New energy automobile battery system anomaly detection method suitable for multiple scenes

By using the Trans-GAN model and CUSUM method in the Tensorflow deep learning framework, the abnormal situation of the new energy vehicle battery system in multiple vehicle use scenarios is solved, and the problem of poor detection effect of the existing technology in many scenarios is solved. Achieve high-precision and low-cost abnormality detection of battery system.

CN118980932AActive Publication Date: 2024-11-19BEIJING INST OF TECH
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
CN202410720925.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-05
Publication Date
2024-11-19
Estimated Expiration
2044-06-05

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively detect abnormal situations in new energy vehicle battery systems in various vehicle use scenarios, especially due to the different characteristics of the parameters of the battery during charging and discharging, resulting in the detection effect of the existing methods in various scenarios. Not good.

Method used

The Trans-GAN model is established using the deep learning framework Tensorflow, combined with the generator and discriminator of the Transformer model, and through data fragment division, feature parameter extraction and CUSUM method, a battery system abnormality detection model is established in multiple scenarios, using sliding windows and cumulative and Method to detect abnormal battery voltage.

Benefits of technology

It realizes effective detection of abnormal voltage inconsistency of battery systems in different vehicle usage scenarios, and has higher accuracy and lower computing costs compared with other GAN model-based methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure SMS_1
    Figure SMS_1
  • Figure SMS_3
    Figure SMS_3
  • Figure SMS_5
    Figure SMS_5
Patent Text Reader

Abstract

The invention provides a new energy automobile battery system anomaly detection method suitable for multiple scenes, which performs data set construction and feature extraction on different automobile use scenes, establishes an unsupervised anomaly detection model based on Trans-GAN, and uses an improved cumulative sum (CUSUM) method on this basis to detect the anomaly of a new energy automobile battery system. Effective detection of the voltage inconsistency abnormity of the battery system in different scenes can be realized. Due to the non-convolution characteristic of the adopted Trans-GAN model, compared with other methods based on the GAN model, the method has obvious advantages in the aspects of precision and calculation cost.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis of power battery systems of new energy vehicles, and specifically relates to a new energy vehicle battery system anomaly detection method applicable to multiple scenarios. Background Art

[0002] At present, various abnormal conditions of automotive power battery systems during use are still difficult to detect directly, and require indirect judgment using parameters such as battery voltage, current, and SOC. However, since the above parameters of new energy vehicles in typical vehicle use scenarios such as charging and discharging have different changing characteristics and trends, there are still few battery system abnormality detection methods in the prior art that can be well applied to a variety of different vehicle use scenarios. Some existing technologies use data-driven anomaly detection methods based on machine learning. Although they can obtain relatively accurate battery anomaly prediction models by training with a large amount of battery sample data, their high dependence on the quantity and quality of sample data also limits the practicality of such methods. Summary of the invention

[0003] In view of this, in view of the technical problems existing in the art, the present invention provides a new energy vehicle battery system abnormality detection method applicable to multiple scenarios, which specifically includes the following steps:

[0004] Step 1: divide the historical data of the normal operation of the new energy vehicle into data segments based on the charging and discharging vehicle scenarios, and use the divided data segments to construct data sets corresponding to the charging scenario and the discharging scenario respectively;

[0005] Step 2: Extract characteristic parameters related to battery voltage using the data sets of charging and discharging scenarios;

[0006] Step 3: Use the deep learning framework Tensorflow to build a Trans-GAN model. The generator and discriminator in the model are both built based on the Transformer model. The generator accepts the input random noise and outputs a forged battery voltage fragment, which is input into the discriminator together with the real battery voltage fragment to identify the authenticity of the fragment. At the same time, the generator and the discriminator generate corresponding reconstruction error loss and identification error loss respectively.

