Electric vehicle battery life prediction method based on deep learning
By collecting multi-source data and building an improved Transformer model, combining the attention mechanism and multi-layer perceptron, and adopting an adaptive learning rate and adversarial training strategy, the problems of incomplete consideration of factors and insufficient adaptability in existing battery life prediction methods are solved, and more accurate battery life prediction and battery maintenance guidance are achieved.
Patent Information
- Application Number
- CN202510862337.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-25
- Publication Date
- 2025-10-10
AI Technical Summary
Existing battery life prediction methods, when utilizing deep learning, fail to fully consider the various influencing factors during actual battery use, resulting in inaccurate prediction results, and the model training and optimization lack specificity, making it difficult to adapt to the dynamic changes in battery data.
Collect multi-source data, build an improved Transformer model and introduce an attention mechanism enhancement module. Combined with a multi-layer perceptron, an adaptive learning rate adjustment algorithm and adversarial training are used, and incremental training is performed regularly to generate battery life warning information.
It improves the accuracy and precision of battery life prediction, enhances the generalization ability of the model, provides timely battery maintenance guidance, improves the safety of electric vehicles and reduces their use costs.
Smart Images
Figure CN120761865A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to the technical field of electric vehicle battery management, and in particular to an electric vehicle battery life prediction method based on deep learning. BACKGROUND
[0002] With the wide application of electric vehicles, the battery as the core component of the electric vehicle directly affects the use performance and economy of the electric vehicle. Accurate prediction of the battery life is of great significance for reasonable planning of battery replacement, improvement of electric vehicle safety and reduction of use cost.
[0003] At present, the commonly used battery life prediction methods mainly include a method based on an electrochemical model and a data-driven method. The method based on the electrochemical model can be modeled from the chemical reaction mechanism inside the battery, but the model parameters are difficult to accurately obtain, the calculation complexity is high, and the generality for different types of batteries is poor. The data-driven method, especially the traditional machine learning method, often has limited prediction accuracy when dealing with complex battery data, as it is difficult to extract deep features. In recent years, deep learning has shown great ability in data processing and feature extraction, but the existing battery life prediction methods based on deep learning mostly only use single type of data for modeling, and cannot fully consider the various influencing factors of the battery in the actual use process, resulting in inaccurate prediction results. At the same time, the training and optimization process of the model lacks pertinence and is difficult to adapt to the dynamic change characteristics of the battery data. Therefore, we propose an electric vehicle battery life prediction method based on deep learning. SUMMARY
[0004] The application aims to solve the problems in the prior art and provides an electric vehicle battery life prediction method based on deep learning.
[0005] In order to achieve the above-mentioned purpose, the application adopts the following technical solutions:
[0006] An electric vehicle battery life prediction method based on deep learning comprises the following steps:
[0007] S1, collecting multi-source data of the electric vehicle battery, including battery voltage, current, temperature, charge-discharge times, charge-discharge depth, battery health state historical data, and road condition information and driving habit data in the driving process of the electric vehicle;
[0008] S2, cleaning, interpolating and normalizing the collected data;
[0009] S3, constructing an improved Transformer model, introducing an attention mechanism enhancement module in the encoder layer, and connecting a multi-layer perception after the output layer;
[0010] S4. Divide the preprocessed data into training set, validation set and test set, use the weighted sum of mean square error and mean absolute error as the loss function, and use the adaptive learning rate adjustment algorithm to train the model;
[0011] S5. Use adversarial training during model training and perform incremental training on the model regularly.
[0012] S6. Input the real-time collected and pre-processed data into the trained model, output the predicted value of the remaining battery life, and generate battery life warning information based on the predicted value.
[0013] Preferably, in step S2, the collected data is cleaned to remove outliers and missing values in the data, and the missing values are filled using an interpolation algorithm based on a time series.
[0014] Preferably, in step S3, the input of the improved Transformer model is a preprocessed multi-source data sequence, and the sequence length is determined according to the time resolution of the battery data and the prediction requirements.
[0015] Preferably, in step S4, the data sets of the training set, validation set and test set are divided into 50% training set, 10% validation set and 40% test set respectively.
[0016] Preferably, in step S4, the loss function is the weighted sum of the mean square error and the mean absolute error, and the formula is: Loss = α × MSE + β × MAE, where α and β are weight coefficients, and the optimal value is determined by cross-validation, MSE is the loss function is the mean square error, MAE is the mean absolute error, and Loss is the loss function is the weighted sum of the mean square error and the mean absolute error.
[0017] Preferably, in step S4, an adaptive learning rate adjustment algorithm is used to train the model. During the training process, the learning rate is dynamically adjusted according to the loss value of the validation set. When the loss value of the validation set no longer decreases for 5 consecutive cycles, the learning rate is multiplied by 0.1 for decay, and the total number of training cycles is set to 100.
