Battery fault diagnosis method and device, electronic equipment and storage medium

Abstract

The invention provides a battery fault diagnosis method, and belongs to the technical field of battery management, a battery fault diagnosis model is constructed, and the battery fault diagnosis model comprises at least two feature extraction modules and a classification layer; the feature extraction module comprises a convolutional layer, an index linear unit and a batch standardization module which are connected in sequence; performing feature extraction on the operation data of the target battery through a convolutional layer to obtain operation features; if the operation feature is a negative value feature, performing exponential transformation through an exponential linear unit to obtain a target feature; if the operation feature is not the negative value feature, performing linear transformation through an index linear unit to obtain a target feature; and based on the target features, performing fault diagnosis on the target battery through the classification layer. The method is suitable for battery operation data possibly having negative values, negative feature information can be effectively reserved, and the detection precision of fault types such as reverse charging is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to a battery fault diagnosis method and device, electronic equipment and storage medium, belonging to the technical field of battery management. BACKGROUND

[0002] Battery failure can cause device performance degradation, and even cause safety accidents such as fire and explosion, so fault diagnosis of the battery is the key to ensure safe operation of the device.

[0003] In related technologies, the model-based diagnosis method can realize real-time monitoring and fault diagnosis of the battery state by modeling the dynamic characteristics of the battery. This method usually selects certain model parameters as indicators to measure battery abnormalities, and has high model accuracy and diagnosis accuracy.

[0004] For example, the Chinese invention patent with publication number CN119827998A discloses a battery fault detection method, device, computer equipment and storage medium method, obtains the target battery parameters of the battery pack to be detected within the historical period and the actual battery voltage of each single battery in the battery pack to be detected; and input the target battery parameters into the voltage prediction model to obtain the predicted average battery voltage; and then according to the difference between the predicted average battery voltage and the actual battery voltage of each single battery, determine the fault detection result of each single battery. On the one hand, more dimensional features can be obtained, making the extracted battery features more accurate and improving the battery fault detection result; on the other hand, the battery fault detection result can be determined more simply and efficiently. However, the model-based diagnosis method needs to accurately model the battery monomer, and for a battery system composed of thousands of battery monomers, the model error may be seriously deviated from the actual operation. Therefore, a more accurate battery fault diagnosis method is needed. SUMMARY

[0005] In order to solve the problems existing in the prior art, the present application provides a battery fault diagnosis method, device, electronic equipment and storage medium.

[0006] The technical solution of the present application is as follows: The first aspect of the present application provides a battery fault diagnosis method, the method comprising: obtaining the running data of the target battery; constructing a battery fault diagnosis model, the battery fault diagnosis model comprising at least two feature extraction modules and a classification layer; the feature extraction modules are alternately connected through a residual network, and each feature extraction module comprises a convolution layer, an exponential linear unit and a batch normalization module connected in sequence; The running data is feature extracted through the convolution layer to obtain a running feature; if the running feature is a negative value feature, the exponential linear unit is used for exponential transformation to obtain a target feature; if the running feature is not a negative value feature, the exponential linear unit is used for linear transformation to obtain the target feature; the target feature is normalized in batches through the batch normalization module to obtain a final target feature; Based on the final target feature, the target battery is diagnosed for faults through the classification layer to obtain a diagnosis result.

[0007] Preferably, the target feature is normalized in batches through the batch normalization module to obtain a final target feature, specifically: The mean and variance of the current batch of target features are calculated, the target feature is normalized based on the mean and variance to obtain a normalized target feature, the normalized target feature is input into a next feature extraction module until the processing of all feature extraction modules is completed, and the normalized target feature output by the last feature extraction module is obtained as the final target feature.

[0008] Preferably, the method further comprises: Obtaining battery historical running data as sample data; inputting the sample data into an initial battery fault diagnosis model, and extracting features layer by layer through the residual network alternately connected feature extraction modules; performing batch normalization processing on the features output by each layer, and inputting the normalized features into the exponential linear unit for nonlinear transformation; calculating a prediction result according to the nonlinearly transformed features, and optimizing the battery fault diagnosis model parameters by combining the prediction result with a loss function to obtain a trained battery fault diagnosis model.

