An lstm-gan-based fault diagnosis method for vehicle power battery system

Abstract

The application provides a vehicle power battery system fault diagnosis method based on LSTM-GAN, utilizes vehicle historical operation data, and performs unsupervised learning training on the relationship between each battery monomer voltage and each characteristic parameter based on the model of LSTM-GAN, and through statistical analysis on the deviation coefficient obtained by the LSTM-GAN model of the vehicle using the same battery system, the threshold detection of the monomer voltage anomaly is realized. The method effectively solves the problem of difficult battery monomer state monitoring in the power battery system, has less dependence on the number and quality of data, and can reduce the calculation cost while improving the accuracy and efficiency of fault diagnosis.

Description

Technical Field

[0001] This invention belongs to the field of vehicle power battery system fault diagnosis technology, specifically relating to a fault diagnosis method for vehicle power battery systems based on LSTM-GAN. Background Technology

[0002] During the use of automotive power battery systems, inconsistencies often arise in the voltage, internal resistance, capacity, and performance of individual cells. These inconsistencies tend to worsen with increasing charge-discharge cycles, leading to problems such as shortened battery life and increased likelihood of malfunctions and safety risks. Therefore, effective monitoring of the state of each cell within the battery system is crucial. However, due to the nonlinear characteristics of power battery systems during use, the real-time changes in cell state parameters impose significant limitations on methods for monitoring cell faults. While existing technologies utilize battery sample data combined with machine learning algorithms for fault diagnosis, their high dependence on the quantity and quality of sample data, coupled with substantial computational overhead, restricts their practical application. Summary of the Invention

[0003] In view of this, and to address the technical problems existing in this field, the present invention provides a fault diagnosis method for automotive power battery systems based on LSTM-GAN, specifically including the following steps:

[0004] Step 1: Extract battery state parameters, load parameters, external environmental parameters, and driving intention parameters from the power battery system during use. The battery state parameters include: individual cell voltage, SOC, individual cell probe temperature, and total current. The load parameter is specifically selected as motor output power, the external environmental parameter is specifically selected as ambient temperature, and the driving intention parameter is specifically selected as vehicle speed. Based on the time series of individual cell voltages, the extracted data is divided into two datasets: normal individual cell voltage segments and abnormal individual cell voltage segments.

[0005] Step 2: Use the deep learning framework Tensorflow to build an LSTM-GAN model. The model includes an LSTM network as a generator and another LSTM network as a discriminator. The generator accepts input random noise and outputs fake battery cell voltage segments, which are then input into the discriminator along with real battery cell voltage segments to distinguish between real and fake segments. At the same time, both the generator and the discriminator generate corresponding error losses.

[0006] Step 3: Use the normal battery cell voltage segment obtained in Step 1 to train the LSTM-GAN model to establish the correspondence between the battery cell voltage and other extracted parameters; during training, continuously adjust the model parameters of the discriminator and the generator based on the authenticity identification results of the discriminator, so that the fake battery cell voltage segment output by the generator gradually approaches the real battery cell voltage segment; training continues until the preset number of training times is reached or the error stabilizes and converges.

[0007] Step 4: Statistically sum the reconstruction error of the generator outputting the fake battery cell voltage segment during training with the discrimination error of the discriminator, and use this sum as the bias coefficient of the constructed LSTM-GAN model.

[0008] Step 5: Input the parameters of a vehicle's battery system into the LSTM-GAN model to obtain the deviation coefficients. By comparing the deviation coefficients with those of other vehicles using the same power battery system, determine whether the vehicle has abnormal battery cell voltages.

[0009] Furthermore, in step two, for the GAN model composed of the generator and discriminator, the Adam optimizer is specifically used to minimize its error function, thereby iteratively updating the parameters of the generator and discriminator during the training process.

[0010] Furthermore, in step three, the training of the LSTM-GAN model involves alternating between the discriminator and the generator, with the discriminator being trained first and the generator trained next.

