Lithium ion battery life early prediction method and system based on meta learning

Abstract

The invention discloses a meta-learning-based lithium ion battery life early prediction method and system, and belongs to the technical field of batteries. The method comprises the following steps: collecting historical data of charge and discharge aging tests of the lithium ion battery under different working conditions; after data screening and preprocessing are carried out on the collected data, training data set and test data set samples are constructed; constructing and training a meta-learning framework based on a gating cycle unit and a random forest as a prediction model; and inputting the test data set into the trained prediction model to obtain a prediction result of the service life of the lithium ion battery, and evaluating the obtained prediction result. According to the invention, the deep features of the battery data are extracted and analyzed through the meta-learning model, and the early and accurate prediction of the service life of the lithium ion battery is realized, so that the efficiency and safety of a battery management system are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of batteries. Specifically, the present invention relates to a method and system for early prediction of the lifespan of lithium-ion batteries based on meta-learning. Background Art

[0002] In the fields of electric vehicles and renewable energy storage, the key role of lithium-ion batteries has become increasingly prominent. As the usage time increases, the gradual decline in battery performance has become a major challenge. When the battery capacity drops to 80% of the initial capacity, its application in electric vehicles may no longer be ideal. Therefore, predicting the remaining useful life of the battery in advance is of great significance for ensuring vehicle performance, optimizing battery usage strategies, and enhancing user confidence. In addition, early lifespan prediction can also help manufacturers improve production processes, and users can more effectively plan battery replacements. Exploring an innovative method that can provide early warnings before the battery performance begins to decline has a profound impact on improving the overall performance and sustainability of the battery. The implementation of this method is expected to bring significant progress to battery technology and enhance its application value in modern energy solutions.

[0003] The existing early prediction of the lifespan of lithium batteries is mainly based on model-driven and data-driven methods. These methods have the following defects:

[0004] 1. Model-driven methods require knowledge of various physical and chemical changes inside the battery and a large amount of knowledge reserve for construction. At the same time, when dealing with different working conditions and battery types, their generalization ability is poor and it is difficult to adapt to changing working conditions.

[0005] 2. Data-driven methods require a large amount of data when constructing models. At the same time, the timeliness and accuracy of model prediction are also worthy of consideration. A large amount of operating data may be required for traditional prediction to accurately predict the final remaining life.

[0006] Therefore, the present invention proposes a method and system for early prediction of the lifespan of lithium-ion batteries based on meta-learning. Summary of the Invention

[0007] The present invention aims to overcome the deficiencies of the prior art and proposes a method and system for early prediction of the lifespan of lithium-ion batteries based on meta-learning to achieve the following objectives: Extract and analyze the deep features of battery data through a meta-learning model to achieve early and accurate prediction of the lifespan of lithium-ion batteries, thereby significantly improving the efficiency and safety of the battery management system.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is: A method for early prediction of the lifespan of lithium-ion batteries based on meta-learning, the method comprising the following steps:

[0009] Step S1: Collect historical data of charge and discharge aging tests of lithium-ion batteries under different working conditions, including cycle life, discharge capacity, voltage, current, and temperature;

[0010] Step S2: After data screening and preprocessing of the data collected in Step S1, construct training dataset and test dataset samples;

[0011] Step S3: Based on the training dataset, construct and train a meta-learning framework based on gated recurrent unit and random forest as a prediction model;

[0012] Step S4: Input the test dataset into the prediction model trained in Step S3 to obtain the prediction result of the lithium-ion battery life;

[0013] Step S5: Evaluate the prediction result obtained in Step S4.

[0014] Preferably, Step S2 includes:

[0015] Step S21: Perform data cleaning and screening on the data collected in Step S1. The screening includes calculating the correlation coefficient between various data and battery life through correlation analysis, and eliminating the data with a correlation coefficient lower than the preset threshold;

[0016] Step S22: Extract the required feature data from the screened data;

[0017] Step S23: Standardize the selected feature data;

[0018] Step S24: Divide the standardized feature data into training data and test data, and use the battery cycle life as the prediction index.

[0019] Preferably, the feature data extracted in Step S22 includes the variance of the discharge capacity difference.

[0020] Preferably, the standardization method in Step S23 is Z-score standardization, and the mathematical expression is:

[0021]

[0022] where x is the feature value in the original data, μ is the mean of this feature, σ is the standard deviation of this feature, and x ′ is the value after standardization.

