of a target leakage identification model of a battery leakage identification method according to an embodiment of the present invention;

[0053] Figure 3 This is a graph showing the accuracy of a target leakage recognition model during training of a battery leakage recognition method according to an embodiment of the present invention;

[0054] Figure 4 : is a curve showing a change in loss value during the training process of a target leakage recognition model of a battery leakage recognition method according to an embodiment of the present invention;

[0055] Figure 5 : is the ROC curve of the battery leakage identification method in an embodiment of the present invention;

[0056] Figure 6 1. A comparison chart of recognition results of various model test sets of the battery leakage recognition method according to an embodiment of the present invention;

[0057] Figure 74 is a structural block diagram of a battery leakage identification device in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0059] In the description of the present invention, it should be noted that the terms "first", "second" and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0060] The battery leakage identification method provided by the embodiment of the present invention is as follows: Figure 1 As shown, Figure 1 This is a flow chart of a battery leakage identification method, including steps S101 to S104. The details of each step are as follows:

[0061] S101. Obtain voltage signals and current signals of several batteries to be identified when they are in operation, extract the voltage signals and current signals, and obtain electrochemical impedance spectroscopy data, wherein the electrochemical impedance spectroscopy data includes frequency, real impedance part, and imaginary impedance part.

[0062] In this embodiment, electrochemical impedance spectroscopy (EIS) data of the battery to be identified is collected by an electrochemical workstation. Specifically, the EIS data is obtained by applying an AC disturbance signal of a specific frequency to the battery, measuring its current response signal or voltage response signal, and then calculating it through frequency domain analysis. Since there are many methods to obtain EIS data, how to obtain EIS data based on the current response signal or voltage response signal analysis is not described here in detail.

[0063] S102. Perform channel separation on the electrochemical impedance spectroscopy data to obtain a frequency channel, a complex real channel, and a complex imaginary channel. Take the logarithm of the frequency channel to obtain a logarithmic frequency. Determine a target logarithmic frequency based on the logarithmic frequency. Based on the target logarithmic frequency, perform calculations using the complex real channel and the complex imaginary channel to obtain impedance data.

[0064] In this embodiment, the frequency points of the EIS data collected from different batteries are not necessarily consistent. Therefore, the EIS data needs to be preprocessed. Existing resampling methods are generally targeted at time domain signals and are mostly used for signal processing, which is not suitable for resampling EIS data. To ensure the shape characteristics of the EIS impedance spectrum, a complex frequency domain interpolation resampling method for EIS data is customized: first, the electrochemical impedance spectroscopy data is channel-separated to obtain a frequency channel, a complex real part channel, and a complex imaginary part channel. Then, the frequency channel is logarithmized to obtain the logarithmic frequency.

[0065] Then, a real part interpolation function and an imaginary part interpolation function are constructed respectively. According to the target logarithmic frequency defined by the logarithmic frequency, the target logarithmic frequency is input into the constructed real part interpolation function and the imaginary part interpolation function respectively to obtain impedance data, wherein the impedance data includes the real part of impedance and the imaginary part of impedance.

[0066] In one embodiment, impedance data is obtained by performing calculations based on a target logarithmic frequency using a complex real channel and a complex imaginary channel, including:

[0067] Based on the complex real part channel and the complex imaginary part channel, the corresponding real part interpolation function and imaginary part interpolation function are constructed respectively;

[0068] The target logarithmic frequency is determined according to the logarithmic frequency, and the target logarithmic frequency is input into the real part interpolation function and the imaginary part interpolation function for processing to obtain impedance data, wherein the impedance data includes the real part of impedance and the imaginary part of impedance.

[0069] In this embodiment, a piecewise cubic polynomial is used to construct a real part interpolation function and an imaginary part interpolation function, respectively, and the required target logarithmic frequency points, such as 26 frequency points, are defined. The target logarithmic frequency points are input into the real part interpolation function and the imaginary part interpolation function, respectively, to obtain impedance data.

[0070] The above steps can eliminate differences in frequency points and data length between different devices or different test batches. For example, some data may have 26 points between 0.01 and 1000 Hz, some between 0.01 and 500 Hz, and some between 0.01 and 2000 Hz. The data length and frequency are inconsistent. After resampling, they are all unified to 26 points between 0.01 and 1000 Hz. This ensures consistent data length when input into the target leakage identification model.