[0018] Preferably, in step S5, adversarial training introduces a discriminator network, which is used to determine whether the input data comes from the real data distribution or the data distribution generated by the model, and the adversarial training is used to encourage the model to learn a more real and accurate data feature distribution.
[0019] Preferably, in step S5, the model is incrementally trained regularly, that is, new battery data is collected regularly, and parameters of the trained model are updated to adapt to changes in battery performance over time.
[0020] Preferably, in step S6, battery life warning information is generated based on the predicted value, and a warning prompt is issued when the predicted remaining service life is lower than a set threshold.
[0021] Preferably, in step S1, the road condition information includes slope and road surface type, the driving habit data includes the frequency of rapid acceleration and deceleration, the voltage acquisition sensor uses a TIBQ40Z80 chip with a sampling rate of 10Hz, and the temperature acquisition sensor model is a DS18B20 digital sensor.
[0022] Compared with the prior art, the present invention has the following beneficial effects:
[0023] 1. The present invention fully considers the various influencing factors of the battery during actual use by collecting multi-source data. Compared with the prediction method that only uses a single type of data, it can more comprehensively reflect the actual state of the battery and improve the accuracy of the prediction;
[0024] 2. In this invention, the improved Transformer model combined with the attention mechanism enhancement module and the multi-layer perceptron can effectively extract deep features from battery data. At the same time, the adaptive learning rate adjustment algorithm and optimized loss function design make model training more efficient and further improve prediction accuracy.
[0025] 3. In the present invention, adversarial training and incremental training strategies are adopted. On the one hand, the model's ability to learn the distribution of real data features is improved, and on the other hand, the model can adapt to the dynamic changes in battery performance, thereby enhancing the model's generalization ability and prediction reliability.
[0026] 4. In the present invention, by generating battery life warning information, timely battery maintenance guidance is provided to users, which helps to improve the safety of electric vehicles and reduce the cost of use. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 This is a flowchart of a method for predicting electric vehicle battery life based on deep learning proposed by the present invention. DETAILED DESCRIPTION
[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.
[0029] Reference Figure 1 , a method for predicting the battery life of an electric vehicle based on deep learning, comprising the following steps:
[0030] S1. Collect multi-source data of electric vehicle batteries, including battery voltage, current, temperature, charge and discharge times, charge and discharge depth, battery health status history data, as well as road condition information and driving habit data during electric vehicle driving;
[0031] S2, cleaning, interpolation and normalization of the collected data;
[0032] S3. Build an improved Transformer model, introduce an attention mechanism enhancement module in the encoder layer, and connect a multi-layer perceptron after the output layer;
[0033] S4. Divide the preprocessed data into training set, validation set and test set, use the weighted sum of mean square error and mean absolute error as the loss function, and use the adaptive learning rate adjustment algorithm to train the model;
[0034] S5. Use adversarial training during model training and perform incremental training on the model regularly.
[0035] S6. Input the real-time collected and pre-processed data into the trained model, output the predicted value of the remaining battery life, and generate battery life warning information based on the predicted value.
[0036] In step S2, the collected data is cleaned to remove outliers and missing values in the data, and the missing values are filled using an interpolation algorithm based on time series.
[0037] In step S3, the input of the improved Transformer model is the preprocessed multi-source data sequence, and the sequence length is determined according to the time resolution of the battery data and the prediction requirements.
[0038] In step S4, the training set, validation set and test set data sets are divided into 50% training set, 10% validation set and 40% test set respectively.
[0039] In step S4, the loss function is the weighted sum of the mean square error and the mean absolute error. The formula is: Loss = α × MSE + β × MAE, where α and β are weight coefficients, and the optimal value is determined by cross-validation. MSE is the loss function, MAE is the mean absolute error, and Loss is the loss function, which is the weighted sum of the mean square error and the mean absolute error.
[0040] In step S4, the model is trained using an adaptive learning rate adjustment algorithm. During the training process, the learning rate is dynamically adjusted according to the loss value of the validation set. When the loss value of the validation set does not decrease for five consecutive cycles, the learning rate is multiplied by 0.1 for decay. The total number of training cycles is set to 100.
[0041] In step S5, adversarial training introduces a discriminator network, which is used to determine whether the input data comes from the real data distribution or the data distribution generated by the model. Through adversarial training, the model is prompted to learn a more realistic and accurate data feature distribution.
[0042] In step S5, the model is incrementally trained regularly, which means that new battery data is collected regularly and the parameters of the trained model are updated to adapt to changes in battery performance over time.
[0043] In step S6, a battery life warning message is generated based on the predicted value, and a warning prompt is issued when the predicted remaining service life is lower than a set threshold.