[0023] Preferably, the prediction model includes a gated recurrent unit and a meta-model based on random forest. Among them, the output end of the gated recurrent unit is connected to the meta-model based on random forest, and the meta-model based on random forest outputs the prediction result.

[0024] Preferably, the output calculation formula of the gated recurrent unit is as follows:

[0025]

[0026] where h t is the hidden state output at time step t; h t-1 is the hidden state output at time step t - 1; z t is the update gate; is the candidate hidden state, and ⊙ represents element-wise multiplication.

[0027] Preferably, after the prediction model is constructed, the prediction model is trained, where: the training dataset samples constructed in step 2 are input into the gated recurrent unit, the parameters of the gated recurrent unit are randomly initialized, and training is performed using the backpropagation algorithm; during the training process, when the loss function value no longer decreases within the preset number of epochs, stop training and retain the parameters of the gated recurrent unit; then, the features output by the gated recurrent unit are used as the input of the meta-model based on random forest; finally, the meta-model outputs the early prediction result of the lithium-ion battery life.

[0028] Preferably, the mean square error is selected as the loss function, and its mathematical expression is:

[0029]

[0030] where is the output prediction value, y i is the actual value, and n is the total number of samples.

[0031] Preferably, the step S5 includes:

[0032] Step S51: Use the root mean square error RMSE to measure the deviation between the prediction value and the true value, and its mathematical expression is:

[0033]

[0034] where is the prediction value, y i is the actual value, and n is the total number of samples;

[0035] Step S52: Use the mean absolute percentage error MAPE to measure the prediction accuracy of the model, and its mathematical expression is:

[0036]

[0037] where is the prediction value, y i is the actual value, and n is the total number of samples;

[0038] Step S53: Visualize the prediction results and the actual values, that is, draw a line chart to show the relationship between the two, and at the same time draw a residual chart to analyze the prediction ability of the model and check for systematic biases.

[0039] Meanwhile, the present application also proposes an early prediction system for the lifespan of lithium-ion batteries based on meta-learning. The system includes a computer for executing a computer program formed according to the above method.

[0040] The technical effects of the present invention are as follows:

[0041] (1) By introducing a time series feature extraction based on GRU and a meta-learning framework of RF, the model gives full play to the advantages of meta-learning in feature extraction and generalization ability, and can capture the complex dynamic relationships of multi-dimensional charge and discharge data of lithium-ion batteries, thus significantly improving the effect of battery life prediction.

[0042] (2) The method of the present invention has strong robustness through training and optimization of a large amount of experimental data, and can adapt to the prediction requirements under different charge and discharge conditions and data distributions.

[0043] (3) The meta-learning-based framework of the present application has high computational efficiency and can realize real-time life prediction in an actual battery management system, helping to optimize the charge and discharge strategy to extend the battery service life.

[0044] (4) When evaluating the performance of the model, the present method uses the root mean square error (RMSE) and the mean absolute percentage error (MAPE) as evaluation indicators. These indicators can comprehensively measure the prediction accuracy and error distribution of the model, ensuring the reliability and interpretability of the prediction results.

[0045] (5) The present method is applicable to various types of lithium-ion battery systems and can be extended to the field of performance prediction and management of other energy storage systems. Description of the Drawings

[0046] Figure 1 It is a flowchart of an early prediction method for the lifespan of lithium-ion batteries based on meta-learning according to an embodiment of the present invention;

[0047] Figure 2 It is a schematic structural diagram of a prediction model combining a gated recurrent unit and a meta-learning framework according to an embodiment of the present invention. Detailed Embodiments

[0048] The following is a further detailed description of the specific implementation manners of the present invention with reference to the accompanying drawings through the description of embodiments, aiming to help those skilled in the art have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention and facilitate its implementation. It should be noted that the terms "first", "second", etc. described in this application are only for the convenience of describing the technical solution to distinguish different components, and do not limit this application. To make the technical solution of the present invention clearer, the present invention is explained and illustrated through the following embodiments.

[0049] This embodiment provides a method for early prediction of the life of a lithium-ion battery based on meta-learning, as Figure 1 shown, the method includes the following steps:

[0050] Step S1: Collect historical data of charge and discharge aging tests of lithium-ion batteries under different working conditions, including cycle life, discharge capacity, voltage, current, and temperature;

[0051] Step S2: After screening and preprocessing the data collected in Step S1, construct training dataset and test dataset samples;

[0052] Step S3: Based on the training dataset, construct and train a meta-learning framework based on gated recurrent unit and random forest as a prediction model;

[0053] Step S4: Input the test dataset into the prediction model trained in Step S3 to obtain the prediction result of the life of the lithium-ion battery;

[0054] Step S5: Evaluate the prediction result obtained in Step S4.

