A method, system, application method and electronic device for estimating SOH of a power battery of a new energy vehicle

Through the CNN-Transformer model and data cleaning technology, combined with multi-dimensional key feature extraction, a SOH digital twin model was established, which solved the SOH estimation error problem caused by differences in the actual working conditions of the battery and laboratory test conditions, and achieved accurate real-time battery SOH It is estimated that the security of the battery management system has been improved.

CN118483614BActive Publication Date: 2024-11-15ANHUI UNIV
View PDF 12 Cites 0 Cited by

Patent Information

Application Number
CN202410586687.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-13
Publication Date
2024-11-15
Estimated Expiration
2044-05-13

AI Technical Summary

Technical Problem

In the prior art, the actual operating conditions of the battery and the laboratory test conditions are large, resulting in a large calculation error in the SOH estimation of the power battery, and a great impact on the calculation accuracy of the charging start SOC and the end SOC.

Method used

The CNN-Transformer model is used to combine data cleaning and multi-dimensional key feature extraction, and the SOH digital twin model is established through transfer learning, and the SOH of the battery cell and battery pack is estimated in real time, so as to realize the coordinated estimation of the cloud-edge-end architecture.

Benefits of technology

Accurate real-time estimation of battery SOH is achieved, which reduces calculation errors and improves the safety and reliability of the battery management system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118483614B_ABST
    Figure CN118483614B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of battery SOH estimation and application technology, and solves the technical problem that the calculated SOH value has a large error due to the large difference between the actual working conditions of the battery and the laboratory test conditions. In particular, it relates to a method, system, application method and electronic device for estimating the SOH of a power battery of a new energy vehicle, including: performing data cleaning on battery data to obtain battery operation data; extracting health indicators during battery operation and performing aggregation and correlation analysis to obtain multi-dimensional key features; inputting the multi-dimensional key features into a CNN-Transformer model, and establishing a SOH digital twin model through transfer learning. The present invention can accurately and effectively estimate the SOH of battery cells and battery packs in real time, can solve the problem of poor real-time performance of deep learning algorithm models and mechanism models, and realize accurate estimation of the SOH of real vehicles, which is of great significance to the safe and reliable operation of battery management systems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of battery SOH estimation and application technology, and in particular to a power battery SOH estimation method, system, application method and electronic equipment for a new energy vehicle. Background Art

[0002] At present, China is in a stage of rapid development of the new energy vehicle industry. Major automobile manufacturers have focused on R&D of new energy vehicle business to ensure their technological leadership in market competition. The state of health (SOH) of electric vehicle power batteries is a core indicator of battery performance and is of great significance to battery management, optimized operation, maintenance and safety.

[0003] At present, the SOH estimation method is mainly based on the empirical model of experimental data or the simpler charging ampere-hour table lookup method. The empirical model requires a lot of experimental tests. The actual working conditions of the battery are quite different from the laboratory test conditions, and the calculated SOH value has a large error. When calculating SOH based on ampere-hours, the starting SOC and ending SOC of charging have a great influence on the calculation accuracy, and there is rarely full charging and full discharging in actual use. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method, system, application method and electronic device for estimating the SOH of a power battery of a new energy vehicle, which solves the technical problem that the calculated SOH value has a large error due to the large difference between the actual working conditions of the battery and the laboratory test conditions.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for estimating the SOH of a power battery of a new energy vehicle, the method comprising the following steps:

[0006] Obtain battery data of new energy vehicles and clean the battery data to obtain high-quality battery operation data;

[0007] Extract the health indicators of the battery during operation from the battery operation data, aggregate the health indicators and perform correlation analysis to obtain the multi-dimensional key features of the battery;

[0008] The multi-dimensional key features are input into the CNN-Transformer model, and then the SOH digital twin model for real-time estimation of the SOH of battery cells and battery packs is established through transfer learning.

[0009] Furthermore, the data cleaning of the battery data is specifically as follows: analyzing the time characteristics of the battery data time series to detect outlier samples, recovering abnormal samples, and using deduplication, filtering, interpolation, and voltage curve resampling methods to remove cloud data noise to obtain high-quality battery operation data.