[0011] Furthermore, in step four, the reconstruction error L of the generator is specifically calculated using the following formula. G :

[0012]

[0013] In the formula, N is the sample size, and G(z) i ) represents the generator's response to input noise z i The generation result indicates that the generator generates random noise z. i Data generated in x i For the i-th real battery cell voltage segment sample data, |·| represents the L2 norm (Euclidean distance) of the vector;

[0014] The discrimination error L of the discriminator is calculated using the following formula. D :

[0015]

[0016] In the formula, L D To represent the discrimination error, N is the number of samples, and y iThe label for the i-th sample data is 0 or 1. If it is the actual data y, then... i =1; if it is generated data, then y i =0, x i Let D(x) be the data for the i-th sample, which can be real data or generated data. i ) is the discriminator for sample x i The output result indicates that x i The probability of being identified as real data;

[0017] Then, the two errors are added together to obtain the deviation value Er calculated by the model for the i-th battery cell. i :

[0018] Er i =L G +L D

[0019] The deviation coefficient Er is obtained by summing the deviation values ​​of all individual battery cells:

[0020]

[0021] M represents the total number of battery cells in the power battery system.

[0022] Furthermore, step five, determining abnormal cell voltage, specifically includes the following processes:

[0023] 1) Input the usage parameters of a certain number of fault-free vehicles using the same power battery system into the LSTM-GAN model to obtain the corresponding deviation coefficients;

[0024] 2) Calculate the cumulative distribution function (CDF) of the obtained deviation coefficients;

[0025] 3) The deviation coefficient value is calibrated according to the cumulative distribution function of the deviation coefficient distribution, and it is used as the judgment threshold for abnormal battery cell voltage. This threshold is provided to the power battery voltage abnormality diagnosis module of these vehicles, so that the power battery voltage abnormality diagnosis module on the vehicle can compare the deviation coefficient calculated by the vehicle itself in real time with the threshold to determine whether there is an abnormality.

[0026] The LSTM-GAN-based fault diagnosis method for automotive power battery systems provided by this invention utilizes historical vehicle operating data and an LSTM-GAN model to perform unsupervised learning training on the relationship between the voltage of each battery cell and various characterization parameters. By statistically analyzing the deviation coefficients obtained from the LSTM-GAN model for vehicles using the same battery system, threshold detection of abnormal cell voltages is achieved. This method effectively solves the problem of difficult monitoring of the state of individual battery cells in power battery systems, and has less dependence on the quantity and quality of data. Therefore, it can improve the accuracy and efficiency of fault diagnosis while reducing computational costs. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the LSTM-GAN network structure established in the method provided by the present invention.

[0028] Figure 2 Here is a schematic diagram of the method flow provided by the present invention:

[0029] Figure 3 This is a histogram of the distribution of deviation coefficients statistically derived from an example based on the present invention;

[0030] Figure 4 This is a graph of the cumulative distribution function of the deviation coefficient calculated in an example based on the present invention. Detailed Implementation

[0031] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] The fault diagnosis method for automotive power battery systems based on LSTM-GAN provided by this invention, such as... Figure 2 As shown, the specific steps include:

[0033] Step 1: Extract battery state parameters, load parameters, external environmental parameters, and driving intention parameters from the power battery system during use. The battery state parameters include: individual cell voltage, SOC, individual cell probe temperature, and total current. The load parameter is specifically selected as motor output power, the external environmental parameter is specifically selected as ambient temperature, and the driving intention parameter is specifically selected as vehicle speed. Based on the time series of individual cell voltages, the extracted data is divided into two datasets: normal individual cell voltage segments and abnormal individual cell voltage segments.

[0034] Step 2: Use the deep learning framework Tensorflow to build an LSTM-GAN model, such as... Figure 1As shown, it includes an LSTM network as a generator and another LSTM network as a discriminator; the generator accepts input random noise and outputs fake battery cell voltage segments, which are then input together with real battery cell voltage segments into the discriminator to distinguish the authenticity of the segments. At the same time, both the generator and the discriminator generate corresponding error losses.