[0055] Specifically, in Step S1, the lithium-ion battery in this embodiment undergoes charge experiments under a variety of different working conditions to ensure the diversity and comprehensiveness of the collected data. In this embodiment,

[0056] The upper and lower cut-off potentials of the lithium-ion battery are set at 3.6 V and 2.0 V respectively, strictly in line with the manufacturer's specifications. One-step or two-step fast charging strategies were set up in the experiment. The format of the charging strategy is "C1(Q1)-C2", where C1 and C2 are the first and second constant current charging steps respectively, and Q1 represents the state of charge (SOC, %) at the time of current switching. For example, in the first constant current charging step, the battery is charged at 1C constant current to 30% SOC and then switched to the second constant current charging step; after constant current charging, it is charged at constant voltage to the preset SOC and then starts constant current discharging, and this cycle repeats. To cover a variety of actual usage scenarios, the experimental conditions include different temperatures (such as 0 °C, 25 °C and 40 °C), discharge rates (such as 0.5C, 1C and 2C), and cycle depths. During the charge and discharge process of the battery, key data such as voltage, current, and capacity are recorded through a real-time monitoring device connected to a computer. In each cycle, data is collected at fixed intervals of time, and finally, fully charged and discharged cycle data covering a variety of working conditions is obtained until the state of health (SOH) drops to 80%.

[0057] In this embodiment, step S2 includes:

[0058] Step S21, perform data cleaning and screening on the data collected in step S1. Among them, data cleaning includes cleaning invalid data such as outliers and noise, so as to improve the quality of the data. The screening includes calculating the correlation coefficients between various data and the battery life through correlation analysis, and eliminating the data with correlation coefficients lower than the preset threshold, such as the initial voltage, ambient temperature, etc., so as to reduce the overall computation amount and improve the calculation efficiency.

[0059] Step S22, extract the required feature data from the screened data. In addition to common features such as the initial discharge capacity, charging time, and battery temperature, the feature data extracted in this embodiment also includes the variance of the discharge capacity difference, that is, find the discharge capacity curves of the 100th cycle and the 10th cycle in the battery cycle, subtract them to obtain a difference curve and obtain its variance. In the prior art, it has been proved that the variance of the discharge capacity difference is highly correlated with the cycle life (the specific proof process is not elaborated in this application). Therefore, taking the variance of the discharge capacity difference as one of the feature data can improve the prediction accuracy of the battery life.

[0060] Step S23, standardize the selected feature data to improve the consistency of the data. In this embodiment, the standardization method is Z-score standardization, and the mathematical expression is:

[0061]

[0062] where x is the feature value in the original data, μ is the mean of this feature, σ is the standard deviation of this feature, x ′It is the value after standardization. By this method, the data is made to conform to the standard normal distribution (with a mean of 0 and a standard deviation of 1), thereby eliminating the influence of different feature scales.

[0063] Step S24: Divide the standardized feature data into training data and test data, and use the battery cycle life as the prediction index. In this embodiment, 80% is used as training data to train the model, and the remaining 20% is used as test data to evaluate the model performance.

[0064] In step S3 of this embodiment, as Figure 2 shown, the constructed prediction model includes a gated recurrent unit (GRU) and a meta-model based on random forest (RF). Among them, the output end of the gated recurrent unit is connected to the meta-model based on random forest, and the meta-model based on random forest outputs the prediction result.

[0065] The gated recurrent unit is used to extract the time series features of battery data. By selectively retaining information through the gating mechanism, it can extract the time-dependent relationship in the charge and discharge data of lithium-ion batteries, capture long-term and short-term deep dynamic features (high-order features), so as to be used for subsequent modeling and prediction and improve the prediction accuracy. The gated recurrent unit consists of an input layer, a hidden layer and an output layer, and its output calculation formula is:

[0066]

[0067] where, h t is the hidden state output at time step t; h t-1 is the hidden state output at time step t - 1; z t is the update gate; is the candidate hidden state, and ⊙ represents element-wise multiplication.