[0009] Preferably, the running data comprises circulating current voltage data; the running data of the target battery comprises: Obtaining a single battery voltage of the target battery, and calculating a single battery voltage difference according to the single battery voltage; Obtaining a single battery internal resistance of the target battery, and calculating a parallel circulating current voltage difference according to the single battery internal resistance and the single battery voltage difference; Exponentially amplifying the parallel circulating current voltage difference to obtain the circulating current voltage data.

[0010] Preferably, the method further comprises: Obtaining sample data, and counting the number of different fault category labels in the sample data to obtain a first number; Calculating the weight of the fault category label based on the inverse of the first number to obtain a category weight; Obtaining a prediction result corresponding to the sample data through an initial battery fault diagnosis model, performing weighted calculation based on the prediction result, the fault category label and the category weight to obtain a loss value; Optimizing hyperparameters of the initial battery fault diagnosis model based on the loss value to obtain a trained battery fault diagnosis model.

[0011] Preferably, after calculating the weight of the fault category label based on the first number of opposites to obtain the category weight, the method further comprises: Obtaining a preset sliding window, calculating the number of fault category labels in the sliding window to obtain a second number; Calculating the ratio of the first number and the second number to obtain a first ratio; Updating the category weight according to the first ratio to obtain an updated category weight.

[0012] Preferably, the running data further comprises monomer battery voltage data, monomer battery current data and monomer battery temperature data.

[0013] The second aspect of the application provides a battery fault diagnosis device, comprising: A running data acquisition module configured to acquire running data of a target battery; A fault diagnosis module internally configured with a battery fault diagnosis model, wherein the battery fault diagnosis model comprises at least two feature extraction modules and one classification layer; the feature extraction modules are alternately connected through a residual network; each feature extraction module comprises a convolution layer, an exponential linear unit and a batch normalization module connected in sequence, wherein: The running data is subjected to feature extraction through the convolution layer to obtain running features; if the running features are negative features, the running features are subjected to exponential transformation through the exponential linear unit to obtain target features; if the running features are not negative features, the running features are subjected to linear transformation through the exponential linear unit to obtain target features; and the target features are subjected to batch normalization processing through the batch normalization module to obtain final target features; The target battery is subjected to fault diagnosis through the classification layer based on the final target features to obtain a diagnosis result.

[0014] The third aspect of the application provides a battery fault diagnosis device, comprising a memory and at least one processor, wherein the memory is stored with instructions; the at least one processor invokes the instructions in the memory to enable the battery fault diagnosis device to perform the above battery fault diagnosis method.

[0015] The fourth aspect of the present application provides a computer readable storage medium, and the computer readable storage medium stores instructions, when the instructions are run on a computer, the computer executes the battery fault diagnosis method described above.

[0016] The present application has the following beneficial effects: 1. The present application provides a battery fault diagnosis method, device, electronic equipment and storage medium, by constructing a residual module composed of a convolution layer and an exponential linear unit (ELU), and using a skip connection to realize cross-layer feature fusion, the training stability of the deep network is significantly improved, wherein the skip connection allows the original input signal to bypass the nonlinear transformation and directly pass to the deep layer, forming an identity mapping branch, effectively suppressing the gradient dissipation problem, and ensuring the effective return of the gradient in the back propagation process; by stacking multiple residual modules, the model can extract multi-scale abstract features (such as local voltage mutation, global temperature trend, etc.) step by step, while retaining the bottom layer fine-grained information, avoiding feature degradation caused by nonlinear superposition; and because the gradient propagation is improved, the design supports the construction of an ultra-deep network, which captures the complex nonlinear patterns (such as electrochemical noise, time-dependent characteristics) of battery operation data through hierarchical capture, significantly improving the learning ability of the fault classification boundary; 2. The present application provides a battery fault diagnosis method, device, electronic equipment and storage medium, the battery fault diagnosis model constructed introduces a batch normalization module at the end of the feature extraction module, the batch normalization module keeps the input distribution of each layer of the network stable by reducing the internal covariate shift, this feature accelerates the training process of the model, at the same time, the output value of the network is constrained within a reasonable range, effectively reducing the fluctuation in the model training process, and enhancing the stability of the model. In addition, batch normalization promotes the model to learn more robust feature representation, enhances the adaptability of the model to different data sets and environments, and significantly improves the generalization ability of the model; 3. The present application provides a battery fault diagnosis method, device, electronic equipment and storage medium, the exponential linear unit ELU generates a smooth response through an exponential function in the negative interval, avoiding the neuron death problem caused by the hard zero truncation of ReLU, and is especially suitable for feature expression of low-frequency weak signals in battery failure; and ELU has continuous derivability in the global range, and its negative zone unsaturation ensures the stability of the gradient amplitude in the back propagation process, and alleviates the training shock of the deep network; by adjusting the hyperparameters of ELU control the negative value decay rate, which can realize dynamic trade-off between feature sparsity and information integrity, and adapt to the feature distribution difference of different fault modes. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a schematic diagram of the feature extraction module in the present application; Figure 2A schematic diagram of an iterative change curve of the accuracy of the battery fault diagnosis model of the application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the application will be apparently and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0019] It should be understood that the step numbers used herein are only for the convenience of description, and are not limited to the execution sequence of the steps.

