Electric vehicle lithium battery SOC estimation method based on deep learning
By constructing the CNN-GRU-LSTM neural network model, the problems of low computing efficiency and high resource consumption in lithium battery SOC estimation are solved, and higher prediction accuracy and faster processing speed are achieved, supporting the real-time and effective management of lithium batteries in electric vehicles.
Patent Information
- Application Number
- CN202411948020.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-27
- Publication Date
- 2025-04-15
AI Technical Summary
The existing lithium battery SOC estimation methods have problems such as low computational efficiency, gradient disappearance, and storage data occupying space, especially the gated cycle unit (GRU) and long short-term memory network (LSTM) are relatively high in parameter and memory resource consumption.
A CNN-GRU-LSTM neural network model is constructed based on deep learning. The Gated Recurrent Unit (GRU) focuses on early time-related information, and combines the long-term memory ability of the long-term short-term memory network (LSTM) to improve the SOC prediction accuracy and realize real-time estimation.
This method can significantly improve the prediction accuracy of lithium battery SOC, reduce the consumption of computing resources, achieve faster processing speed and smaller memory usage, and support the effective management of lithium batteries in electric vehicles.
Smart Images
Figure CN119827997A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of new energy vehicles, and in particular to a method for estimating the SOC of a lithium battery of an electric vehicle based on deep learning. Background Art
[0002] In recent years, electric vehicles have gradually been widely used in the automotive industry due to their significant advantages such as zero emissions, high efficiency, and low noise. However, with its rapid development, it also faces a series of challenges such as range anxiety, uneven distribution of charging facilities, and lithium battery safety. As the core power source of electric vehicles, the performance status of lithium batteries has become an important indicator for measuring the performance and user experience of electric vehicles. Usually, drivers need to estimate the driving range of electric vehicles by the SOC of lithium batteries. Therefore, accurate state of charge (SOC) estimation of lithium batteries is crucial to prevent lithium batteries from overcharging and discharging, extending lithium battery life, and understanding driving mileage.
[0003] At present, the SOC estimation methods for electric vehicle power lithium batteries are mainly divided into model-based estimation methods, data-driven estimation methods and model-data fusion-based estimation methods. In practical applications, neural network estimation methods have become popular methods for SOC estimation due to their advantages in deeply mining the nonlinear relationship between data, including recurrent neural networks (RNN), long short-term memory networks (LSTM), gated recurrent units (GRU), etc. In the process of using gated recurrent units (GRU) to estimate SOC, there are problems such as low computational efficiency, gradient disappearance, and space occupied by stored data; although the use of long short-term memory networks (LSTM) can effectively solve the shortcomings of traditional recurrent neural networks (RNN) in the long-term dependency problem and show good performance on large-scale data sets, compared with gated recurrent units (GRU), LSTM requires more parameters and memory resources. Therefore, the combination of LSTM and GRU can be an effective solution, which can simultaneously deal with long-term dependency problems and be more efficient and time-saving in implementation. Summary of the invention
[0004] To solve the above problems, this application proposes a SOC estimation method for electric vehicle lithium batteries based on deep learning, and constructs a CNN-GRU-LSTM (Convolutional Neural Network Gated Recurrent Unit-Long Short-Term Memory) neural network model, which can further improve the prediction accuracy of lithium battery state of charge (SOC) and realize real-time accurate estimation of lithium battery SOC.
[0005] In order to achieve the above requirements, the present invention adopts the following technical solutions:
[0006] A method for estimating SOC of a lithium battery of an electric vehicle based on deep learning comprises the following steps:
[0007] S1, obtaining historical voltage data, historical current data, historical temperature data and actual value of the historical state of charge SOC of the lithium battery and performing preprocessing to obtain sample data;
[0008] S2. Construct a convolutional neural network gated recurrent unit-long short-term memory neural network model and a segmentation dataset, train it on the segmentation dataset, and use the mean absolute error (MSE) and root square error (RMSE) indicators to evaluate the model performance to obtain an optimized model.
[0009] S3. Deploy the optimization model to the cloud-based lithium battery management system CBMS for real-time SOC estimation.