[0068] The meta-model based on random forest, as the core part of the meta-learning framework, can model non-linear relationships by parallelly training multiple decision trees when facing the complexity of different battery data. Compared with traditional machine learning methods, random forest has strong generalization ability and can adapt to variable working conditions and data distributions. Each tree learns specific patterns from the data during training, and the final prediction result is the average or voting result of the output results of all decision trees. This enables the model to adapt to different charge and discharge working conditions and the complexity of battery data. At the same time, since random forest can calculate the training process of each tree in parallel, it improves the calculation efficiency of the prediction model and can realize real-time life prediction in the actual battery management system, helping to optimize the charge and discharge strategy to extend the battery service life.

[0069] The core advantage of meta - learning lies in its ability to enhance the model's prediction ability on new tasks by learning the shared information among tasks in a multi - task learning framework. Specifically in this invention, the meta - model based on random forest is trained under different charge - discharge working conditions and data distributions, enabling the model to capture the complex dynamic relationships in lithium - ion battery life prediction. This process not only improves the model's processing ability for multi - dimensional battery data but also enhances the overall prediction performance of the model by gradually optimizing the decision - tree structure.

[0070] After the prediction model is constructed, the prediction model is trained, where: the training dataset samples constructed in step 2 are input into the gated recurrent unit, the parameters of the gated recurrent unit are randomly initialized, and the back - propagation algorithm is used for training; during the training process, when the loss function value no longer decreases within a preset number of epochs (for example, set to 10 epochs in this embodiment, and can be flexibly selected according to actual situations in specific implementations), the training stops and the parameters of the gated recurrent unit are retained; then, the features output by the gated recurrent unit are used as the input of the meta - model based on random forest; finally, the meta - model outputs the early prediction result of the lithium - ion battery life. Through the training and optimization of a large amount of experimental data, this invention has strong robustness and can adapt to the prediction requirements under different charge - discharge working conditions and data distributions.

[0071] Among them, during the training process, the mean squared error (MSE) is selected as the loss function, and its mathematical expression is:

[0072]

[0073] Where, is the output predicted value, y i is the actual value, and n is the total number of samples. The mean squared error is used to measure the deviation between the predicted value and the actual value. In this embodiment, the prediction effect is improved by minimizing the loss function value during model optimization.

[0074] After training, in step S4, 20% of the divided test data is input into the trained gated recurrent unit to extract time - series features; subsequently, the output of the gated recurrent unit is used as the input and passed to the meta - model based on random forest, making full use of the feature extraction and generalization ability of the meta - learning framework to accurately predict the battery life.

[0075] After obtaining the prediction result, in step S5 of this embodiment, multiple model evaluation metrics and visualization methods are used to comprehensively evaluate the prediction result, and the improvement and applicability of the method in this embodiment in terms of prediction performance are analyzed.

[0076] Specifically, step S5 includes:

[0077] Step S51: Use the root mean square error (RMSE) to measure the deviation between the predicted value and the true value. Its mathematical expression is:

[0078]

[0079] where, is the predicted value, y i is the actual value, and n is the total number of samples; the smaller the RMSE, the smaller the prediction error of the model and the better its performance.

[0080] Step S52: Use the mean absolute percentage error (MAPE) to measure the prediction accuracy of the model. Its mathematical expression is:

[0081]

[0082] where, is the predicted value, y i is the actual value, and n is the total number of samples; the smaller the MAPE value, the higher the prediction accuracy of the model.

[0083] Step S53: Visualize the prediction results and the actual values, that is, draw a line chart to show the relationship between the two, and at the same time draw a residual chart to analyze the prediction ability of the model and check for systematic biases.

[0084] Through experimental data testing, this method shows a significant performance improvement compared with traditional machine learning methods. The RMSE in life prediction is 78.98 and the MAPE is 8.69%, while the RMSE of traditional machine learning is 88 and the MAPE is 10.21%. Obviously, the RMSE and MAPE of the method in this application are smaller, and the model performance and prediction accuracy are higher.

[0085] In summary, for the multi-dimensional time series data during the charge and discharge process of lithium-ion batteries, this embodiment proposes a time series feature extraction method unique to gated recurrent units, and combines a random forest model to achieve non-linear modeling. The constructed prediction model can accurately predict key performance indicators such as the life of lithium-ion batteries, significantly improving the performance and usage efficiency of the battery management system.

[0086] At the same time, this application also proposes an early prediction system for the life of lithium-ion batteries based on meta-learning. The system includes a computer for executing a computer program formed according to the above method.