[0020] It should be understood that the terms used in the specification of the application are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in the specification and the appended claims of the application, the singular forms "a", "an" and "the" are intended to include the plural forms, unless the context clearly indicates otherwise.

[0021] The terms "comprise" and "include" indicate the presence of described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0022] The term "and / or" means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0023] The present embodiment provides a battery fault diagnosis method, comprising the following steps: S1, obtaining running data of a target battery; S2, constructing a battery fault diagnosis model, the battery fault diagnosis model comprising at least two feature extraction modules and a classification layer; the feature extraction modules are alternately connected through residual networks, and each feature extraction module comprises a convolution layer, an exponential linear unit and a batch normalization module connected in sequence; S3, performing feature extraction on the running data through the convolution layer to obtain running features; if the running features are negative value features, performing exponential transformation on the running features through the exponential linear unit to obtain target features; if the running features are not negative value features, performing linear transformation on the running features through the exponential linear unit to obtain target features; performing batch normalization processing on the target features through the batch normalization module to obtain final target features; S4, performing fault diagnosis on the target battery based on the final target features through the classification layer to obtain a diagnosis result.

[0024] The embodiment of the application extracts high-order abstract features in the battery operation data layer by layer through a multi-layer convolution structure, captures complex nonlinear relationships between parameters such as voltage, current, temperature, and the like, and adopts differential processing for positive and negative features, thereby ensuring linear propagation efficiency of the positive features and nonlinearly enhancing the negative abnormal features, especially suitable for battery operation data that may have negative values, effectively retaining negative feature information, and improving detection accuracy for fault types such as reverse charging.

[0025] In an optional embodiment, the operation data of the target battery refers to a set of multi-element parameters reflecting the real-time working state of the battery, which can include time series data of physical quantities such as single battery voltage data, single battery current data, and single battery temperature data. For example, the current and voltage curves can be collected in real time through the sensors built in the battery management system (BMS), and the temperature distribution data can be obtained through a thermistor network. The charge and discharge cycle data under specific working conditions can also be obtained through laboratory test equipment, and the data source or collection method is not limited to a single one.

[0026] In an optional embodiment, step S1 includes, but is not limited to, the following steps: obtaining single battery voltage of the target battery, calculating single battery voltage difference according to the single battery voltage, obtaining single battery internal resistance of the target battery, calculating parallel circulating voltage difference according to the single battery internal resistance and the single battery voltage difference, and performing exponential amplification on the parallel circulating voltage difference to obtain the circulating voltage data.

[0027] The embodiment of the application performs exponential amplification on the parallel circulating voltage difference, realizes nonlinear amplification of the tiny but critical circulating voltage difference, strengthens the abnormal signal, and improves the detection accuracy of fault types such as current distribution abnormality (e.g., poor contact of the connecting member, sudden change of internal resistance).

[0028] Referring to Figure 1 In an optional embodiment, the battery fault diagnosis model is a pre-trained convolutional neural network, which includes two feature extraction modules and a classification layer, the feature extraction modules are alternately connected through a residual network, that is, a jump connection is adopted, and each feature extraction module includes convolution layers, exponential linear units, and batch normalization modules connected in sequence.