[0010] Preferably, the historical voltage data, historical current data and historical temperature data of the lithium battery are obtained from the terminal lithium battery management system CBMS;
[0011] The cloud-based lithium battery management system CBMS is used to store operating parameters such as voltage, current, temperature, etc. of the lithium battery of the electric vehicle to form a lithium battery operating parameter database;
[0012] The specific process of data processing by the cloud-based lithium battery management system CBMS is as follows: accurately measuring the voltage, current, temperature and capacity parameters of the electric vehicle lithium battery during the charging and discharging process through sensors;
[0013] The corresponding measured values are logged into a Raspberry Pi 4B+ connected to the Internet, and all information is sent to a centralized cloud server to establish a database of lithium battery operating parameters.
[0014] Preferably, the specific process of pretreatment is:
[0015] S11: Data cleaning: remove erroneous values, outliers and duplicate data to ensure data quality;
[0016] S12: Data normalization, compressing the data to [0,1] to eliminate the large difference in values between different features that leads to different results;
[0017] Among them, the calculation formula for data normalization processing is:
[0018]
[0019] Among them, X norm represents the normalized value, X i Represents the original data, X max and X min Represent the minimum and maximum values of each feature respectively.
[0020] Preferably, the input variables of the convolutional neural network gated recurrent unit-long short-term memory neural network model are the voltage parameter, current parameter and temperature parameter of the lithium battery, and the output variable is the SOC corresponding to the lithium battery;
[0021] Preferably, the ampere-hour integration method is used to calculate the historical state of charge (SOC) of the lithium battery, and the calculated SOC value is used as the true value of the lithium battery. The specific calculation formula is as follows:
[0022]
[0023] Among them, SOC k represents the actual SOC value of the lithium battery calculated at the kth moment, I k-1 represents the current at time k-1, Δt represents the sampling interval, Q max Represents the rated capacity of the lithium battery.
[0024] The calculated actual SOC value is used for subsequent testing of model performance.
[0025] Preferably, the input variables of the convolutional neural network gated recurrent unit-long short-term memory neural network model are the voltage parameter, current parameter and temperature parameter of the lithium battery, and the output variable is the state of charge SOC corresponding to the lithium battery;
[0026] The segmented data set is constructed by acquiring historical voltage data, historical current data, historical temperature data and historical SOC actual values of the lithium battery, including a training set, a test set and a validation set;
[0027] The training, test, and validation data with a ratio of 70:20:10 are selected from the uniform distribution of input variable and output variable pairs to construct the training set, test set, and validation set.
[0028] Preferably, the convolutional neural network gated recurrent unit-long short-term memory neural network model, i.e., the CNN-GRU-LSTM neural network model, includes four types of layer structures, namely, a one-dimensional convolutional layer, i.e., a 1D CNN layer, a long short-term memory layer, i.e., an LSTM layer, a gated recurrent unit layer, i.e., a GRU layer, and a fully connected layer;
[0029] The 1D CNN layer contains 128 filters;
[0030] The GRU layer contains 64 hidden units;
[0031] The LSTM layer contains 64 hidden units;
[0032] The CNN-GRU-LSTM neural network model consists of an input layer, a hidden layer and a fully connected layer;
[0033] The input layer receives the current, voltage and temperature data of the lithium battery at historical moments;
[0034] The hidden layer is composed of a 1D CNN layer, a GRU layer and an LSTM layer, which are used to extract the time series characteristics of the lithium battery state and perform nonlinear modeling;
[0035] The fully connected layer serves as the output layer of the neural network model and is responsible for outputting the SOC prediction result of the lithium battery.
[0036] Preferably, the operation steps of the convolutional neural network gated recurrent unit-long short-term memory neural network model include:
[0037] S21: After the input data is standardized, the one-dimensional 1D convolution layer of the convolutional neural network CNN is used to extract features of the sample data and construct a feature vector with time series characteristics;
[0038] S22: Input the obtained time series feature vector into the gated recurrent unit GRU module to capture the short-term and medium-term time series dependencies in the data and filter out invalid information;
[0039] S23: Further model the long-term dependencies in the data through the long short-term memory network LSTM layer to enhance the ability to capture long-term span information;
[0040] S24: The output of the LSTM layer is flattened into one-dimensional data through the Flatten layer so as to be passed to the subsequent fully connected layer;
[0041] S25: Use a fully connected layer as the output layer to output the SOC prediction result.