[0010] Furthermore, the process of extracting the health indicators of the battery during operation from the battery operation data, aggregating and analyzing the correlation of the health indicators to obtain the multi-dimensional key features of the battery includes the following steps:

[0011] Divide the battery operation data into charging data and driving data;

[0012] Extracting voltage-related features, current-related features, and IC curve-related features from charging data to form health indicators during charging;

[0013] Aggregate the health indicators during charging and driving to form a multi-dimensional health indicator;

[0014] The Spearman rank correlation coefficient was used to analyze the correlation between multidimensional health indicators and battery available capacity. S ;

[0015] Select the correlation R S The multidimensional health indicator whose absolute value is closest to 1 is taken as the multidimensional key feature.

[0016] 4. The power battery SOH estimation method according to claim 3, characterized in that the correlation R S The calculation formula is:

[0017]

[0018]

[0019] In the above formula, M is the total number of multidimensional health indicators; X n Y is the available capacity of the battery each time it is discharged; n Input characteristics for each charge and discharge cycle; They are the average values ​​of the health indicators during charging and driving, respectively.

[0020] Furthermore, the CNN-Transformer model is organized into a 2x4 matrix by combining a convolutional neural network CNN and a Transformer model, and convolves the multi-dimensional key features using a 2x2 convolution kernel with a step size of 2 to capture the local correlation of the multi-dimensional key features;

[0021] Then, the convolutional multi-dimensional key features are introduced into the Transformer architecture by adding a position encoding layer, so that the CNN-Transformer model can effectively capture the global dependencies in the multi-dimensional key features.

[0022] Furthermore, the CNN-Transformer model includes:

[0023] An input layer that accepts four input multi-dimensional key features and organizes them into a 2x4 input matrix;

[0024] A convolutional layer that extracts local correlations of multi-dimensional key features of the input to generate a convolutional feature map;

[0025] Introducing position information, adding the convolutional feature map to the position encoding to preserve the relative position relationship between features;

[0026] Transformer model that models global dependencies between input sequences using multi-dimensional key features processed by convolution and positional encoding;

[0027] The output layer generates predictions or classification results for the input sequence.

[0028] Furthermore, the SOH digital twin model for real-time estimation of the SOH of battery cells and battery packs is established through transfer learning, and the specific process includes the following steps:

[0029] Use the source data set in the battery operation data as the input data of the CNN-Transformer model, and obtain a base network A through training;

[0030] Use the target data set in the battery operation data as the input data of the CNN-Transformer model, and obtain a target network B through training;

[0031] Determine the number of layers n that can be migrated in the base network A to obtain the AnB network, that is, retain the first n layers of the base network A and parameters for transfer learning to the target network B;

[0032] The battery operation data are used to train the AnB network and the target network B for N times respectively, and the model with the smallest prediction error between the AnB network and the target network B is selected as the SOH digital twin model.

[0033] The technical solution also provides an application method for the SOH estimation method of power batteries in new energy vehicles. By deploying the SOH digital twin model to the edge device - the vehicle-side controller or Tbox, it is used to estimate the SOH of battery cells and battery packs in real time, and finally interacts with the on-board BMS to realize the practical application of the cloud-edge-end architecture.

[0034] The deployment to the edge device - vehicle controller or Tbox includes the following steps:

[0035] Model conversion: converting the trained SOH digital twin model into an ONNX model;

[0036] Model compression, using the quantization tools provided by ONNX Runtime to adapt to the resource limitations of edge computing devices;

[0037] Edge computing platform selection: If the computing power required to run the software developed by the entire project exceeds the computing power of the chip used by the current vehicle-side BMS, the software needs to be deployed on other vehicle controllers;

[0038] Deployment tool to infer ONNX models using the ONNX Runtime inference engine provided by Microsoft.