[0029] In this embodiment, the principles of convolutional layer 110 and convolutional layer 111 are the same, and similarly, the principles of exponential linear unit 120 and exponential linear unit 121 are the same, and the principles of batch normalization module 130 and batch normalization module 131 are the same, and there may be differences in their respective structures, for example, convolutional layer 110 may use a 3x3 convolutional kernel, which can better capture local features, such as small fluctuation patterns in battery data; while convolutional layer 111 may use a 5x5 convolutional kernel, which can obtain more extensive context information, which may be used to capture larger range trend features in battery data; for example, the value of in exponential linear unit 120 may be 0.1, and its saturation characteristics on the negative half-axis are relatively weak; while the value of in exponential linear unit 121 may be 0.2, and the saturation degree on the negative half-axis is deeper, and the response to negative input will be different. Convolutional layers and exponential linear units are alternately connected through residual connections to form a residual block. Multiple residual blocks can be stacked together to build a deeper network.

[0030] The convolutional layer slides a set of filters (also known as convolutional kernels or kernels) over the input data and performs element-wise multiplication (dot product) with the local regions of the input data, then weighted sum and adds a bias term to get an output value. This output value is processed by a nonlinear activation function (such as ReLU) to get the input feature map. Each filter is responsible for detecting specific features in the input data, such as edges, textures, etc. Each neuron in the convolutional layer is only connected to a local region of the input data, which helps to reduce the number of parameters and computational complexity while preserving the spatial structure information of the image. For the same filter, the weights used when sliding over the entire input data are the same, reducing the number of parameters and making the model invariant to image translation. Since the convolutional kernel slides over the input data and extracts features, if the input data is translated, the convolutional layer can still detect the same features. The exponential linear unit is a nonlinear activation function that can increase the nonlinear characteristics of the network and alleviate the problem of neuron death caused by the ReLU activation function when the input is negative. The residual network simplifies the optimization process of deep networks by introducing residual connections, which allow the network layers to learn the "residual" between the input and output, so that information can be directly passed from the previous layer to the next layer without being affected by subsequent layers. The batch normalization module is configured to standardize a small batch of data, making the mean of the data close to 0 and the variance close to 1, which helps to reduce the problems of gradient vanishing and gradient explosion, speed up the convergence speed of the network, and improve the stability and generalization ability of the model.

[0031] Further, in an optional embodiment, the pre-training process comprises: obtaining historical operation data of the battery as sample data; inputting the sample data into an initial battery fault diagnosis model, and extracting features layer by layer through a feature extraction module alternately connected by a residual network; performing batch normalization processing on the features output by each layer, and inputting the normalized features into an exponential linear unit for nonlinear transformation; calculating a prediction result according to the nonlinearly transformed features, and optimizing the battery fault diagnosis model parameters by combining the prediction result with a loss function (such as a cross-entropy loss function) to obtain a trained battery fault diagnosis model.

[0032] In an optional embodiment, the mathematical expression of the nonlinear transformation performed by the exponential linear unit is: ; wherein, is an expression of an exponential linear unit ELU function, is input data of the exponential linear unit, is a constant less than zero, is a natural constant. The ELU function introduces a nonlinear characteristic, so that the model can learn complex patterns. Since the ReLU function sets negative values to zero, it can cause neurons to die, i.e., no longer respond to input. The ELU function introduces an exponential function in the negative interval, ensuring that all neurons have gradients, thereby effectively avoiding this problem. The negative interval output of the ELU function is close to zero, but not zero, which helps to speed up the learning process of the model. Specifically, in the positive value region, the ELU function keeps the input unchanged, allowing positive signals to pass through the network. When the input is less than zero, the ELU function introduces an exponential term , which ensures that there is a non-zero gradient in the negative value region, thereby alleviating the problem of neuron saturation and effectively preventing gradient disappearance, thereby effectively improving the accuracy of the battery fault diagnosis model training.

[0033] In an optional embodiment, the batch normalization module performs batch normalization processing on the target features to obtain final target features, specifically as follows: When processing a large amount of data, due to the limitation of computing resources (such as memory, GPU processing capability, etc.), all data will not be input into the model at one time for training or inference. Instead, a large-scale data set is divided into several small subsets, each of which is called a "batch". When the model processes the operation data of the target battery, the operation data is divided into multiple batches, and the target features obtained are also divided into multiple batches.