[0042] Preferably, the feature extraction expression in the process of extracting features from sample data using a one-dimensional 1D convolutional layer of a convolutional neural network CNN is:
[0043] Y = f(X*W+b);
[0044] Among them, Y represents the extracted feature vector, W represents the weight matrix, X represents the input, * represents the convolution operation, b represents the bias vector, and f represents the activation function.
[0045] The gated recurrent unit GRU module performs filtering through its update gate and reset gate;
[0046] The update gate can determine the extent to which the previous state information is passed to the current state estimation. For abnormal data or invalid information generated by interference such as slight error fluctuations of the sensor, the update gate can reduce the continuous impact of these erroneous information on the SOC estimation.
[0047] The reset gate can control the degree of forgetting of historical states. When encountering historical data that may be invalid, the reset gate can properly reset the previous state, so that the model focuses on the current more effective input data to estimate the SOC.
[0048] Preferably, when the mean absolute error MSE and root square error RMSE indicators are used to evaluate the model performance, the expressions of the mean absolute error MSE and root square error RMSE are:
[0049]
[0050] Among them, N represents the length of the data, SOC (m)k Represents the measured value of lithium battery SOC, SOC (p)k Represents the predicted value of lithium battery SOC.
[0051] In summary, the electric vehicle lithium battery SOC estimation method based on deep learning of the present invention, compared with the traditional technology, the present invention adopts a new lightweight combined neural network model (CNN-GRU-LSTM), which extracts the features of lithium battery charging and discharging data through CNN, uses GRU to focus on early time related information, improves computational efficiency, and combines the long-term memory ability of LSTM for in-depth analysis. This model combines the hierarchical feature extraction of CNN, the computational efficiency of GRU and the long-term memory storage potential of LSTM network. Compared with traditional algorithms, it has higher prediction accuracy and faster processing speed, and can be used to estimate the SOC value of electric vehicle lithium batteries online. It can estimate the SOC of lithium batteries more reliably and accurately in the process of self-learning network parameters.
[0052] Due to the small number of parameters of the model, it is able to minimize the SOC estimation error with shorter processing time and smaller memory usage, thus providing strong support for the effective management of lithium batteries in electric vehicles.
[0053] The technical method of the present invention is further described in detail below through the drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] Figure 1 A flow chart of the electric vehicle lithium battery SOC estimation method based on deep learning provided by the present invention.
[0055] Figure 2 Schematic diagram of the system architecture of the CNN-GRU-LSTM neural network model in the present invention. DETAILED DESCRIPTION
[0056] The technical method of the present invention is further described below by means of the accompanying drawings and embodiments. It should be noted that unless otherwise specifically stated, the relative arrangement of components and S, numerical expressions and values described in these embodiments do not limit the scope of the present application.
[0057] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.
[0058] Technologies, systems, and devices known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, systems, and devices should be considered part of the specification.
[0059] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.
[0060] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.
[0061] The present invention provides a method for estimating the SOC of a lithium battery of an electric vehicle based on deep learning, comprising the following steps:
[0062] S1. Obtain the actual value of the historical voltage data, historical current data, historical temperature data and historical state of charge SOC of the lithium battery and pre-process it to obtain sample data, and pre-process the data to ensure consistency and eliminate noise;
[0063] Preferably, the historical voltage data, historical current data and historical temperature data of the lithium battery are obtained from the terminal lithium battery management system CBMS;
[0064] The cloud-based lithium battery management system CBMS is used to store the voltage, current, temperature and other operating parameters of the lithium battery of the electric vehicle to form a lithium battery operating parameter database. Compared with the on-board BMS, it can provide unlimited data storage capacity and more powerful computing processing capabilities;
[0065] The specific process of data processing by the cloud-based lithium battery management system CBMS is as follows: accurately measuring the voltage, current, temperature and capacity parameters of the electric vehicle lithium battery during the charging and discharging process through sensors;
[0066] The corresponding measured values are logged into a Raspberry Pi 4B+ connected to the Internet, and all information is sent to a centralized cloud server to establish a database of lithium battery operating parameters.