[0044] In step S1, the road condition information includes slope and road type, the driving habit data includes the frequency of sudden acceleration and deceleration, the voltage acquisition sensor uses a TIBQ40Z80 chip with a sampling rate of 10Hz, and the temperature acquisition sensor model is a DS18B20 digital sensor.
[0045] By collecting multi-source data, the invention fully considers the various influencing factors of the battery during actual use. Compared with the prediction method that only uses a single type of data, it can more comprehensively reflect the actual state of the battery and improve the accuracy of the prediction; the improved Transformer model combined with the attention mechanism enhancement module and the multi-layer perceptron can effectively extract deep-level features in the battery data. At the same time, the adaptive learning rate adjustment algorithm and the optimized loss function design make the model training more efficient and further improve the prediction accuracy; the use of adversarial training and incremental training strategies, on the one hand, improves the model's ability to learn the distribution of real data features, and on the other hand, enables the model to adapt to the dynamic changes in battery performance, thereby enhancing the model's generalization ability and the reliability of the prediction; by generating battery life warning information, timely battery maintenance guidance is provided to users, which helps to improve the safety of electric vehicles and reduce their use costs.
[0046] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.
Claims
1. A method for predicting the battery life of an electric vehicle based on deep learning, characterized in that: The following steps are involved: S1. Collect multi-source data of electric vehicle batteries, including battery voltage, current, temperature, charge and discharge times, charge and discharge depth, battery health status history data, as well as road condition information and driving habit data during electric vehicle driving; S2, cleaning, interpolation and normalization of the collected data; S3. Build an improved Transformer model, introduce an attention mechanism enhancement module in the encoder layer, and connect a multi-layer perceptron after the output layer; S4. Divide the preprocessed data into training set, validation set and test set, use the weighted sum of mean square error and mean absolute error as the loss function, and use the adaptive learning rate adjustment algorithm to train the model; S5. Use adversarial training during model training and perform incremental training on the model regularly. S6. Input the real-time collected and pre-processed data into the trained model, output the predicted value of the remaining battery life, and generate battery life warning information based on the predicted value.
2. The electric vehicle battery life prediction method based on deep learning according to claim 1, characterized in that: In step S2, the collected data is cleaned to remove outliers and missing values in the data, and the missing values are filled using an interpolation algorithm based on time series.
3. The electric vehicle battery life prediction method based on deep learning according to claim 1, characterized in that: In step S3, the input of the improved Transformer model is a preprocessed multi-source data sequence, and the sequence length is determined according to the time resolution of the battery data and the prediction requirements.
4. The electric vehicle battery life prediction method based on deep learning according to claim 1, characterized in that: In step S4, the data sets of the training set, the validation set and the test set are divided into 50% training set, 10% validation set and 40% test set ratios respectively.
5. The electric vehicle battery life prediction method based on deep learning according to claim 1, characterized in that: In step S4, the loss function is the weighted sum of the mean square error and the mean absolute error, and the formula is: Loss = α × MSE + β × MAE, where α and β are weight coefficients, and the optimal value is determined by cross-validation. MSE is the loss function, MAE is the mean absolute error, and Loss is the loss function, which is the weighted sum of the mean square error and the mean absolute error.
6. The electric vehicle battery life prediction method based on deep learning according to claim 1, characterized in that: In step S4, the model is trained using an adaptive learning rate adjustment algorithm. During the training process, the learning rate is dynamically adjusted according to the loss value of the validation set. When the loss value of the validation set no longer decreases for five consecutive cycles, the learning rate is multiplied by 0.1 for decay. The total number of training cycles is set to 100.
7. The electric vehicle battery life prediction method based on deep learning according to claim 1, characterized in that: In step S5, adversarial training introduces a discriminator network, which is used to determine whether the input data comes from the real data distribution or the data distribution generated by the model. Through adversarial training, the model is prompted to learn a more real and accurate data feature distribution.
8. The electric vehicle battery life prediction method based on deep learning according to claim 1, characterized in that: In step S5, the model is incrementally trained regularly, which means that new battery data is collected regularly and parameters of the trained model are updated to adapt to changes in battery performance over time.
9. The electric vehicle battery life prediction method based on deep learning according to claim 1, characterized in that: In step S6, a battery life warning message is generated based on the predicted value, and a warning prompt is issued when the predicted remaining service life is lower than a set threshold.
10. The electric vehicle battery life prediction method based on deep learning according to claim 1, characterized in that: In step S1, the road condition information includes slope and road surface type, the driving habit data includes the frequency of rapid acceleration and deceleration, the voltage acquisition sensor uses a TIBQ40Z80 chip with a sampling rate of 10 Hz, and the temperature acquisition sensor is a DS18B20 digital sensor.