[0035] Step 3: Use the normal battery cell voltage segment obtained in Step 1 to train the LSTM-GAN model to establish the correspondence between the battery cell voltage and other extracted parameters; during training, continuously adjust the model parameters of the discriminator and the generator based on the authenticity identification results of the discriminator, so that the fake battery cell voltage segment output by the generator gradually approaches the real battery cell voltage segment; training continues until the preset number of training times is reached or the error stabilizes and converges.

[0036] Step 4: Statistically sum the reconstruction error of the generator outputting the fake battery cell voltage segment during training with the discrimination error of the discriminator, and use this sum as the bias coefficient of the constructed LSTM-GAN model.

[0037] Step 5: Input the parameters of a vehicle's battery system into the LSTM-GAN model to obtain the deviation coefficients. By comparing the deviation coefficients with those of other vehicles using the same power battery system, determine whether the vehicle has abnormal battery cell voltages.

[0038] In a preferred embodiment of the present invention, in step two, for the GAN model composed of the generator and the discriminator, the Adam optimizer is specifically used to minimize its error function, thereby iteratively updating the parameters of the generator and the discriminator during the training process.

[0039] In a preferred embodiment of the present invention, step three, training the LSTM-GAN model, specifically involves alternating between the discriminator and the generator, with the discriminator trained first and the generator trained later each time.

[0040] In a preferred embodiment of the present invention, step four specifically involves calculating the generator's reconstruction error L using the following formula. G :

[0041]

[0042] In the formula, N is the sample size, and G(z) i ) represents the generator's response to input noise z i The generation result indicates that the generator generates random noise z. i Data generated in x i For the i-th real battery cell voltage segment sample data, |·| represents the L2 norm (Euclidean distance) of the vector;

[0043] The discrimination error L of the discriminator is calculated using the following formula. D :

[0044]

[0045] In the formula, L D To represent the discrimination error, N is the number of samples, and y i The label for the i-th sample data is 0 or 1. If it is the actual data y, then... i =1; if it is generated data, then y i =0, x i Let D(x) be the data for the i-th sample, which can be real data or generated data. i ) is the discriminator for sample x i The output result indicates that x i The probability of being identified as real data;

[0046] Then, the two errors are added together to obtain the deviation value Er calculated by the model for the i-th battery cell. i :

[0047] Er i =L G +L D

[0048] The deviation coefficient Er is obtained by summing the deviation values ​​of all individual battery cells:

[0049]

[0050] M represents the total number of battery cells in the power battery system.

[0051] In a preferred embodiment of the present invention, step five, determining the abnormal voltage of a single battery cell, specifically includes the following process:

[0052] 1) Input the usage parameters of a certain number of fault-free vehicles using the same power battery system into the LSTM-GAN model to obtain the corresponding deviation coefficients;

[0053] 2) Calculate the cumulative distribution function (CDF) of the obtained deviation coefficients; specifically, you can first statistically analyze the probability distribution of the deviation coefficients, such as... Figure 3 The histogram shown can be obtained by integrating the probability distribution. Figure 4 The cumulative distribution function curve of the deviation coefficient is shown below;

[0054] 3) The deviation coefficient value is calibrated according to the cumulative distribution function of the deviation coefficient distribution. For example, the probability interval [5%, 95%] is used as the fault-free interval. This probability interval can cover 90% of the deviation coefficient data records. The corresponding interval [-3.33%, 3.21%] is set as the allowable interval of the deviation coefficient of the battery system under normal state. Based on this interval, the judgment threshold of abnormal battery cell voltage is determined and provided to the power battery voltage abnormality diagnosis module of these vehicles. The power battery voltage abnormality diagnosis module on the vehicle compares the deviation coefficient calculated by the vehicle itself in real time with the threshold to determine whether there is an abnormality.