[0067] Furthermore, the specific process of preprocessing is:
[0068] S11: Data cleaning: remove erroneous values, outliers and duplicate data to ensure data quality;
[0069] S12: Data normalization, compressing the data to [0,1] to eliminate the large difference in values between different features that leads to different results;
[0070] Among them, the calculation formula for data normalization processing is:
[0071]
[0072] Among them, X norm represents the normalized value, X i Represents the original data, X max and X min Represent the minimum and maximum values of each feature respectively.
[0073] S2. Construct a convolutional neural network gated recurrent unit-long short-term memory neural network model and a segmentation dataset, train it on the segmentation dataset, and use the mean absolute error (MSE) and root square error (RMSE) indicators to evaluate the model performance to obtain an optimized model.
[0074] Furthermore, the input variables of the convolutional neural network gated recurrent unit-long short-term memory neural network model are the voltage parameter, current parameter and temperature parameter of the lithium battery, and the output variable is the SOC corresponding to the lithium battery;
[0075] Furthermore, the ampere-hour integration method is used to calculate the historical state of charge (SOC) of the lithium battery, and the calculated SOC value is used as the true value of the lithium battery. The specific calculation formula is as follows:
[0076]
[0077] Among them, SOC k represents the actual SOC value of the lithium battery calculated at the kth moment, I k-1 represents the current at time k-1, Δt represents the sampling interval, Q max Represents the rated capacity of the lithium battery.
[0078] The calculated actual SOC value is used for subsequent testing of model performance.
[0079] Furthermore, the input variables of the convolutional neural network gated recurrent unit-long short-term memory neural network model are the voltage parameter, current parameter and temperature parameter of the lithium battery, and the output variable is the state of charge SOC corresponding to the lithium battery;
[0080] The segmented data set is constructed by acquiring historical voltage data, historical current data, historical temperature data and historical SOC actual values of the lithium battery, including a training set, a test set and a validation set;
[0081] The training, test, and validation data with a ratio of 70:20:10 are selected from the uniform distribution of input variable and output variable pairs to construct the training set, test set, and validation set.
[0082] Furthermore, the convolutional neural network gated recurrent unit-long short-term memory neural network model, namely the CNN-GRU-LSTM neural network model, includes four types of layer structures, namely, a one-dimensional convolutional layer, namely, a 1D CNN layer, a long short-term memory layer, namely, an LSTM layer, a gated recurrent unit layer, namely, a GRU layer, and a fully connected layer;
[0083] The CNN-GRU-LSTM neural network model includes an input layer, a hidden layer, and a fully connected layer:
[0084] The input layer receives the current, voltage and temperature data of the lithium battery at historical moments;
[0085] The hidden layer is composed of a 1D CNN layer, a GRU layer and an LSTM layer, which are used to extract the time series characteristics of the lithium battery state and perform nonlinear modeling;
[0086] The fully connected layer serves as the output layer of the neural network model and is responsible for outputting the SOC prediction result of the lithium battery.
[0087] Furthermore, the operation steps of the convolutional neural network gated recurrent unit-long short-term memory neural network model include:
[0088] S21: After the input data is standardized, the one-dimensional 1D convolution layer of the convolutional neural network CNN is used to extract features of the sample data and construct a feature vector with time series characteristics;
[0089] S22: Input the obtained time series feature vector into the gated recurrent unit GRU module to capture the short-term and medium-term time series dependencies in the data and filter out invalid information;
[0090] S23: Further model the long-term dependencies in the data through the long short-term memory network LSTM layer to enhance the ability to capture long-term span information;
[0091] S24: The output of the LSTM layer is flattened into one-dimensional data through the Flatten layer so as to be passed to the subsequent fully connected layer;
[0092] S25: Use a fully connected layer as the output layer to output the SOC prediction result.