[0007] Step 4: Use the data sets of charging and discharging scenarios to train the established Trans-GAN model respectively until the predetermined number of training times is reached or the loss functions of the generator and the discriminator reach a certain convergence threshold;

[0008] Step 5. Compare the reconstructed voltage data obtained by the trained Trans-GAN model with the original voltage data in the actual vehicle operation, and calculate the residual between the two; set the sliding window and calculate the cumulative sum CUSUM (Cumulative Sum) value of the residual in each window; compare the calculated CUSUM value with the corresponding threshold. If it exceeds the threshold, record that the battery is abnormal in the corresponding scenario.

[0009] Furthermore, in step 2, for the charging scenario, the Pearson correlation analysis method is specifically used to extract SOC and current data as characteristic parameters;

[0010] For the discharge scenario, the following feature extraction method based on the shape of time series data is used:

[0011] First, find the midpoint of every two frames of the discharge data segment, recorded as:

[0012]

[0013] Where j represents the jth data frame;

[0014] Then calculate the slope at the midpoint

[0015]

[0016] And calculate the slope change rate between two adjacent slopes:

[0017]

[0018] Calculate the average slope change rate of the discharge data segment according to the obtained slope change rate, and set it as the threshold of the slope change rate;

[0019] Identify turning points in a data segment based on the set slope change rate threshold and the following criteria:

[0020]

[0021] Based on the identified turning points, the peaks and valleys of the data segment are identified based on the following criteria:

[0022]

[0023] All turning points are connected to achieve data dimensionality reduction and filter out unclear features, and the retained data curve is used as the characteristic parameter corresponding to the discharge scenario.

[0024] Furthermore, in the Trans-GAN model established in step 3, the reconstruction error loss function of the generator is specifically expressed as follows:

[0025]

[0026] Among them, L G is the reconstruction error loss, N is the number of samples, G(z i ) is the generator's response to the input noise z i The generated result indicates that the generator is generated from random noise z i The data generated in i is the data of the i-th sample, which is the real data, |·| represents the L2 norm of the vector, that is, the Euclidean distance;

[0027] The specific form of the discriminator's identification error loss function is as follows:

[0028]

[0029] Among them, L D is the identification error loss, N is the number of samples, y i is the label of the i-th sample, which takes a value of 0 or 1. If it is real data, then y i =1; if it is generated data, then y i =0;x i is the data of the i-th sample, which is real data or generated data, D(x i ) is the discriminator for sample x i The output result indicates that x i The probability of being judged as real data.

[0030] Furthermore, in step 4, the training of the generator and the discriminator is performed alternately, and the discriminator is trained first and then the generator each time; the Adam optimizer is used to minimize the two-error loss function in the model to iteratively update the parameters of the generator and the discriminator during the training process.

[0031] Furthermore, the CUSUM threshold set in step 5 is determined based on the 3σ principle using the distribution of the calculated CUSUM values.

[0032] The above-mentioned new energy vehicle battery system anomaly detection method applicable to multiple scenarios provided by the present invention performs data set construction and feature extraction for different vehicle use scenarios, and establishes an unsupervised anomaly detection model based on Trans-GAN. On this basis, the improved cumulative sum (CUSUM) method is used to achieve effective detection of battery system voltage inconsistency anomalies in different scenarios. The non-convolution characteristics of the adopted Trans-GAN model also make the present invention have obvious advantages in accuracy and computational cost compared with other methods based on the GAN model. DETAILED DESCRIPTION

[0033] The technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] The present invention provides a method for detecting abnormalities in a new energy vehicle battery system applicable to multiple scenarios, and specifically comprises the following steps:

[0035] Step 1: divide the historical data of new energy vehicles (vehicles that are still running until the data is obtained) during normal operation into data segments based on charging and discharging vehicle scenarios, and use the divided data segments to construct data sets corresponding to the charging scenario and the discharging scenario respectively;

[0036] Step 2: Extract characteristic parameters related to battery voltage using the data sets of charging and discharging scenarios;

[0037] Step 3: Use the deep learning framework Tensorflow to build a Trans-GAN model. The generator and discriminator in the model are both built based on the Transformer model. The generator accepts the input random noise and outputs a forged battery voltage fragment, which is input into the discriminator together with the real battery voltage fragment to identify the authenticity of the fragment. At the same time, the generator and the discriminator generate corresponding reconstruction error loss and identification error loss respectively.