[0039] The technical solution also provides a system for implementing the above-mentioned power battery SOH estimation method, including:

[0040] Data processing module, used to obtain high-quality battery operation data of electric vehicles through data cleaning;

[0041] The feature extraction module is used to extract the health indicators of the battery during operation from the battery operation data, and aggregate and analyze the health indicators to obtain the multi-dimensional key features of the battery;

[0042] The model training module is used to initialize the parameters of the CNN-Transformer model, take the multi-dimensional key features as the input of the CNN-Transformer model, and train the SOH digital twin model;

[0043] The result verification module uses the test set to test the SOH digital twin model and compares it with the prediction results of other models to verify the superiority of the SOH digital twin model;

[0044] The system deployment module is used to deploy the SOH digital twin model to embedded devices to realize the cloud-edge-end framework.

[0045] By means of the above technical solution, the present invention provides a method, system, application method and electronic device for estimating the SOH of a power battery of a new energy vehicle, which has at least the following beneficial effects:

[0046] 1. The power battery SOH estimation method proposed in the present invention can accurately and effectively estimate the SOH of battery cells and battery packs in real time, which can solve the problem of poor real-time performance of deep learning algorithm models and mechanism models, and realize accurate estimation of the SOH of real vehicles, which is of great significance to the safe and reliable operation of battery management systems.

[0047] 2. The application method proposed in the present invention solves the problem of poor real-time performance of deep learning algorithm models. By studying the cloud-edge-end collaborative estimation mechanism of battery SOH, an integrated mechanism of cloud training-edge testing-on-board BMS interaction is formed, thereby realizing real-time and accurate estimation of battery SOH under big data and actual vehicle conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0049] Figure 1 This is a flow chart of the power battery SOH estimation method of the present invention;

[0050] Figure 2 It is a principle block diagram of cloud data processing and SOH modeling of the present invention;

[0051] Figure 3 This is a structural block diagram of the cloud-edge-end collaborative architecture of the present invention;

[0052] Figure 4 It is a curve chart of the SOH estimation results of the actual vehicle data of the present invention. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific implementation methods, so that the implementation process of how the present application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0054] At present, China is in a stage of rapid development of the new energy vehicle industry. Major automobile manufacturers have focused on R&D of new energy vehicle business to ensure their technological leadership in market competition. With the development of new energy vehicles and energy digitalization technology in my country, intelligent connected vehicles have been regarded as one of the important directions for the future transformation and upgrading of the automobile industry. In order to strengthen the safety supervision of new energy enterprises and vehicles, it is required to establish files for each new energy vehicle sold and track its use and maintenance through data. Big data and artificial intelligence technologies have shown good application prospects in the field of new energy vehicles. The state of health (SOH) of electric vehicle power batteries is a core indicator of battery performance and is of great significance to battery management, optimized operation, maintenance and safety.

[0055] At present, the SOH estimation method is mainly based on the empirical model of experimental data or the simpler charging ampere-hour table lookup method. The empirical model requires a lot of experimental tests. The actual working conditions of the battery are quite different from the laboratory test conditions, and the calculated SOH value has a large error. When calculating SOH based on ampere-hours, the starting SOC and ending SOC of charging have a great influence on the calculation accuracy, and there is rarely full charging and full discharging in actual use.

[0056] Due to the large difference between the actual battery working conditions and the laboratory test conditions, the calculated SOH value has a large error. Please refer to Figure 1-Figure 2 This embodiment proposes a method for estimating the SOH of a power battery of a new energy vehicle. This embodiment can accurately and effectively estimate the SOH of a battery cell and a battery pack in real time, solve the problem of poor real-time performance of deep learning algorithm models and mechanism models, and realize accurate estimation of the SOH of a real vehicle. The method includes the following steps:

[0057] S1. Obtain battery data of new energy vehicles, and perform data cleaning on the battery data to obtain high-quality battery operation data; wherein, data cleaning of the battery data is specifically: analyzing the time characteristics of the battery data time series to detect outlier samples, recovering abnormal samples, and using deduplication, filtering, interpolation, and voltage curve resampling methods to remove cloud data noise to obtain high-quality battery operation data, wherein deduplication, filtering, and interpolation methods are conventional technical means in data cleaning technology, and will not be described in detail in this embodiment. In this embodiment, a voltage curve resampling method is added on the basis of the conventional data cleaning method to fit the data points corresponding to the time characteristics in the battery data, thereby smoothing the data.

