Battery RUL prediction method based on MH-MoE model and EIS data
Through the multi-head hybrid expert model (MH-MoE), the generalization capability and efficiency problems of traditional models in battery RUL prediction are solved, and more accurate and efficient battery status prediction is achieved, which is suitable for battery health management.
Patent Information
- Application Number
- CN202411280084.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2025-01-03
AI Technical Summary
When traditional machine learning models process complex and heterogeneous battery EIS data, their generalization capabilities are limited, which easily leads to overfitting. They are not ideal for large-scale data processing, making it difficult to accurately predict the remaining service life of the battery. (RUL).
The multi-head hybrid expert model (MH-MoE) combined with EIS data is used to process and weight sum the RUL prediction value of the battery through 3 expert sub-models and 3 gating network models. The expert network is dynamically selected by using the gating mechanism to perform Linear regression prediction.
It improves the accuracy and computing efficiency of battery RUL prediction, enhances the expressiveness and adaptability of the model, and can capture battery status information more comprehensively, and is suitable for battery health management.
Smart Images
Figure CN119247143A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of electrochemical technology, and in particular to a battery RUL prediction method based on an MH-MoE model and EIS data. Background Art
[0002] The remaining useful life (RUL) of a battery is one of the key indicators for measuring the health of a battery. Usually, the battery RUL can be expressed by the number of remaining cycles. The importance of battery RUL is mainly reflected in the following aspects: it is used to evaluate the health of the battery, guide the use and maintenance of the battery, provide early warning of potential battery failures, evaluate the capacity attenuation of the battery, optimize the battery management strategy, and evaluate the value of battery reuse.
[0003] Predicting the remaining useful life (RUL) of a battery plays an extremely critical role in practical applications. Electrochemical impedance spectroscopy (EIS) technology is an efficient analytical tool that can reveal the electrochemical reactions inside the battery. As the active substances in the battery are consumed and the performance decreases, the EIS curve will change accordingly. Therefore, using EIS data to predict the RUL of the battery is an effective technical approach. Typically, we use EIS data to train machine learning algorithms, such as neural networks or support vector machines, to learn and predict the changing trends of the battery RUL.
[0004] Traditional machine learning models usually rely on a single model to process the entire dataset, which often does not work well when dealing with complex and highly heterogeneous data. Such models have limited generalization capabilities, are prone to overfitting, and are not efficient when dealing with large-scale data.
[0005] The Multi-Head Mixture of Experts (MH-MoE) model is an advanced neural network design that aims to improve the expressiveness and computational efficiency of the model. The architecture of MH-MoE is summarized as follows: 1. Expert network: It consists of multiple independent sub-networks, and the expert network processes the input data from different perspectives.
[0006] 2. Gating network: Each gating network acts as a "head" to select and activate the appropriate expert network based on the input data. Multiple gating networks (heads) can process input data in parallel.
[0007] 3. Multi-head mechanism: This mechanism allows multiple gating networks to operate simultaneously, and each gating network generates its own weight distribution to select and activate a part of the expert network.
[0008] The workflow of MH-MoE is as follows: First, the input data passes through multiple gating networks (heads), each of which generates a set of weights based on the input data for selecting and activating the expert network.
[0009] Next, each gating network selects and activates a few expert networks, which process the input data.
[0010] Finally, the outputs of all activated expert networks are aggregated to form the final output result. Summary of the invention
[0011] In order to solve the above technical problems, the present invention proposes a battery RUL prediction method based on the MH-MoE model and EIS data, and uses the battery EIS data to predict the battery RUL.
[0012] The technical solution adopted by the present invention to solve its technical problem is: A battery RUL prediction method based on MH-MoE model and EIS data includes the following steps: Step S1. Test and collect battery EIS and RUL data: Use professional impedance spectrum test instrument to test and collect battery EIS data, and cover the frequency range of 10 -2 HZ to 10 5 HZ, select 5 to 60 frequency values in the middle of the frequency range, preferably 60 as samples;
[0013] When collecting data, record the real and imaginary data of the impedance corresponding to each frequency, as well as the target value to be predicted, which is the corresponding battery RUL.
[0014] Step S2. Constructing the MH-MoE model: The MH-MoE model contains three expert sub-models and three gating network models, and each sub-model is an MLP model.