[0071] In one embodiment, the corresponding real part interpolation function and imaginary part interpolation function are constructed based on the complex real part channel and the complex imaginary part channel, respectively, including:

[0072] Based on the complex real part channel, the real part interpolation function is constructed, where the real part interpolation function is:

[0073] S real (F) = a i (FF i ) 3 +b i (FF i ) 2 +c i (FF i )+d i

[0074] Where, F i is the logarithmic frequency value of the left endpoint of the i-th interval, a i is the curvature change rate of the real part interpolation function, b i is the curvature of the real part interpolation function, c i is the slope of the real interpolation function, d i is a constant term, F is the logarithmic frequency value, and represents the independent variable of the function;

[0075] Based on the complex imaginary part channel, the imaginary part interpolation function is constructed, where the imaginary part interpolation function is:

[0076] S imag (F) = a j (FF j ) 3 +b j (FF j ) 2 +c j (FF j )+d j

[0077] Where, F j is the logarithmic frequency value of the left endpoint of the jth interval, a j is the curvature change rate of the imaginary interpolation function, b j is the curvature of the imaginary interpolation function, c j is the slope of the imaginary interpolation function, d j is a constant term, F is the logarithmic frequency value, and represents the independent variable of the function.

[0078] In this embodiment, when using piecewise cubic polynomials to construct real part interpolation functions and imaginary part interpolation functions respectively, both can be implemented through the interpolation function of the scipy library. Specifically, based on the real impedance data in the complex imaginary part channel, a smooth real part interpolation function is constructed. The real part interpolation function is:

[0079] S real (F) = a i (FF i ) 3 +b i (FF i )2 +c i (FF i )+d i

[0080] Where, F i is the logarithmic frequency value of the left endpoint of the i-th interval, a i is the curvature change rate of the real part interpolation function, b i is the curvature of the real part interpolation function, c i is the slope of the real interpolation function, d i is a constant term, F is the logarithmic frequency value, and represents the independent variable of the function;

[0081] Based on the complex imaginary part channel, the imaginary part interpolation function is constructed, where the imaginary part interpolation function is:

[0082] S imag (F) = a j (FF j ) 3 +b j (FF j ) 2 +c j (FF j )+d j

[0083] Where, F j is the logarithmic frequency value of the left endpoint of the jth interval, a j is the curvature change rate of the imaginary interpolation function, b j is the curvature of the imaginary interpolation function, c j is the slope of the imaginary interpolation function, d j is a constant term, F is the logarithmic frequency value, and represents the independent variable of the function.

[0084] It should be noted that the coefficient a i 、b i 、c i d i 、a j 、b j 、c j and d j By solving the three-moment equations, it is found that in this piecewise interpolation function, the value of the coefficient corresponding to each interval is different.

[0085] S103 , restoring the target logarithmic frequency to a linear frequency to obtain a final frequency, and standardizing the final frequency and impedance data to obtain standardized frequency data and standardized impedance data.

[0086] In this embodiment, the target logarithmic frequency is restored to a linear frequency to obtain a final frequency, and the final frequency and impedance data are respectively standardized to obtain standardized frequency data and standardized impedance data. Specifically, the target logarithmic frequency and impedance data are respectively subjected to Z-score standardization to eliminate the differences between different data scales, and obtain standardized frequency data and standardized impedance data, wherein the impedance data includes the real part of the impedance and the imaginary part of the impedance. The Z-score standardization formula is:

[0087]

[0088] Where z is the transformed sequence, x is the original sequence, μ is the sample mean, and σ is the standard deviation of the sample.

[0089] S104. Input the standardized frequency data and the standardized impedance data into a target leakage recognition model for recognition to obtain a recognition result, and determine whether the battery is leaking based on the recognition result. The target leakage recognition model is trained based on an improved HFC-CNN model.