[0038] Step 4: Use the data sets of charging and discharging scenarios to train the established Trans-GAN model respectively until the predetermined number of training times is reached or the loss functions of the generator and the discriminator reach a certain convergence threshold;

[0039] Step 5. Compare the reconstructed voltage data obtained by the trained Trans-GAN model with the original voltage data in the actual vehicle operation, and calculate the residual between the two; set the sliding window and calculate the cumulative sum CUSUM (Cumulative Sum) value of the residual in each window; compare the calculated CUSUM value with the corresponding threshold. If it exceeds the threshold, record that the battery is abnormal in the corresponding scenario.

[0040] In a preferred embodiment of the present invention, in step 2, for the charging scenario, since the working condition during charging is stable, the voltage data curve is stable and less affected by external interference, the Pearson correlation analysis method is used to extract SOC and current data as characteristic parameters to improve the convergence speed of the model;

[0041] For the discharge scenario, due to the complex working conditions during discharge, it will be affected by factors such as traffic conditions and personal driving habits, resulting in large fluctuations in the voltage data curve during discharge. If used directly, it will reduce the model calculation efficiency and produce large errors. Therefore, it is necessary to reduce the dimension of the data of the discharge process and retain important features. Therefore, the following feature extraction method based on the shape of time series data is used:

[0042] First, find the midpoint of every two frames of the discharge data segment, recorded as:

[0043]

[0044] Where j represents the jth data frame;

[0045] Then calculate the slope at the midpoint

[0046]

[0047] And calculate the slope change rate between two adjacent slopes:

[0048]

[0049] Calculate the average slope change rate of the discharge data segment according to the obtained slope change rate, and set it as the threshold of the slope change rate;

[0050] Identify turning points in a data segment based on the set slope change rate threshold and the following criteria:

[0051]

[0052] Based on the identified turning points, the peaks and valleys of the data segment are identified based on the following criteria:

[0053]

[0054] All turning points are connected to achieve data dimensionality reduction and filter out unclear features, and the retained data curve is used as the characteristic parameter corresponding to the discharge scenario.

[0055] In a preferred embodiment of the present invention, in the Trans-GAN model established in step 3, the reconstruction error loss function of the generator is specifically in the following form:

[0056]

[0057] Among them, L G is the reconstruction error loss, N is the number of samples, G(z i ) is the generator's response to the input noise z i The generated result indicates that the generator is generated from random noise zi The data generated in i is the data of the i-th sample, which is the real data, |·| represents the L2 norm of the vector, that is, the Euclidean distance;

[0058] The specific form of the discriminator's identification error loss function is as follows:

[0059]

[0060] Among them, L D is the identification error loss, N is the number of samples, y i is the label of the i-th sample, which takes a value of 0 or 1. If it is real data, then y i =1; if it is generated data, then y i =0;x i is the data of the i-th sample, which is real data or generated data, D(x i ) is the discriminator for sample x i The output result indicates that x i The probability of being judged as real data.

[0061] In a preferred embodiment of the present invention, in step four, the training of the generator and the discriminator is performed alternately, and the discriminator is trained first and then the generator each time; the Adam optimizer is used to minimize the two-error loss function in the model to iteratively update the parameters of the generator and the discriminator during the training process.

[0062] In a preferred embodiment of the present invention, the CUSUM threshold set in step 5 is determined based on the 3σ principle using the distribution of the calculated CUSUM value. By comparing the calculated CUSUM value with the corresponding threshold, if it exceeds the set abnormality detection threshold, the battery abnormality is recorded and maintenance is performed in a timely manner to ensure the safe operation of the new energy vehicle battery system.

[0063] It should be understood that the size of the serial numbers of the steps in the embodiment of the present invention does not mean 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 embodiment of the present invention.