[0015] The adopted MLP sub-expert model consists of 1 output layer, 2 middle layers and 1 output layer, and the number of neurons in the middle layers is 32.
[0016] The three gating network models all use a multi-layer perceptron model, which is a three-layer network structure. The input layer receives input data, and the middle layer is a 32-dimensional vector.
[0017] The three gating network models independently calculate the weight of each expert sub-model. The output of each gating network is a three-dimensional vector, corresponding to the weights of the three expert sub-models, and the sum of the three weights is 1.
[0018] The RUL prediction values of the three expert sub-models are weighted and summed with the expert weights output by the three gating networks to obtain the weighted sum values of the three RULs.
[0019] Perform linear regression prediction on the weighted sum of the three RULs to obtain the final predicted value RUL; Step S3. Training and verifying the MH-MoE model: Implement the MH-MoE model code and train it on the Keras platform. The training data is divided into a training set and a test set in a ratio of 8:2. The model is trained on the training set and verified on the test set.
[0020] Furthermore, the number of training times is 100 and the number of batch samples is 2. The optimizer for model training is the adam algorithm, and the loss function is the mean square error.
[0021] Step S4. Predicting the battery RUL: After the MH-MoE model is trained, if the model effect is good, deploy the MH-MoE model and use MH-MoE to predict the battery RUL. The weighted sum of the three RUL values is linearly regressed, and the predicted value is the RUL prediction value.
[0022] Furthermore, the loss function of the model is mean square error; the optimizer is adam; and the evaluation indicator is the mean absolute error, which is the sum of the absolute values of the differences between the target value and the predicted value.
[0023] The beneficial effects of the present invention are The present invention utilizes its structural advantages and can more effectively process battery EIS data and provide accurate and efficient RUL prediction, which is of great significance for battery health management.
[0024] The details are as follows: 1. Multi-angle feature learning: Each “head” can focus on learning different aspects of EIS data, thereby capturing more comprehensive information about the battery state and improving prediction accuracy.
[0025] 2. Flexible gating mechanism: The gating network of MH-MoE can dynamically select the appropriate expert network according to the characteristics of the input data, which enhances the model's adaptability to changes in battery status.
[0026] 3. Enhanced model expressiveness: By combining the outputs of multiple expert networks, MH-MoE is able to learn more complex battery degradation patterns, thereby improving the expressiveness of RUL prediction.
[0027] 4. Computational efficiency: Due to the sparsity of the gating network, only part of the expert network is activated, which reduces the consumption of computing resources and improves computing efficiency.
[0028] 5. Easy to expand: If you need to process more data features or more complex tasks, you can expand the MH-MoE model by increasing the number of "heads" without drastically changing the model structure. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1It is a structural diagram of the battery RUL prediction model based on the MH-MoE model and EIS of the present invention.
[0030] Figure 2 It is the overall flow chart of predicting battery RUL based on MH-MoE model and EIS of the present invention.
[0031] Figure 3 This is the MLP sub-model structure diagram.
[0032] Figure 4 It is a structural diagram of the gated network model.
[0033] Figure 5 This is a sample diagram of the battery RUL prediction results based on the MH-MoE model and EIS of the present invention. DETAILED DESCRIPTION
[0034] refer to Figure 1 , the input data is the EIS data of the battery. EIS consists of a real part and an imaginary part. real_1 to real_m represent the real part data of EIS, imaginary_1 to imaginary_m represent the imaginary part data of EIS, and m represents the number of frequencies of EIS.
[0035] Next is the expert layer and the gated network layer. The present invention takes three multi-layer perceptron (MLP) expert sub-models and three gated networks based on multi-layer perceptron (MLP) as examples.
[0036] The input data are input into three expert sub-models and three gating network models respectively.
[0037] The three expert sub-models extract corresponding features from different angles, and the three gating network models independently calculate the weight of each expert sub-model.
[0038] The outputs of the three expert sub-models are RUL_1, RUL_2 and RUL_3. These three outputs are all continuous variables, representing the predicted values of RUL for each sub-expert model.
[0039] The three gating network models independently calculate the weight of each expert sub-model. The output of each gating network is a three-dimensional vector, corresponding to the weights of the three expert sub-models, and the sum of the three weights is 1.
[0040] Then, the RUL prediction values of the three expert sub-models are weighted and summed with the expert weights output by the three gating networks to obtain the weighted sum values of the three RULs.