[0055] It should be understood that the sequence number of each step in the embodiments of the present invention does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0056] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fault diagnosis method for automotive power battery systems based on LSTM-GAN, characterized in that: Specifically, the following steps are included: Step 1: Extract battery state parameters, load parameters, external environment parameters, and driving intention parameters from the power battery system during use. The battery state parameters include: battery cell voltage, SOC, cell probe temperature, and total current. The load parameter is the motor output power. The external environment parameter is the ambient temperature. The driving intention parameter is the vehicle speed. Based on the battery cell voltage time series, the extracted data is divided into two datasets: normal battery cell voltage segments and abnormal battery cell voltage segments. Step 2: Use the deep learning framework Tensorflow to build an LSTM-GAN model. The model includes an LSTM network as a generator and another LSTM network as a discriminator. The generator accepts input random noise and outputs fake battery cell voltage segments, which are then input into the discriminator along with real battery cell voltage segments to distinguish between real and fake segments. At the same time, both the generator and the discriminator generate corresponding error losses. Step 3: Use the normal battery cell voltage segment obtained in Step 1 to train the LSTM-GAN model to establish the correspondence between the battery cell voltage and other extracted parameters; during training, continuously adjust the model parameters of the discriminator and the generator based on the authenticity identification results of the discriminator, so that the fake battery cell voltage segment output by the generator gradually approaches the real battery cell voltage segment; training continues until the preset number of training times is reached or the error stabilizes and converges. Step 4: The reconstruction error of the generator's output of fake battery cell voltage segments during training is statistically summed with the discrimination error of the discriminator, and this sum is used as the bias coefficient of the constructed LSTM-GAN model; the generator's reconstruction error... Specifically, it is calculated using the following formula: In the formula, N For the sample size, For the generator to handle input noise The generation result indicates that the generator generates random noise. Data generated in For the first i Sample data of voltage segments from a real battery cell The L2 norm of a vector; The discrimination error of the discriminator is calculated using the following formula. : In the formula, To represent the discrimination error, N is the number of samples. For the first i The label for each sample data point takes a value of 0 or 1; if it is real data... =1; if it is generating data, then =0, Let be the data for the i-th sample, representing either real data or generated data. For the discriminator to sample The output result indicates that... The probability of being identified as real data; Then add the two errors together to obtain the model for the first... i Deviation value calculated for each individual battery cell : The deviation coefficient is obtained by summing the deviation values ​​of all individual battery cells. : M This represents the total number of battery cells in the power battery system. Step 5: Input the parameters of a vehicle's battery system into the LSTM-GAN model to obtain the deviation coefficients. By comparing the deviation coefficients with those of other vehicles using the same power battery system, determine whether the vehicle has abnormal battery cell voltages.

2. The method as described in claim 1, characterized in that: In step two, for the GAN model composed of the generator and discriminator, the Adam optimizer is used to minimize its error function, thereby iteratively updating the parameters of the generator and discriminator during the training process.

3. The method as described in claim 1, characterized in that: Step three involves training the LSTM-GAN model by alternating between the discriminator and the generator, training the discriminator first and then the generator each time.

4. The method as described in claim 3, characterized in that: Step five, determining abnormal cell voltage, specifically includes the following processes: 1) Input the usage parameters of a certain number of fault-free vehicles using the same power battery system into the LSTM-GAN model to obtain the corresponding deviation coefficients; 2) Calculate the cumulative distribution function of the obtained deviation coefficients; 3) The deviation coefficient value is calibrated according to the cumulative distribution function of the deviation coefficient, and it is used as the judgment threshold for abnormal battery cell voltage. This threshold is provided to the power battery voltage abnormality diagnosis module of these vehicles, so that the power battery voltage abnormality diagnosis module on the vehicle can compare the deviation coefficient calculated by the vehicle itself in real time with the threshold to determine whether there is an abnormal cell voltage.

Images