[0058] Among them, the voltage curve resampling method adopts Savitzky-Golay filtering, which is a smoothing method based on polynomial fitting. It smoothes the data by performing polynomial fitting on the data points in the data window. Here we use 11th-order fitting.

[0059] S2. Extract the health indicators of the battery during operation from the battery operation data, aggregate the health indicators and perform correlation analysis to obtain the multi-dimensional key features of the battery. The specific process includes the following steps:

[0060] S21. Divide the battery operation data into charging data and driving data. Specifically, in order to obtain a more comprehensive health indicator, the battery operation data is divided into charging data and driving data, and the two are processed separately.

[0061] S22. Extract voltage-related features, current-related features, temperature-related features, and IC curve-related features from the charging data to form health indicators during the charging process; for relatively stable charging data, extract voltage-related features, current-related features, and IC curve-related features, respectively.

[0062] Voltage-related features include the time interval of equal charge voltage rise (TI-ECVR), the charge capacity of equal charge voltage rise (CC-ECVR), and the slope of the charge voltage curve (SCVC). However, the data of TI-ECVR and CC-ECVR are slightly different after standardization, and some data points are almost the same. The area under the charging voltage curve (ACVC) can be used instead of the CC-ECVR value as a feature to estimate the battery SOH. The voltage segment range is selected according to the battery OCV characteristic curve. The time intervals of equal charge voltage rise TI-ECVR in each stage are Δt1-Δt6 respectively; the slopes of the charging voltage curve SCVC are K1-K6 respectively; and ACVC are A1-A6 respectively.

[0063] The calculation formula of the slope of the charging voltage curve SCVC is:

[0064] K i =Δt i / |U i -U0|

[0065] Among them, K i is the slope of the charging voltage curve; Δt i is the time interval of each charging stage; U0 is the initial voltage point selected in this paper; U i For other voltage points; i=1,2,3.

[0066] Features related to the IC curve include the IC peak point (IC peak, ICP), the peak point location (IC peak location, ICPL) and the area in the IC curve (Area in the IC curve, AICC).

[0067] The current-related features include extracting the time interval of equal charge current drop (TI-ECCD), the charge capacity of equal charge current drop (CC-ECCD) and the slope of the charge current curve (SCCC) from the CV charging stage to characterize the aging phenomenon of the battery. During the entire CV period, TI-ECCD is Δt7, CC-ECCD is Q7, and SCCC is K7. The calculation formula of SCCC in this embodiment is as follows:

[0068] K7=|I2-I1| / Δt7

[0069] Among them, I1 is the current value at time t, and I2 is the current value at time t+Δt7.

[0070] The driving data is segmented, and the charging segment data with an SOC of 35% to 55% is extracted. The corresponding segment feature values ​​are extracted from the vehicle speed, voltage, current, and vehicle status data to form a health indicator during driving.

[0071] S23. Use the cluster analysis k-means algorithm to aggregate the health indicators during charging and the health indicators during driving to form a multi-dimensional health indicator. The implementation process of the aggregation is as follows:

[0072] First, randomly select K objects as the initial cluster centers, then calculate the norm of other samples to the cluster center point and minimize it, that is:

[0073] min||x (i) -μ k || 2

[0074] Among them, x (i) is the i-th variable in sample X; μ k is the corresponding cluster center point.

[0075] Set c by moving the cluster centroid position (i) and μ kReduce the cost function value J to obtain a suitable clustering result, c (i) For x (i) The distance to the cluster center is:

[0076]

[0077] Where m is the total number of variables i; x (i) is the i-th variable in sample X; Represents the cluster c to which the i-th sample point belongs (i) The center point of .