[0090] In this embodiment, an initial leakage recognition model is constructed based on the improved HFC-CNN model, and the initial leakage recognition model is trained to obtain a target leakage recognition model. Figure 2 As shown in the figure, the network layer structure of the target leakage recognition model includes complex convolutional layers, pooling layers, and fully connected layers. Compared with other deep learning models, convolutional neural networks are more lightweight due to their parameter sharing characteristics. CNNs also have strong local feature extraction capabilities and can effectively capture local features in sequence data. CNNs are also translationally invariant, meaning they are invariant to small translations of sequence data, which can cope with data translation caused by wiring harness contact resistance. In addition, custom complex convolutional layers can specifically process data in the complex frequency domain, preserving its complex characteristics.

[0091] Then input the standardized frequency data and the standardized impedance data into the target leakage recognition model for recognition, and obtain the recognition result. According to the recognition result, it is judged whether the battery is leaking.

[0092] In one embodiment, the normalized frequency data and the normalized impedance data are input into a leakage identification model for identification to obtain an identification result, including:

[0093] The standardized frequency data and the standardized impedance data are convolved using a custom complex convolution layer of the leakage recognition model to obtain a first convolution result, a second convolution result, and a third convolution result;

[0094] The pooling layer of the liquid leakage recognition model is used to fuse the first convolution result, the second convolution result, and the third convolution result to obtain a fusion result;

[0095] The fusion result is processed using the fully connected layer of the leakage recognition model to obtain the recognition result.

[0096] In this embodiment, after the standardized frequency data and the standardized impedance data are input into the leakage recognition model, a custom complex convolution layer is used to perform convolution operations on the input data to obtain a first convolution result, a second convolution result, and a third convolution result.

[0097] It should be noted that the first convolution result is the frequency channel convolution result, the second convolution result is the real part convolution result, and the third convolution result is the imaginary part convolution result.

[0098] Then, in the pooling layer, the first convolution result, the second convolution result, and the third convolution result are feature fused to obtain the fusion result. The pooling layer is the global maximum pooling, and the output data dimension is (batch_size, 90). The fusion result is:

[0099] O=[F|C real |C imag ]+b∈R T×3F

[0100] Where F is the first convolution result, C real is the second convolution result, C imag is the result of the third convolution, and b is the bias term.

[0101] The fusion result is processed using the fully connected layer and the recognition result is output. The output dimension of the fully connected layer is None, the data length remains unchanged, and the number of channels = the number of frequency convolution kernels + the number of impedance convolution kernels, where

[0102] The recognition results are:

[0103] Y=σ(O)∈R T×3F

[0104] Where Y is the recognition result, σ is the activation function, and O is the fusion result.

[0105] In one embodiment, a custom complex convolution layer of a leakage recognition model is used to convolve the normalized frequency data and the normalized impedance data to obtain a first convolution result, a second convolution result, and a third convolution result, including:

[0106] Divide the input normalized frequency data and normalized impedance data into frequency data channel and complex impedance channel data according to the channel dimension;

[0107] Convolve the frequency data channel to obtain the first convolution result;

[0108] The complex impedance channel data is split to obtain a real component and an imaginary component, and a cross convolution operation is performed on the real component and the imaginary component to obtain a second convolution result and a third convolution result.

[0109] In this embodiment, when the custom complex convolution layer performs convolution operations, the core idea of ​​convolution is: the frequency channel is convolved separately, and the impedance channel is cross-convolved with the real and imaginary parts. Specifically, first, the input normalized frequency data and normalized impedance data are split into frequency data channels and complex impedance data channels according to the channel dimension, with the first 1 / 3 being the frequency data channel and the last 2 / 3 being the complex impedance data channel, such as:

[0110] X f =X[:,:,0:K],X c =X[:,:,K:3K]

[0111] Where, X f is the frequency data channel, X c is the complex impedance data channel.

[0112] Convolve the frequency data channel to obtain the first convolution result, where the convolution process is:

[0113]

[0114] Where F is the result of the first convolution, X f is the frequency data channel, W f is the weight matrix.

[0115] When performing convolution on a complex impedance data channel, define the complex component splitting:

[0116] X c =[X r ||X i ],X r ,X i ∈R T×K

[0117] Where, X r is the real part, X i is the imaginary component.