[0064] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting abnormality in a new energy vehicle battery system applicable to multiple scenarios, characterized in that: The specific steps include: Step 1: divide the historical data of the normal operation of the new energy vehicle into data segments based on the charging and discharging vehicle scenarios, and use the divided data segments to construct data sets corresponding to the charging scenario and the discharging scenario respectively; Step 2: Extract characteristic parameters related to battery voltage using the data sets of charging and discharging scenarios; Step 3: Use the deep learning framework Tensorflow to build a Trans-GAN model. The generator and discriminator in the model are both built based on the Transformer model. The generator accepts the input random noise and outputs a forged battery voltage fragment, which is input into the discriminator together with the real battery voltage fragment to identify the authenticity of the fragment. At the same time, the generator and the discriminator generate corresponding reconstruction error loss and identification error loss respectively. Step 4: Use the data sets of charging and discharging scenarios to train the established Trans-GAN model respectively until the predetermined number of training times is reached or the loss functions of the generator and the discriminator reach a certain convergence threshold; Step 5. Compare the reconstructed voltage data obtained by the trained Trans-GAN model with the original voltage data in the actual vehicle operation, and calculate the residual between the two; set the sliding window and calculate the cumulative sum CUSUM value of the residual in each window; compare the calculated CUSUM value with the corresponding threshold, and if it exceeds the threshold, record that the battery is abnormal in the corresponding scenario.

2. The method according to claim 1, characterized in that: In step 2, for the charging scenario, the Pearson correlation analysis method is used to extract SOC and current data as characteristic parameters; For the discharge scenario, the following feature extraction method based on the shape of time series data is used: First, find the midpoint of every two frames of the discharge data segment, recorded as: Where j represents the jth data frame; Then calculate the slope at the midpoint And calculate the slope change rate between two adjacent slopes: Calculate the average slope change rate of the discharge data segment according to the obtained slope change rate, and set it as the threshold of the slope change rate; Identify turning points in a data segment based on the set slope change rate threshold and the following criteria: Based on the identified turning points, the peaks and valleys of the data segment are identified based on the following criteria: All turning points are connected to achieve data dimensionality reduction and filter out unclear features, and the retained data curve is used as the characteristic parameter corresponding to the discharge scenario.

3. The method according to claim 1, characterized in that: In the Trans-GAN model established in step 3, the reconstruction error loss function of the generator is specifically expressed as follows: Among them, L G is the reconstruction error loss, N is the number of samples, G(z i ) is the generator's response to the input noise z i The generated result indicates that the generator is generated from random noise z i The data generated in i is the data of the i-th sample, which is the real data, and |·| represents the L2 norm of the vector, that is, the Euclidean distance; The specific form of the discriminator's identification error loss function is as follows: Among them, L D is the identification error loss, N is the number of samples, y i is the label of the i-th sample, which takes a value of 0 or 1. If it is real data, then y i =1; if it is generated data, then y i =0;x i is the data of the i-th sample, which is real data or generated data, D(x i ) is the discriminator for sample x i The output result indicates that x i The probability of being judged as real data.

4. The method according to claim 1, characterized in that: In step 4, the training of the generator and the discriminator is performed alternately, each time the discriminator is trained first and then the generator; the Adam optimizer is used to minimize the two-error loss function in the model to iteratively update the parameters of the generator and the discriminator during the training process.

5. The method according to claim 1, characterized in that: The CUSUM threshold set in step 5 is determined based on the 3σ principle using the distribution of the calculated CUSUM values.

Citation Information

Patent Citations

  • PLC anomaly detection method based on electromagnetic side channel

    CN112649675A

  • Aging early warning method and system based on lithium ion battery capacity diving point, automobile and computer storage medium

    CN113030758A

  • Battery management device and battery management method

    CN115917341A

  • Iterative unsupervised non-intrusive load identification method and system

    CN115932428A

  • Vehicle power battery system fault diagnosis method based on LSTM-GAN

    CN117110888A