[0078] S24. Use Spearman rank correlation coefficient to analyze the correlation between multidimensional health indicators and battery available capacity R S ; Since different features have a great influence on the prediction accuracy of the model, high-quality feature information can reduce the complexity of the model to a certain extent and improve the prediction accuracy of the model. If the relationship between a feature and the response variable is nonlinear, some machine learning algorithms can be used to select features as inputs to the prediction model based on the contribution of the features to the label vector. The Spearman rank correlation coefficient can be used to analyze the correlation between multidimensional health indicators and the available capacity of the battery. Among them, the correlation R S The calculation formula is:

[0079]

[0080]

[0081] In the above formula, M is the total number of multidimensional health indicators; X n Y is the available capacity of the battery each time it is discharged; n Input characteristics for each charge and discharge cycle; They are the average values ​​of the health indicators during charging and driving, respectively.

[0082] S25. Select correlation R S The multidimensional health index with the absolute value closest to 1 is taken as the multidimensional key feature, among which the correlation R S The closer the absolute value of is to 1, the higher the correlation between the multidimensional health indicator and the available battery capacity, and the closer their relationship is. S The closer the absolute value of is to 0, the lower the correlation between the two.

[0083] This embodiment processes massive multi-source heterogeneous real vehicle data on the cloud of new energy vehicles, extracts multi-dimensional key features related to battery operating conditions through data cleaning and correlation analysis, uses deep learning algorithms to perform SOH digital twin modeling, and explains the relationship between battery operation and SOH under actual working conditions through multi-dimensional and multi-scale real vehicle and experimental data.

[0084] S3. Input the multi-dimensional key features into the CNN-Transformer model, and then establish the SOH digital twin model for real-time estimation of battery cell and battery pack SOH through transfer learning; Convolutional neural network CNN is a special type of feedforward neural network, and its network hierarchy usually includes convolution layer, pooling layer and fully connected layer. CNN processes and analyzes image data by simulating the natural visual perception mechanism of organisms, so that the network can automatically recognize complex features in the image.

[0085] In this embodiment, the CNN-Transformer model is organized into a 2x4 matrix by combining a convolutional neural network CNN and a Transformer model, and convolves the multi-dimensional key features by using a 2x2 convolution kernel with a step size of 2 to capture the local correlation of the multi-dimensional key features;

[0086] Then, the convolutional multi-dimensional key features are introduced into the Transformer architecture by adding a position encoding layer, so that the CNN-Transformer model can effectively capture the global dependencies in the multi-dimensional key features.

[0087] The CNN-Transformer model includes: an input layer for accepting four input multidimensional key features and organizing them into a 2x4 input matrix; a convolution operation is performed on the input matrix using a 2x2 convolution kernel with a step size of 2. This step is intended to extract the local correlation of the input multidimensional key features, thereby generating a convolution layer of a convolution feature map. In order to introduce position information, the convolution feature map is added to the position encoding to retain the relative position relationship between the features. The position encoding is usually calculated by sine and cosine functions. The multidimensional key features after convolution and position encoding are input into the Transformer model. The Transformer model consists of a multi-layer self-attention mechanism and a fully connected feedforward network, which can effectively model the global dependency between input sequences. Finally, through the output layer of the Transformer model, an output layer for predicting or classifying the input sequence is generated.

[0088] Among them, as needs to be disclosed, the convolution operation of the convolutional neural network CNN can be expressed as:

[0089]

[0090] In the above formula, the input matrix is ​​I; the filter is K; * represents the convolution operation; i and j represent pixel points; m and n represent the index variables that traverse the input matrix I.

[0091] The Transformer model consists of a position encoding layer, an encoder-decoder, a residual connection layer, and a normalization module. Each encoder-decoder consists of two sublayers: a self-attention layer and a feedforward neural network.

[0092] The mathematical formula of the self-attention mechanism is:

[0093]

[0094] In the above formula, Q, K and V represent the matrices of query, key and value respectively; d n is the dimension of the key vector, used to scale the dot product result to prevent the gradient from disappearing; K T represents the transpose of matrix K; softmax is the activation function.