[0034] The mean and variance of the target features of the current batch are calculated, the target features are normalized based on the mean and variance to obtain normalized target features, the normalized target features are input into the next feature extraction module, and the processing of all feature extraction modules is completed to obtain the normalized target features output by the last feature extraction module, and the feature is taken as the final target feature, wherein: The mean is the average of all data points, and the variance is a statistical measure of the degree of difference between data points and their mean. The standardization process converts data into a form with zero mean and unit variance. The above steps are applied to each feature of the input data to ensure that each feature has a zero mean and a unit variance, which helps the model better learn the relative importance between features and improves the convergence speed and performance of the model.

[0035] The batch normalization module is used to accelerate training and enhance stability. By adjusting the mean and variance of each mini-batch in the network layer to standardize its input, it ensures that the network layer obtains consistent input distribution at different training stages. The normal form of the normalization process first subtracts the mean of the input data; second, divide the result by the standard deviation of the input data. However, normalizing all input data will result in excessive computation. Therefore, inspired by the idea of mini-batch gradient descent, the data is normalized in batches, which not only reduces the amount of calculation, but also improves the processing efficiency. The mathematical expression of the batch normalization module includes: Mean formula: ; Variance formula: ; Normalization formula: ; Where, is the mean, is the batch number, is the input data of the th batch, is the standard deviation, is the variance, is the normalized input data, is a small quantity introduced to prevent division by zero.

[0036] The embodiment of the present application reduces the internal covariant bias by batch normalization, making the input distribution of each layer more stable, thereby accelerating the training process of the model. And the output value of the network is limited within a reasonable range, reducing the fluctuation of the model during the training process, improving the stability of the model. In addition, batch normalization helps the model to learn more robust feature representation, thereby improving the generalization ability of the model.

[0037] In an optional embodiment, the battery fault diagnosis method further comprises: obtaining sample data, counting the number of different fault category labels in the sample data to obtain a first number; the fault category labels include overcharge, overdischarge, reverse charging, poor contact of a connecting piece, internal resistance mutation, etc.; calculating the weight of the fault category label based on the inverse number of the first number to obtain a category weight; obtaining a prediction result corresponding to the sample data through an initial battery fault diagnosis model, performing weighted calculation based on the prediction result, the fault category label and the category weight to obtain a loss value; and optimizing the hyperparameters of the initial battery fault diagnosis model based on the loss value to obtain a trained battery fault diagnosis model.

[0038] The embodiment of the present application calculates the weight of the fault category label as the category weight based on the inverse number of the number of fault category labels, and performs weighted calculation on the loss value of different category faults through the category weight, so that the influence of a minority class fault (such as thermal runaway, accounting for <3%) in the loss function is improved, and the influence caused by the uneven distribution of fault data in the actual scene is reduced.

[0039] In an optional embodiment, after the weight of the fault category label is calculated based on the inverse number of the first number to obtain the category weight, the battery fault diagnosis method further comprises: obtaining a preset sliding window, calculating the number of fault category labels in the sliding window to obtain a second number; calculating the ratio of the first number and the second number to obtain a first ratio; and updating the category weight according to the first ratio to obtain an updated category weight.

[0040] The embodiment of the present application calculates the number of fault category labels to obtain a second number through a sliding window, calculates the ratio of the first number and the second number to obtain a first ratio, and updates the category weight according to the first ratio, thereby enhancing the sensitivity of the model to a minority fault class in the current window and enabling more fine processing of local imbalance. For example, the frequency of the thermal runaway fault in the window is significantly higher than the historical average level, the updated weight increases the contribution of this type of loss, enhances the learning of this type of feature, and thus improves the fault diagnosis accuracy.

[0041] In an optional embodiment, to further verify the improvement of the training accuracy and generalization ability of the battery fault diagnosis model of the application, an ablation study is designed to evaluate the contribution of each technology to the effectiveness of the model. Among them, the unoptimized convolutional neural network, the convolutional neural network model only introducing batch normalization (BN), only introducing jump connection (RC), and only introducing exponential linear unit (ELU) are respectively referred to as CNN, CNN+BN, CNN+RC, and CNN+ELU. The model introducing two or three of BN, RC and ELU at the same time is respectively named as CNN+BR, CNN+BE, CNN+RE, and CNN+BRE, and the application is CNN+BRE.