[0093] The gated recurrent unit GRU module performs filtering through its update gate and reset gate;
[0094] The update gate can determine the extent to which the previous state information is passed to the current state estimation. For abnormal data or invalid information generated by interference such as slight error fluctuations of the sensor, the update gate can reduce the continuous impact of these erroneous information on the SOC estimation.
[0095] The reset gate can control the degree of forgetting of historical states. When encountering historical data that may be invalid, the reset gate can properly reset the previous state, so that the model focuses on the current more effective input data to estimate the SOC.
[0096] Furthermore, the feature extraction expression in the process of extracting features from sample data using the one-dimensional 1D convolutional layer of the convolutional neural network CNN is:
[0097] Y = f(X*W+b);
[0098] Among them, Y represents the extracted feature vector, W represents the weight matrix, X represents the input, * represents the convolution operation, b represents the bias vector, and f represents the activation function.
[0099] Furthermore, the convolutional neural network CNN includes 128 filters, the gated recurrent unit GRU includes 64 hidden units, and the long short-term memory LSTM includes 64 hidden units.
[0100] Furthermore, when using the mean absolute error MSE and root square error RMSE indicators to evaluate model performance, the expressions of the mean absolute error MSE and root square error RMSE are:
[0101]
[0102] Among them, N represents the length of the data, SOC (m)k Represents the measured value of lithium battery SOC, SOC (p)k Represents the predicted value of lithium battery SOC.
[0103] S3. Deploy the optimization model to the cloud-based lithium battery management system CBMS for real-time SOC estimation.
[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical method of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical method to deviate from the spirit and scope of the technical method of the present invention.
Claims
1. A method for estimating SOC of lithium batteries of electric vehicles based on deep learning, characterized in that: The following steps are involved: S1, obtaining historical voltage data, historical current data, historical temperature data and actual value of the historical state of charge SOC of the lithium battery and performing preprocessing to obtain sample data; S2. Construct a convolutional neural network gated recurrent unit-long short-term memory neural network model and a segmentation dataset, train it on the segmentation dataset, and use the mean absolute error (MSE) and root square error (RMSE) indicators to evaluate the model performance to obtain an optimized model. S3. Deploy the optimization model to the cloud-based lithium battery management system (CBMS) to perform real-time state of charge (SOC) estimation.
2. The electric vehicle lithium battery SOC estimation method based on deep learning according to claim 1 is characterized in that: The historical voltage data, historical current data and historical temperature data of the lithium battery are obtained from the terminal lithium battery management system CBMS; The cloud-based lithium battery management system CBMS is used to store operating parameters such as voltage, current, temperature, etc. of the lithium battery of the electric vehicle to form a lithium battery operating parameter database; The specific process of data processing by the cloud-based lithium battery management system CBMS is as follows: accurately measuring the voltage, current, temperature and capacity parameters of the electric vehicle lithium battery during the charging and discharging process through sensors; The corresponding measured values are logged into a Raspberry Pi 4B+ connected to the Internet, and all information is sent to a centralized cloud server to establish a database of lithium battery operating parameters.
3. The electric vehicle lithium battery SOC estimation method based on deep learning according to claim 1 is characterized in that: The specific process of preprocessing is: S11: Data cleaning: remove erroneous values, outliers and duplicate data to ensure data quality; S12: Data normalization, compressing the data to [0,1] to eliminate the large difference in values between different features that leads to different results; Among them, the calculation formula for data normalization processing is: Among them, X norm represents the normalized value, X i represents the original data, X max and X min Represent the minimum and maximum values of each feature respectively.
4. The electric vehicle lithium battery SOC estimation method based on deep learning according to claim 1, characterized in that: The historical state of charge SOC value of the lithium battery is calculated by the ampere-hour integration method, and the calculated historical state of charge SOC value of the lithium battery is used as the actual value of the historical state of charge SOC of the lithium battery. The specific calculation formula is as follows: Among them, SOC k represents the actual value of the state of charge SOC calculated at the kth moment, I k-1 represents the current at time k-1, Δt represents the sampling interval, Q max Represents the rated capacity of the lithium battery; The actual value of the state of charge (SOC) obtained by calculation is used for subsequent testing of the model performance.