[0095] When multiple self-attention layers are stacked together, each layer receives the output of the previous layer as input and produces a new output. This stacking structure enables the model to gradually build and refine its understanding of the input sequence. The multi-attention mechanism formula is as follows:

[0096] MultiHedad(Q,K,V)=Concat(head1,…,head i )W O

[0097]

[0098] In the above formula, Concat represents the concatenation operation; Q, K, and V represent the query, key, and value matrices respectively; W O , are all trainable weight matrices.

[0099] In step S3, a SOH digital twin model for real-time estimation of the SOH of battery cells and battery packs is established through transfer learning. The specific process includes the following steps:

[0100] S31, using the source data set in the battery operation data as the input data of the CNN-Transformer model, and obtaining a base network A through training;

[0101] S32, using the target data set in the battery operation data as the input data of the CNN-Transformer model, and obtaining a target network B through training;

[0102] S33, determining the number of layers n that can be migrated in the base network A to obtain the AnB network, that is, retaining the first n layers of the base network A and parameters for transfer learning to the target network B;

[0103] S34. Use the battery operation data to train the AnB network and the target network B N times respectively, and select the model with the smallest error in the prediction results between the AnB network and the target network B as the SOH digital twin model.

[0104] The power battery SOH estimation method proposed in the present invention can accurately and effectively estimate the SOH of battery cells and battery packs in real time, can solve the problem of poor real-time performance of deep learning algorithm models and mechanism models, and realize accurate estimation of the SOH of real vehicles, which is of great significance to the safe and reliable operation of battery management systems.

[0105] like Figure 3 As shown, in order to solve the problem of poor real-time performance of deep learning algorithm models, this embodiment proposes an application method for the SOH estimation method of power batteries of new energy vehicles. By studying the cloud-edge-end collaborative estimation mechanism of battery SOH, an integrated mechanism of cloud training-edge-end testing-on-board BMS interaction is formed, and a battery SOH digital twin model is established to achieve real-time and accurate estimation of battery SOH under big data and actual vehicle conditions. By deploying the SOH digital twin model to the edge device-vehicle controller or Tbox, it is used to estimate the SOH of battery cells and battery packs in real time, and finally interacts with the on-board BMS to realize the practical application of the cloud-edge-end architecture. Specifically, firstly, empirical model modeling and calculation and data-driven model modeling and training are performed in the cloud, and then the constructed SOH digital twin model is deployed to the embedded device at the edge for model training and fusion, and finally the prediction results of the SOH digital twin model are passed to the on-board BMS for data interaction.

[0106] The deployment to the edge device - vehicle controller or Tbox includes the following steps:

[0107] Model conversion: converting the trained SOH digital twin model into an ONNX model;

[0108] Model compression, using the quantization tools provided by ONNX Runtime to adapt to the resource limitations of edge computing devices;

[0109] Edge computing platform selection: If the computing power required to run the software developed by the entire project exceeds the computing power of the chip used by the current vehicle-side BMS, the software needs to be deployed on other vehicle controllers;

[0110] Deployment tool to infer ONNX models using the ONNX Runtime inference engine provided by Microsoft.

[0111] In this embodiment, the experimental data used comes from the actual vehicle data of a certain automobile company, which records the actual vehicle speed, cumulative mileage, voltage, current, gear and other information. After data cleaning, these data information are subjected to feature value extraction and correlation analysis, and then input into the CNN-Transformer model for training and testing, and then the SOH digital twin model of the battery cell and battery pack is established through transfer learning.

[0112] The simulation results show that Figure 4 As shown, the application method proposed in the present invention can accurately and effectively estimate the battery SOH.

[0113] Corresponding to the power battery SOH estimation method provided in the above-mentioned embodiment, the present embodiment also provides a system for the power battery SOH estimation method. Since the power battery SOH estimation system provided in the present embodiment corresponds to the power battery SOH estimation method provided in the above-mentioned embodiment, the implementation method of the aforementioned power battery SOH estimation method is also applicable to the power battery SOH estimation system provided in the present embodiment, and will not be described in detail in the present embodiment.