[0042] Table 1 shows the performance indicators of the fault diagnosis model of the application. The performance of the basic CNN model is evaluated first, and then the improved modules are gradually added. It can be observed that with the addition of each new module, the accuracy (Acc), precision (Pre), recall (Rec) and F1 score of the model will increase. When only one of BN, ELU and RC is added, the accuracy increases to 0.9071, 0.9286 and 0.9286, respectively, slightly higher than the basic model. In addition, the results show that BN and RC have obvious enhancement on the other three performance indicators. When two of BN, ELU and RC are added, the accuracy reaches more than 0.93, and the other three performance indicators also improve significantly.

[0043] Table 1 Fault diagnosis model performance evaluation table

[0044] It can be seen that the fault diagnosis model of the application shows the best performance, with an accuracy of 0.9714, a precision of 0.9595, a recall of 0.9662, and an F1 score of 0.9629. These important performance indicators emphasize the effectiveness of the BRE module in enhancing the CNN model, and the accuracy and stability of the model have been greatly improved. From the subsequent accuracy curve, it can be known that the accuracy of the model has been close to stable after about 100 iterations, and the convergence performance of the model has also been improved.

[0045] In one example, please refer to Figure 2In the figure, testacc is a test accuracy, representing the correct prediction proportion of the model on the test set, and trainacc is a training accuracy, representing the correct prediction proportion of the model on the training set. Taking the final improved CNN+BRE fault diagnosis model as an example, the training set and the test set have strong stability in the iteration process, which is obviously better than the other models. The accuracy is already very high at about 100 iterations, and in order to compare with other diagnosis models, the early stopping method is not adopted in the research process. It has been proved that the CNN+BRE model is relatively stable in the subsequent iteration process and tends to converge. When the iteration is close to 200 times, it has been stabilized at more than 97%, so the improved model has a faster convergence speed, a very high accuracy and a good stability, and can be applied to the fault diagnosis of the internal short circuit of the battery pack.

[0046] The embodiment of the present application also provides a battery fault diagnosis device, which can realize the battery fault diagnosis method. The running data acquisition module is used for acquiring the running data of the target battery. The battery fault diagnosis module is internally constructed with a battery fault diagnosis model, and the battery fault diagnosis model comprises at least two feature extraction modules and a classification layer; the feature extraction modules are alternately connected through a residual network, and each feature extraction module comprises a convolution layer, an exponential linear unit and a batch normalization module connected in sequence, wherein: The running data is subjected to feature extraction through the convolution layer to obtain running features; if the running features are negative value features, the exponential linear unit is used for exponential transformation to obtain target features; if the running features are not negative value features, the exponential linear unit is used for linear transformation to obtain target features; and the batch normalization module is used for batch normalization processing on the target features to obtain final target features; Based on the final target features, the classification layer is used for fault diagnosis of the target battery to obtain a diagnosis result.

[0047] The specific implementation of the battery fault diagnosis device is basically the same as that of the above-mentioned battery fault diagnosis method, and will not be repeated here.

[0048] The embodiment of the present application also provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor realizes the above-mentioned battery fault diagnosis method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer and the like.

[0049] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the above-mentioned battery fault diagnosis method.

[0050] In the embodiments of the present application, "at least one" means one or more, and "multiple" means two or more. The "and / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the cases of A alone, A and B together, and B alone. Wherein A and B can be singular or plural. The character " / " generally represents an "or" relationship between the front and rear associated objects. "At least one of the following" and the like means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0051] Those skilled in the art can appreciate that the units and algorithm steps described in the embodiments disclosed herein can be realized by electronic hardware, computer software and combination of electronic hardware and computer software. Whether the functions are realized in hardware or software mode depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0052] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0053] In several embodiments provided by the present application, any function realized in the form of a software function unit and sold or used as an independent product can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory; hereinafter referred to as: ROM), a random access memory (Random Access Memory; hereinafter referred to as: RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0054] The above merely illustrates the embodiments of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, which are made by using the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. A battery fault diagnosis method, characterized in that: The method comprises: Obtaining the operating data of the target battery; Constructing a battery fault diagnosis model, the battery fault diagnosis model comprising at least two feature extraction modules and a classification layer; the feature extraction modules are alternately connected via a residual network, and each feature extraction module comprises a convolutional layer, an exponential linear unit, and a batch normalization module connected in sequence; Performing feature extraction on the operation data through the convolution layer to obtain an operation feature; if the operation feature is a negative feature, performing exponential transformation through the exponential linear unit to obtain a target feature; if the operation feature is not a negative feature, performing linear transformation through the exponential linear unit to obtain a target feature; performing batch normalization processing on the target feature through the batch normalization module to obtain a final target feature; Based on the final target features, the target battery is diagnosed for faults through the classification layer to obtain a diagnosis result.