[0118] Real and imaginary cross convolution operation:

[0119]

[0120] C real =R r -I i ,C imag =I r -R i

[0121] Where R r 、R i is the real part of the output, I r , I i is the imaginary part of the output, X r is the real part, X i is the imaginary component, is the real convolution kernel, which contains K different filters. is the real convolution kernel, containing K+1 to 2K different filters, is the imaginary convolution kernel, which contains K different filters. is the imaginary convolution kernel, which contains K+1 to 2K different filters, C real is the second convolution result, C imag is the result of the third convolution.

[0122] In one embodiment, the target leakage recognition model is trained based on an improved HFC-CNN model, including:

[0123] Obtaining electrochemical impedance spectroscopy sample data of a plurality of batteries to be identified, and dividing the electrochemical impedance spectroscopy sample data into a training set and a test set according to a preset ratio;

[0124] Preprocess the electrochemical impedance spectroscopy sample data in the training set to obtain standardized sample data;

[0125] An initial leakage recognition model is constructed based on the improved HFC-CNN model. The standardized sample data is input into the initial leakage recognition model for training to obtain a trained initial leakage recognition model.

[0126] The initial leakage recognition model is verified using the electrochemical impedance spectroscopy sample data in the test set. If the verification conditions are met, the target leakage recognition model is obtained.

[0127] In this embodiment, when collecting electrochemical impedance spectroscopy sample data, the hardware equipment is: an electrochemical workstation, and the frequency range is recommended to cover 0.01Hz~100kHz. The specific frequency range depends on the specific situation of the battery cell. Special attention should be paid to ensure stable electrode contact during testing and avoid contact group anti-interference. The environment is: constant temperature (25±1℃), low humidity (<30%RH), or according to actual conditions. The sample size is: batteries with known status (leakage / normal), and the amount of electrochemical impedance spectroscopy sample data collected is ≥500 groups. The positive and negative sample sizes need to be balanced to cover different leakage levels and different SOC situations as much as possible. The EIS test data includes impedance information at different frequencies, and the data is generally saved as frequency, real impedance part and imaginary impedance part. The electrochemical impedance spectroscopy sample data is then divided into training set and test set according to a preset ratio, such as the training set accounts for 80% of the total number and the test set accounts for 20% of the total number.

[0128] The electrochemical impedance spectroscopy sample data in the training set is preprocessed to obtain standardized sample data. The preprocessing method is resampling and standardization. The specific preprocessing process has been described in S102 to S103 and will not be repeated here. The electrochemical impedance spectroscopy data before preprocessing is shown in Table 1. Table 1 lists the frequency channel data of some data. The differences are mainly reflected in the value of the frequency point and the length of the data. If it is to be input into the target leakage identification model, the data dimension needs to be unified. Therefore, the interpolation resampling method is adopted to unify all data into the frequency points in the last column of the table, and the impedance values ​​corresponding to the frequency points are calculated by the defined interpolation function. After preprocessing, the data is effectively unified.

[0129] Table 1 Electrochemical impedance spectroscopy data before pretreatment

[0130] Serial number freq. / Hz freq. / Hz freq. / Hz 1 2000.0000 2 1273.0000 3 809.8000 1000.0000 4 515.3000 631.0000 5 500.0000 327.9000 398.1000 6 312.4000 208.6000 251.2000 7 195.1000 132.7000 158.5000 8 121.9000 84.4700 100.0000 9 76.1700 53.7500 63.1000 10 47.5800 34.2000 39.8100 11 29.7300 21.7600 25.1200 12 18.5700 13.8500 15.8500 13 11.6000 8.8110 10.0000 14 7.2480 5.6060 6.3100 15 4.5280 3.5670 3.9810 16 2.8290 2.2700 2.5120 17 1.7670 1.4440 1.5850 18 1.1040 0.9191 1.0000 19 0.6898 0.5848 0.6310 20 0.4309 0.3721 0.3981 21 0.2692 0.2368 0.2512 22 0.1682 0.1507 0.1585 23 0.1051 0.0959 0.1000 24 0.0657 0.0610 0.0631 25 0.0410 0.0388 0.0398 26 0.0256 0.0247 0.0251 27 0.0160 0.0157 0.0159 28 0.0100 0.0100 0.0100

[0131] The core idea of ​​complex domain interpolation resampling is to first take the logarithm of the frequency channel, then perform cubic spline interpolation on the complex impedance in the EIS data, and finally resample the data at specified frequency points.