[0114] On the basis of the above embodiments, the present invention further provides a power battery SOH estimation system for new energy vehicles based on a cloud-edge-end collaborative artificial intelligence big model, which is used to support the power battery SOH estimation method for new energy vehicles based on a cloud-edge-end collaborative artificial intelligence big model of the above embodiments, including:

[0115] Data processing module, used to obtain high-quality battery operation data of electric vehicles through data cleaning;

[0116] The feature extraction module is used to extract the health indicators of the battery during operation from the battery operation data, and aggregate and analyze the health indicators to obtain the multi-dimensional key features of the battery;

[0117] The model training module is used to initialize the parameters of the CNN-Transformer model, take the multi-dimensional key features as the input of the CNN-Transformer model, and train the SOH digital twin model;

[0118] The result verification module uses the test set to test the SOH digital twin model and compares it with the prediction results of other models to verify the superiority of the SOH digital twin model;

[0119] The system deployment module is used to deploy the SOH digital twin model to embedded devices to realize the cloud-edge-end framework.

[0120] It should be noted that the system provided in the above embodiment, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0121] The present invention also provides an electronic device, which is a physical device and includes: a processor, a memory, and the processor, an embedded device, a vehicle-mounted BMS system, and the memory are communicatively connected with the processor; the memory is used to store executable instructions executed by at least one processor, and the processor is used to execute the executable instructions to implement a power battery SOH estimation method for new energy vehicles.

[0122] This embodiment also provides a computer device, which includes a processor, a memory, a network interface, an input device and a display screen connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory includes a storage medium and an internal memory. The storage medium can be a non-volatile storage medium or a volatile storage medium. The storage medium stores an operating system and can also store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can implement the power battery SOH estimation method. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the storage medium. The internal memory can also store computer-readable instructions, which can enable the processor to execute the power battery SOH estimation method when the computer-readable instructions are executed by the processor. The network interface of the computer device is used to communicate with an external server through a network connection. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0123] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, so the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0124] The above implementation methods have been described in detail. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for estimating SOH of a power battery of a new energy vehicle, characterized in that: The method comprises the following steps: Obtain battery data of new energy vehicles and clean the battery data to obtain high-quality battery operation data; The health indicators of the battery during operation are extracted from the battery operation data, and the health indicators are aggregated and analyzed for correlation to obtain the multi-dimensional key features of the battery. The specific process includes the following steps: Divide the battery operation data into charging data and driving data; The features related to voltage, current and IC curve are extracted from the charging data to form health indicators during the charging process, where: Voltage-related features include the time interval of equal charging voltage rise, the area under the charging voltage curve, and the slope of the charging voltage curve as features to estimate the battery SOH; Features associated with the IC curve include the IC peak point, the location of the peak point, and the area within the IC curve; The current-related features include extracting the time interval of equal charging current decline, the charging capacity of equal charging current decline and the slope of charging current curve from the CV charging stage to characterize the battery aging phenomenon; The driving data is segmented, and the charging segment data with SOC of 35% to 55% is extracted. The corresponding segment feature values ​​are extracted from the vehicle speed, voltage, current, and vehicle status data to form the health index during driving; Aggregate the health indicators during charging and driving to form a multi-dimensional health indicator; The Spearman rank correlation coefficient was used to analyze the correlation between multidimensional health indicators and battery available capacity. , correlation The calculation formula is: ; In the above formula, M is the total number of multidimensional health indicators; The available capacity of the battery per discharge; Input characteristics for each charge and discharge cycle; are the average values ​​of the health indicators during charging and driving, respectively; Select relevance The multidimensional health index whose absolute value is closest to 1 is taken as the multidimensional key feature; The multi-dimensional key features are input into the CNN-Transformer model, and then the SOH digital twin model for real-time estimation of battery cell and battery pack SOH is established through transfer learning. The CNN-Transformer model is organized into a 2x4 matrix by combining the convolutional neural network CNN and the Transformer model, and the multi-dimensional key features are convolved by using a 2x2 convolution kernel with a step size of 2 to capture the local correlation of the multi-dimensional key features; Then, the convolutional multi-dimensional key features are introduced into the Transformer architecture by adding a position encoding layer, so that the CNN-Transformer model can effectively capture the global dependencies in the multi-dimensional key features; The CNN-Transformer model includes: an input layer for accepting four input multi-dimensional key features and organizing them into a 2x4 input matrix; A convolutional layer that extracts local correlations of multi-dimensional key features of the input to generate a convolutional feature map; Introducing position information, adding the convolutional feature map to the position encoding to preserve the relative position relationship between features; Transformer model that models the global dependencies between input sequences using multi-dimensional key features processed by convolution and positional encoding; An output layer that generates predictions or classification results for the input sequence; The SOH digital twin model for real-time estimation of the SOH of battery cells and battery packs is established through transfer learning, and the specific process includes the following steps: Use the source data set in the battery operation data as the input data of the CNN-Transformer model, and obtain a base network A through training; Use the target data set in the battery operation data as the input data of the CNN-Transformer model, and obtain a target network B through training; Determine the number of layers n that can be migrated in the base network A to obtain the AnB network, that is, retain the first n layers of the base network A and parameters for transfer learning to the target network B; The battery operation data are used to train the AnB network and the target network B for N times respectively, and the model with the smallest prediction error between the AnB network and the target network B is selected as the SOH digital twin model.