2. A battery fault diagnosis method according to claim 1, characterized in that: The target features are normalized in batches by the batch normalization module to obtain the final target features, specifically: Calculate the mean and variance of the target features of the current batch, normalize the target features based on the mean and variance to obtain normalized target features, input the normalized target features into the feature extraction module of the next layer, and continue processing until all feature extraction modules are completed. Finally, obtain the normalized target features output by the last feature extraction module, and use this feature as the final target feature.

3. A battery fault diagnosis method according to claim 1, characterized in that: The method further comprises: Historical battery operation data is obtained as sample data; the sample data is input into the initial battery fault diagnosis model, and features are extracted layer by layer through feature extraction modules alternately connected with the residual network; the features output by each layer are normalized in batches, and the normalized features are input into the exponential linear unit for nonlinear transformation; the prediction results are calculated based on the features after the nonlinear transformation, and the prediction results are combined with the loss function to optimize the parameters of the initial battery fault diagnosis model to obtain the trained battery fault diagnosis model.

4. A battery fault diagnosis method according to claim 1, characterized in that: The operating data includes circulating voltage data. The acquiring the operating data of the target battery includes: Obtaining the single cell voltage of the target battery, and calculating the single cell voltage difference based on the single cell voltage; Obtaining the internal resistance of a single cell of the target battery, and calculating the parallel circulating current voltage difference according to the internal resistance of the single cell and the voltage difference of the single cell; The parallel circulating current voltage difference is exponentially amplified to obtain the circulating current voltage data.

5. A battery fault diagnosis method according to claim 1, characterized in that: The method further comprises: Acquire historical battery operation data as sample data, and count the number of different fault category labels in the sample data to obtain a first number; Calculating the weight of the fault category label based on the opposite of the first number to obtain a category weight; Obtaining a prediction result corresponding to the sample data through an initial battery fault diagnosis model, and performing a weighted calculation based on the prediction result, the fault category label, and the category weight to obtain a loss value; The hyperparameters of the initial battery fault diagnosis model are optimized based on the loss value to obtain a trained battery fault diagnosis model.

6. A battery fault diagnosis method according to claim 5, characterized in that: After calculating the weight of the fault category label based on the inverse of the first number to obtain the category weight, the method further includes: Obtaining a preset sliding window, and calculating the number of fault category labels within the sliding window to obtain a second number; Calculate the ratio of the first quantity to the second quantity to obtain a first ratio; The category weight is updated according to the first ratio to obtain an updated category weight.

7. A battery fault diagnosis method according to claim 4, characterized in that: The operation data also includes single cell voltage data, single cell current data, and single cell temperature data.

8. A battery fault diagnosis device, characterized in that: include: An operating data acquisition module, used to acquire operating data of a target battery; A fault diagnosis module has a battery fault diagnosis model built inside. The battery fault diagnosis model includes at least two feature extraction modules and a classification layer. The feature extraction modules are alternately connected through a residual network. Each feature extraction module includes a convolutional layer, an exponential linear unit, and a batch normalization module connected in sequence, wherein: Performing feature extraction on the operation data through the convolution layer to obtain an operation feature; if the operation feature is a negative feature, performing exponential transformation through the exponential linear unit to obtain a target feature; if the operation feature is not a negative feature, performing linear transformation through the exponential linear unit to obtain a target feature; performing batch normalization processing on the target feature through the batch normalization module to obtain a final target feature; Based on the final target features, the target battery is diagnosed for faults through the classification layer to obtain a diagnosis result.

9. A battery fault diagnosis device, characterized in that: The battery fault diagnosis device includes: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the battery fault diagnosis device executes the steps of the battery fault diagnosis method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the battery fault diagnosis method according to any one of claims 1 to 7 are implemented.

Images