[0132] Then, an initial leakage recognition model is constructed based on the improved HFC-CNN model. The input data dimension of the model input layer is: batch_size = 26, and the feature dimension of each sample is 3; the parameters of the first convolutional layer are set as: the number of convolution kernels is 30, of which 1 / 3 is the number of frequency channel convolution kernels and 2 / 3 is the number of complex impedance channel convolution kernels; the convolution kernel size is 3, the activation function is Rule, padding = 'same', the output dimension is: batch_size = 13, and the feature dimension of each data is 30.

[0133] The parameters of the first pooling layer are set to: global maximum pooling, output data dimension (batch_size, 90)

[0134] The parameters of the second convolutional layer are set as follows: the number of convolution kernels is 90, of which 1 / 3 is the number of frequency channel convolution kernels, 2 / 3 is the number of complex impedance channel convolution kernels, the convolution kernel size is 3, the activation function is Rule, padding = 'same', the output dimension is: batch_size = 13, and the feature dimension of each data is 90.

[0135] The parameters of the second pooling layer are set to: global maximum pooling, output data dimension (batch_size, 90)

[0136] Dropout layer: Randomly discard the layer without changing the data dimension

[0137] The fully connected layer has 32 neurons, 32 output dimensions, and the activation function is Rule.

[0138] The output layer is: a fully connected layer with 1 neuron, a sigmoid activation function, and outputs the binary classification probability of the sample.

[0139] The standardized sample data is input into the initial leakage recognition model for training to obtain a trained initial leakage recognition model. The training parameters are set as follows: the cross entropy loss function is used for training, and the formula is as follows:

[0140]

[0141] Where N is the number of electrochemical impedance spectroscopy sample data in the training set, y i is the recognition result of the i-th sample data, p i is the predicted probability of the i-th sample data.

[0142] The model was trained using hyperparameter optimization and early stopping strategies. The hyperparameter training parameters were as follows: 500 training rounds, 64 batches, 3 convolution kernels, 2 pooling kernels, and an initial learning rate of 0.001, which decayed by 5% every 100 rounds and can be adjusted based on actual results. An early stopping strategy was also used, with patience = 50. If the loss of the test set did not decrease after 50 consecutive rounds, early stopping was considered, and a dropout layer was added with dropout = 0.2. Figure 3 and Figure 4 As shown, Figure 3 is the accuracy curve of the model training process, Figure 4 The loss value variation curve is shown in Figure 2. The initial leakage identification model is then verified using the electrochemical impedance spectroscopy sample data in the test set. If the verification conditions are met, the target leakage identification model is obtained.

[0143] In one embodiment, the initial leakage identification model is verified using electrochemical impedance spectroscopy sample data in a test set. If the verification conditions are met, a target leakage identification model is obtained, including:

[0144] The trained initial leakage recognition model is used to identify the electrochemical impedance spectroscopy sample data in the test set to obtain the initial recognition results. Based on the initial recognition results and the actual results, the recall rate and false positive rate are obtained.

[0145] Based on the recall rate and false positive rate, the Youden index is calculated using the Youden index calculation formula. The Youden index calculation formula is:

[0146] J(τ)=α*TPR+(1-α)*(1-FPR)

[0147] Where TPR is the recall rate, FPR is the false positive rate, α is the weight, J(τ) is the Youden index, and τ is the decision threshold;

[0148] According to the Youden index, the final decision threshold is obtained. If the final decision threshold is greater than or equal to the preset threshold, the trained initial leakage recognition model is determined to be the target leakage recognition model. If the final decision threshold is less than the preset threshold, the initial leakage recognition model is continued to be trained using the electrochemical impedance spectroscopy sample data in the training set until the target leakage recognition model is obtained.