2. The power battery SOH estimation method according to claim 1, characterized in that: The data cleaning of the battery data is specifically as follows: analyzing the time characteristics of the battery data time series to detect outlier samples, recovering abnormal samples, and using deduplication, filtering, interpolation, and voltage curve resampling methods to remove cloud data noise to obtain high-quality battery operation data.

3. The application method of the power battery SOH estimation method according to any one of claims 1-2, characterized in that: Deploy the SOH digital twin model to the edge device - vehicle-side controller or Tbox to estimate the SOH of battery cells and battery packs in real time, and finally interact with the on-board BMS to realize the practical application of the cloud-edge-end architecture; The implementation of deployment to the edge device - vehicle controller or Tbox includes the following steps: Model conversion: converting the trained SOH digital twin model into an ONNX model; Model compression, using the quantization tools provided by ONNX Runtime to adapt to the resource limitations of edge computing devices; Edge computing platform selection: If the computing power required to run the software developed by the entire project exceeds the computing power of the chip used by the current vehicle-side BMS, the software needs to be deployed on other vehicle controllers; Deployment tool that uses the ONNX Runtime inference engine provided by Microsoft to infer ONNX models.

4. A system for implementing the power battery SOH estimation method according to any one of claims 1 to 2, characterized in that: include: Data processing module, used to obtain high-quality battery operation data of electric vehicles through data cleaning; The feature extraction module is used to extract the health indicators of the battery during operation from the battery operation data, and aggregate and analyze the correlation of the health indicators to obtain the multi-dimensional key features of the battery; The model training module is used to initialize the parameters of the CNN-Transformer model, take the multi-dimensional key features as the input of the CNN-Transformer model, and train the SOH digital twin model; The result verification module uses the test set to test the SOH digital twin model and compares it with the prediction results of other models to verify the superiority of the SOH digital twin model; The system deployment module is used to deploy the SOH digital twin model to embedded devices to realize the cloud-edge-end framework.

5. An electronic device, wherein the electronic device is a physical device, and the electronic device comprises: A processor and a memory, wherein the processor, the embedded device, the vehicle-mounted BMS system, the memory and the processor are communicatively connected; The memory is used to store at least one executable instruction executed by the processor, and the processor is used to execute the executable instruction to implement the power battery SOH estimation method for a new energy vehicle as described in any one of claims 1 to 2.

Citation Information

Patent Citations

  • Electric vehicle lithium battery SOH estimation method based on deep network self-adaption

    CN113740736A

  • Transform-based energy storage battery life prediction algorithm

    CN115267575A

  • Lithium battery health state estimation method based on feature selection and time sequence attention

    CN116068399A

  • Lithium battery health state estimation method integrating unsupervised learning and supervised learning

    CN116699410A

  • Sodium ion battery cloud SOH estimation and residual life prediction method, device and program

    CN116754960A