[0087] The above has made an exemplary description of the present invention in conjunction with the accompanying drawings. Obviously, the specific implementation of the present invention is not limited by the above methods. As long as various non-substantive improvements are made using the method concept and technical solution of the present invention; or without improvement, the above concept and technical solution of the present invention are directly applied to other occasions, all fall within the protection scope of the present invention.

Claims

1. A method for early prediction of lithium-ion battery life based on meta-learning, characterized by: The method comprises the following steps: Step S1, collecting historical data of lithium-ion battery charge and discharge aging tests under different working conditions, including cycle life, discharge capacity, voltage, current, and temperature; Step S2: After data screening and preprocessing of the data collected in step S1, a training data set and a test data set sample are constructed; Step S3: Based on the training data set, construct and train a meta-learning framework based on gated recurrent units and random forests as a prediction model; Step S4, inputting the test data set into the prediction model trained in step S3 to obtain a prediction result of the lithium-ion battery life; Step S5: Evaluate the prediction result obtained in step S4.

2. The method for early prediction of lithium-ion battery life based on meta-learning according to claim 1, characterized in that: The step S2 comprises: Step S21, cleaning and screening the data collected in step S1, wherein the screening includes calculating the correlation coefficient between various data and battery life through correlation analysis, and eliminating data with a correlation coefficient lower than a preset threshold; Step S22, extracting required feature data from the screened data; Step S23, standardizing the selected feature data; Step S24: divide the standardized feature data into training data and test data, and use the battery cycle life as a prediction indicator.

3. The method for early prediction of lithium-ion battery life based on meta-learning according to claim 2, characterized in that: The characteristic data extracted in step S22 includes the variance of the discharge capacity difference.

4. A method for early prediction of lithium-ion battery life based on meta-learning according to claim 2 or 3, characterized in that: The standardization method in step S23 is Z-score standardization, and the mathematical expression is: Among them, x is the eigenvalue in the original data, μ is the mean of the feature, σ is the standard deviation of the feature, and x ′ is the standardized value.

5. The method for early prediction of lithium-ion battery life based on meta-learning according to claim 1, characterized in that: The prediction model includes a gated recurrent unit and a meta-model based on random forests, wherein an output end of the gated recurrent unit is connected to the meta-model based on random forests, and the meta-model based on random forests outputs a prediction result.

6. The method for early prediction of lithium-ion battery life based on meta-learning according to claim 5, characterized in that: The output calculation formula of the gated recurrent unit is: Among them, h t is the hidden state output at time step t; h t-1 is the hidden state output at time step t-1; z t It is the update gate; is a candidate hidden state and ⊙ represents element-wise multiplication.

7. A method for early prediction of lithium-ion battery life based on meta-learning according to claim 5 or 6, characterized in that: After the prediction model is constructed, the prediction model is trained, wherein: the training data set samples constructed in step 2 are input into the gated recurrent unit, the parameters of the gated recurrent unit are randomly initialized, and the back propagation algorithm is used for training; during the training process, when the loss function value no longer decreases within a preset number of cycles, the training is stopped and the gated recurrent unit parameters are retained; then, the features output by the gated recurrent unit are used as the input of the random forest-based meta-model; finally, the meta-model outputs an early prediction result for the life of the lithium-ion battery.

8. The method for early prediction of lithium-ion battery life based on meta-learning according to claim 7, characterized in that: The mean square error is selected as the loss function, and its mathematical expression is: in, is the output prediction value, y i is the actual value, and n is the total sample size.

9. The method for early prediction of lithium-ion battery life based on meta-learning according to claim 1, characterized in that: The step S5 comprises: Step S51: Use the root mean square error RMSE to measure the deviation between the predicted value and the true value. Its mathematical expression is: in, is the predicted value, y i is the actual value, n is the total sample size; Step S52: Use the mean absolute percentage error (MAPE) to measure the prediction accuracy of the model. The mathematical expression is: in, is the predicted value, y i is the actual value, n is the total sample size; Step S53: visualize the predicted results and actual values, that is, draw a line graph to show the relationship between the two, and draw a residual graph to analyze the prediction ability of the model and check whether there is any systematic deviation.

10. A lithium-ion battery life early prediction system based on meta-learning according to any one of claims 1 to 9, characterized in that: The system comprises a computer for executing a computer program formed according to the method according to any one of claims 1-9.

Images