[0149] In this embodiment, the electrochemical impedance spectroscopy sample data (hereinafter referred to as "sample data") in the test set is input into the trained initial leakage recognition model for recognition to obtain an initial recognition result. Based on the initial recognition result and the true result, the recall rate and false positive rate are obtained. The recall rate is calculated as follows:

[0150]

[0151] Where TPR is the recall rate, TP is the sample data that actually has leakage and is correctly identified as having leakage, and FN is the sample data that actually has leakage and is correctly identified as having leakage.

[0152] The formula for calculating the false positive rate is:

[0153]

[0154] Where FP is the sample data that is actually not leaking but is mistakenly judged to have leaked liquid, and TN is the sample data that is actually leaking but is mistakenly judged to have not leaked liquid.

[0155] The Youden index corresponding to different thresholds (τ) is calculated using the recall rate and false positive rate. The Youden index focuses on both the detection rate and the false positive rate. In order to make the target leakage recognition model pay more attention to the detection rate, the calculation formula of the Youden index is improved. As an evaluation indicator of the target leakage recognition model, the final decision threshold is obtained to represent the comprehensive score of the model. The Youden index calculation formula is:

[0156] J(τ)=α*TPR+(1-α)*(1-FPR)

[0157] Where TPR is the recall rate, FPR is the false positive rate, α is the weight, J(τ) is the Youden index, and τ is the threshold;

[0158] It should be noted that when α is set to 0.8 (focusing more on the detection rate), the model calculation result is a probability value ranging from 0 to 1. The default threshold (0.5) is generally used as the cutoff point, but here the decision threshold τ corresponding to the maximum value of the Youden index J(τ) is used as the final decision threshold, as shown in Figure 5 As shown, Figure 5 It is the ROC curve-optimal threshold diagram.

[0159] The corresponding final decision threshold is determined according to the Youden index. Based on the final decision threshold, the initial leakage recognition model is adjusted and retrained until the comprehensive score reaches more than 95%, that is, the detection rate is greater than 95% and the false detection rate is less than 10%.

[0160] It should be noted that specific indicators can be adjusted according to actual needs. After completing parameter adjustment and retraining, it is of great significance to sort out the parameter amount and average accuracy data of different models (Table 2). The parameter amount directly affects the expressiveness and computational cost of the model. Generally, models with large parameter amounts perform well in complex tasks, but they also face a higher risk of overfitting. The average accuracy intuitively reflects the overall performance of the model in multi-category tasks. The recognition results of the test sets of each model are compared. Figure 6 shown.

[0161] Table 2 Comparison of parameters and average accuracy of different models

[0162]

[0163]

[0164] Comparative analysis reveals that the target leakage recognition model provided by this invention, HFC-CNN-1, has only 10% to 20% of the parameters of other models. Other metrics, such as accuracy, are similar to those of other models, but the number of parameters is significantly reduced. This indicates that this application uses a custom complex convolutional neural network layer as the primary layer of the model architecture. Due to the parameter-sharing nature of convolutional neural networks, the model is more lightweight than other deep learning networks. Furthermore, this solution eliminates the need for complex feature engineering and directly utilizes convolutional neural networks to extract features from the original sequence for recognition.

[0165] The battery leakage identification device provided by the embodiment of the present invention is as follows: Figure 7 As shown, Figure 7 The device block diagram of the battery leakage identification device 700 includes an acquisition module 701, a separation module 702, a standardization processing module 703 and an identification module 704, wherein:

[0166] The acquisition module 701 is used to obtain voltage signals and current signals of several batteries to be identified when they are in operation, extract the voltage signals and current signals, and obtain electrochemical impedance spectroscopy data, wherein the electrochemical impedance spectroscopy data includes frequency, real impedance part, and imaginary impedance part;

[0167] The separation module 702 is used to perform channel separation on the electrochemical impedance spectroscopy data to obtain a frequency channel, a complex real channel, and a complex imaginary channel, take the logarithm of the frequency channel to obtain a logarithmic frequency, determine a target logarithmic frequency based on the logarithmic frequency, and perform calculations based on the target logarithmic frequency using the complex real channel and the complex imaginary channel to obtain impedance data;

[0168] The standardization processing module 703 is used to restore the target logarithmic frequency to a linear frequency to obtain a final frequency, and to standardize the final frequency and impedance data to obtain standardized frequency data and standardized impedance data respectively;

[0169] The identification module 704 is used to input the standardized frequency data and the standardized impedance data into the target leakage identification model for identification, obtain an identification result, and determine whether the battery is leaking based on the identification result. The target leakage identification model is trained based on the improved HFC-CNN model.