5. The electric vehicle lithium battery SOC estimation method based on deep learning according to claim 4 is characterized in that: The input variables of the convolutional neural network gated recurrent unit-long short-term memory neural network model are the voltage parameter, current parameter and temperature parameter of the lithium battery, and the output variable is the state of charge SOC corresponding to the lithium battery; The segmented data set is constructed by acquiring historical voltage data, historical current data, historical temperature data and actual values of the historical state of charge (SOC) of the lithium battery, and includes a training set, a test set and a validation set; The training, test, and validation data with a ratio of 70:20:10 are selected from the uniform distribution of input variable and output variable pairs to construct the training set, test set, and validation set.
6. The electric vehicle lithium battery SOC estimation method based on deep learning according to claim 1, characterized in that: The convolutional neural network gated recurrent unit-long short-term memory neural network model, namely the CNN-GRU-LSTM neural network model, includes four types of layer structures, namely, a one-dimensional convolutional layer, namely a 1D CNN layer, a long short-term memory layer, namely an LSTM layer, a gated recurrent unit layer, namely a GRU layer, and a fully connected layer; The 1D CNN layer contains 128 filters; The GRU layer contains 64 hidden units; The LSTM layer contains 64 hidden units; The CNN-GRU-LSTM neural network model consists of an input layer, a hidden layer and a fully connected layer; The input layer receives the current data, voltage data, and temperature data of the lithium battery at historical moments; The hidden layer is composed of a 1D CNN layer, a GRU layer and an LSTM layer, which are used to extract the time series characteristics of the lithium battery state and perform nonlinear modeling; The fully connected layer serves as the output layer of the neural network model and is responsible for outputting the SOC prediction result of the lithium battery.
7. The electric vehicle lithium battery SOC estimation method based on deep learning according to claim 1, characterized in that: The steps of running the Convolutional Neural Network Gated Recurrent Unit-Long Short-Term Memory Neural Network model include: S21: After the input data is standardized, the one-dimensional 1D convolution layer of the convolutional neural network CNN is used to extract features of the sample data and construct a feature vector with time series characteristics; S22: Input the obtained time series feature vector into the gated recurrent unit GRU module to capture the short-term and medium-term time series dependencies in the data and filter out invalid information; S23: Further model the long-term dependencies in the data through the long short-term memory network LSTM layer to enhance the ability to capture long-term span information; S24: The output of the LSTM layer is flattened into one-dimensional data through the Flatten layer so as to be passed to the subsequent fully connected layer; S25: Use a fully connected layer as the output layer to output the SOC prediction result.
8. The electric vehicle lithium battery SOC estimation method based on deep learning according to claim 7, characterized in that: The feature extraction expression in the process of extracting features from sample data using the one-dimensional 1D convolutional layer of the convolutional neural network CNN is: Y = f(X*W+b); Among them, Y represents the extracted feature vector, W represents the weight matrix, X represents the input, * represents the convolution operation, b represents the bias vector, and f represents the activation function.
9. The electric vehicle lithium battery SOC estimation method based on deep learning according to claim 7, characterized in that: The gated recurrent unit GRU module performs filtering through its update gate and reset gate; The update gate can determine the extent to which the previous state information is passed to the current state estimation. For abnormal data or invalid information generated by interference such as slight error fluctuations of the sensor, the update gate can reduce the continuous impact of these erroneous information on the SOC estimation. The reset gate can control the degree of forgetting of historical states. When encountering historical data that may be invalid, the reset gate can properly reset the previous state, so that the model focuses on the current more effective input data to estimate the SOC.
10. The electric vehicle lithium battery SOC estimation method based on deep learning according to claim 1, characterized in that: When using the mean absolute error MSE and root square error RMSE indicators to evaluate model performance, the expressions of the mean absolute error MSE and root square error RMSE are: Among them, N represents the length of the data, SOC (m)k Represents the measured value of lithium battery SOC, SOC (p)k Represents the predicted value of lithium battery SOC.