[0170] The specific implementation of the battery leakage identification device is basically the same as the specific embodiment of the above-mentioned battery leakage identification method, and will not be repeated here.

[0171] In one embodiment of the present application, a computer device is provided, which includes a memory and a processor, wherein a computer program is stored in the memory, and the above steps are implemented when the processor executes the computer program; the computer device provided in this embodiment has an implementation principle and technical effects similar to those of the above method embodiment, and will not be repeated here.

[0172] In one embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and the above steps are implemented when the computer program is executed by a processor; the computer-readable storage medium provided in this embodiment has an implementation principle and technical effects similar to those of the above method embodiment, and will not be repeated here.

[0173] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0174] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A battery leakage identification method, characterized in that: include: Obtaining voltage signals and current signals of a plurality of batteries to be identified when they are in operation, extracting the voltage signals and the current signals to obtain electrochemical impedance spectroscopy data, wherein the electrochemical impedance spectroscopy data includes frequency, real impedance part, and imaginary impedance part; performing channel separation on the electrochemical impedance spectroscopy data to obtain a frequency channel, a complex real channel, and a complex imaginary channel, taking a logarithm of the frequency channel to obtain a logarithmic frequency, determining a target logarithmic frequency based on the logarithmic frequency, and performing calculations based on the target logarithmic frequency using the complex real channel and the complex imaginary channel to obtain impedance data; Restoring the target logarithmic frequency to a linear frequency to obtain a final frequency, and standardizing the final frequency and the impedance data to obtain standardized frequency data and standardized impedance data; The standardized frequency data and the standardized impedance data are input into a target leakage recognition model for recognition to obtain a recognition result, and whether the battery is leaking is determined based on the recognition result, wherein the target leakage recognition model is trained based on an improved HFC-CNN model.

2. The battery leakage identification method according to claim 1, wherein: The step of calculating based on the target logarithmic frequency using a complex real channel and a complex imaginary channel to obtain impedance data includes: Based on the complex real part channel and the complex imaginary part channel, respectively, constructing to obtain corresponding real part interpolation function and imaginary part interpolation function; A target logarithmic frequency is determined according to the logarithmic frequency, and the target logarithmic frequency is input into the real part interpolation function and the imaginary part interpolation function for processing to obtain impedance data, wherein the impedance data includes a real part of impedance and an imaginary part of impedance.

3. The battery leakage identification method according to claim 2, wherein: The constructing is performed based on the complex real part channel and the complex imaginary part channel respectively to obtain corresponding real part interpolation function and imaginary part interpolation function, including: Based on the complex real part channel, a real part interpolation function is constructed, wherein the real part interpolation function is: S real (F)=a i (F-F i ) 3 +b i (F-F i ) 2 +c i (F-F i )+d i Where, F i is the logarithmic frequency value of the left endpoint of the i-th interval, a i is the curvature change rate of the real part interpolation function, b i is the curvature of the real part interpolation function, c i is the slope of the real interpolation function, d i is a constant term, F is the logarithmic frequency value, and represents the independent variable of the function; Based on the complex imaginary part channel, an imaginary part interpolation function is constructed, wherein the imaginary part interpolation function is: S imag (F)=a j (F-F j ) 3 +b j (F-F j ) 2 +c j (F-F j )+d j Where, F j is the logarithmic frequency value of the left endpoint of the jth interval, a j is the curvature change rate of the imaginary interpolation function, b j is the curvature of the imaginary interpolation function, c j is the slope of the imaginary interpolation function, d j is a constant term, F is the logarithmic frequency value, and represents the independent variable of the function.

4. The battery leakage identification method according to claim 1, wherein: The step of inputting the standardized frequency data and the standardized impedance data into a leakage identification model for identification to obtain an identification result includes: Convolving the standardized frequency data and the standardized impedance data using a custom complex convolution layer of the leakage recognition model to obtain a first convolution result, a second convolution result, and a third convolution result; Using the pooling layer of the liquid leakage recognition model, the first convolution result, the second convolution result, and the third convolution result are subjected to feature fusion to obtain a fusion result; The fusion result is processed using the fully connected layer of the leakage recognition model to obtain a recognition result.

5. The battery leakage identification method according to claim 4, wherein: The user-defined complex convolution layer of the leakage recognition model is used to convolve the standardized frequency data and the standardized impedance data to obtain a first convolution result, a second convolution result, and a third convolution result, including: Dividing the input normalized frequency data and the input normalized impedance data into frequency data channels and complex impedance channel data according to channel dimensions; Performing convolution on the frequency data channel to obtain a first convolution result; The complex impedance channel data is split to obtain a real component and an imaginary component, and a cross convolution operation is performed on the real component and the imaginary component to obtain a second convolution result and a third convolution result.

6. The battery leakage identification method according to claim 1, wherein: The target leakage recognition model is trained based on the improved HFC-CNN model, including: Obtaining electrochemical impedance spectroscopy sample data of a plurality of batteries to be identified, and dividing the electrochemical impedance spectroscopy sample data into a training set and a test set according to a preset ratio; Preprocessing the electrochemical impedance spectroscopy sample data in the training set to obtain standardized sample data; Building an initial liquid leakage recognition model based on the improved HFC-CNN model, and using the standardized sample data to input the initial liquid leakage recognition model for training to obtain a trained initial liquid leakage recognition model; The initial leakage identification model is verified using the electrochemical impedance spectroscopy sample data in the test set, and if the verification conditions are met, a target leakage identification model is obtained.

7. The battery leakage identification method according to claim 6, wherein: The initial leakage identification model is verified using the electrochemical impedance spectroscopy sample data in the test set, and if the verification conditions are met, a target leakage identification model is obtained, including: Using the trained initial leakage recognition model to recognize the electrochemical impedance spectroscopy sample data in the test set to obtain an initial recognition result, and obtaining a recall rate and a false positive rate based on the initial recognition result and the true result; Based on the recall rate and the false positive rate, the Youden index is calculated using the Youden index calculation formula, and the Youden index calculation formula is: J(τ)=α*TPR+(1-α)*(1-FPR) Where TPR is the recall rate, FPR is the false positive rate, α is the weight, J(τ) is the Youden index, and τ is the decision threshold; A final decision threshold is obtained based on the Youden index. If the final decision threshold is greater than or equal to a preset threshold, the trained initial leakage recognition model is determined to be the target leakage recognition model. If the final decision threshold is less than the preset threshold, the initial leakage recognition model is continued to be trained using the electrochemical impedance spectroscopy sample data in the training set until the target leakage recognition model is obtained.

8. A battery leakage identification device, characterized in that: It includes an acquisition module, a separation module, a standardization processing module and an identification module, among which, The acquisition module is used to acquire voltage signals and current signals of a plurality of batteries to be identified when they are in operation, extract the voltage signals and the current signals, and obtain electrochemical impedance spectroscopy data, wherein the electrochemical impedance spectroscopy data includes frequency, real impedance part, and imaginary impedance part; The separation module is used to perform channel separation on the electrochemical impedance spectroscopy data to obtain a frequency channel, a complex real part channel, and a complex imaginary part channel, take the logarithm of the frequency channel to obtain a logarithmic frequency, determine a target logarithmic frequency based on the logarithmic frequency, and perform calculations based on the target logarithmic frequency using the complex real part channel and the complex imaginary part channel to obtain impedance data; The standardization processing module is used to restore the target logarithmic frequency to a linear frequency to obtain a final frequency, and to standardize the final frequency and the impedance data to obtain standardized frequency data and standardized impedance data respectively; The recognition module is used to input the standardized frequency data and the standardized impedance data into a target leakage recognition model for recognition, obtain a recognition result, and determine whether the battery is leaking based on the recognition result, wherein the target leakage recognition model is trained based on an improved HFC-CNN model.

9. A computer device, characterized in that: include: memory for storing computer programs; A processor, configured to implement the battery leakage identification method according to any one of claims 1 to 7 when executing the computer program.

10. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of the battery leakage identification method according to any one of claims 1 to 7.

Images