[0041] Finally, the weighted sum of the three RULs is used for linear regression prediction to obtain the final predicted value SOH.
[0042] refer to Figure 2 Step S1. Test and collect battery EIS and RUL data: Use professional impedance spectrum test instrument to test and collect battery EIS data, and it needs to cover a certain frequency range. The frequency range is set to 10 -2 HZ to 10 5 HZ, multiple frequency values are selected at a certain interval in the middle of the frequency range, and the frequency values are generally selected from 5 to 60. The present invention selects 60 as a sample.
[0043] When collecting data, it is necessary to record the real and imaginary data of the impedance corresponding to each frequency, as well as the target value that needs to be predicted, which is the corresponding battery RUL.
[0044] Step S2. Constructing the MH-MoE model: The MH-MoE model includes three expert sub-models and three gated network models. In the present invention, each sub-model is an MLP model.
[0045] The present invention uses three expert sub-models, each of which is an MLP model. The MLP sub-expert model used in the present invention comprises one output layer, two middle layers and one output layer, wherein the number of neurons in the middle layer is 32.
[0046] The three gated network models all use a multi-layer perceptron model, which is a three-layer network structure. The input layer receives input data, the middle layer is a 32-dimensional vector, and the output layer is a 3-dimensional vector, which corresponds to the weights of the three expert sub-models, and the sum of the three weights is 1.
[0047] The three gating network models independently calculate the weight of each expert sub-model. The output of each gating network is a three-dimensional vector, corresponding to the weights of the three expert sub-models, and the sum of the three weights is 1.
[0048] The RUL prediction values of the three expert sub-models are weighted and summed with the expert weights output by the three gating networks to obtain the weighted sum values of the three RULs.
[0049] The weighted sum of the three RULs is used for linear regression prediction to obtain the final predicted value RUL.
[0050] refer to Figure 3 , each sub-model is an MLP model. The expert sub-model adopts a multi-layer perceptron model, which is a 4-layer network structure. The input layer receives input data, there are 2 layers in the middle layer, both are 32-dimensional vectors, the activation function is the relu function, the output layer is a continuous variable, the activation function is linear, and the output value is the RUL prediction value of the sub-expert model.
[0051] refer to Figure 4The gated network model adopts a multi-layer perceptron model, which is a 3-layer network structure. The input layer receives input data, the middle layer is a 32-dimensional vector, the activation function is the relu function, and the output layer is a 3-dimensional vector. The activation function is softmax, which corresponds to the weights of the three expert sub-models, and the weighted summation is performed. The sum of the three weights is 1.
[0052] The weighted summation process is as follows: The outputs of the three expert sub-models are RUL_1, RUL_2, and RUL_3. These three outputs are all continuous variables, representing the predicted value of RUL by each sub-expert model. The three gating network models independently calculate the weight of each expert sub-model. The output of each gating network is a three-dimensional vector, corresponding to the weights of the three expert sub-models, and the sum of the three weights is 1.
[0053] The RUL prediction values of the three expert sub-models are weighted and summed with the expert weights output by the three gating networks to obtain the weighted sum values of the three RULs.
[0054] The weighted sum of the three RULs is used for linear regression prediction to obtain the final predicted value RUL.
[0055] Step S3. Training and verifying the MH-MoE model: The present invention implements the MH-MoE model code and trains it on the Keras platform. The training data is divided into a training set and a test set in a ratio of 8:2. The model is trained on the training set, and the model is verified on the test set.
[0056] Furthermore, the number of training epochs is 100, and the number of batch samples is batch_size=32. The optimizer for model training is the adam algorithm, and the loss function is mean square error (mse).
[0057] Step S4. Predicting the battery RUL: After the MH-MoE model is trained, if the model effect is good, the MH-MoE model can be deployed and the battery RUL can be predicted using MH-MoE. The weighted sum of the three RUL values is linearly regressed, and the predicted value is the RUL prediction value.
[0058] Furthermore, the loss function of the model is Mean Square Error (MSE); the optimizer is Adam; and the evaluation metric is Mean Absolute Error (MAE), which is the sum of the absolute values of the differences between the target value and the predicted value. Example
[0059] 1. Prediction data display The present invention tests the effect on a public data set (https: / / github.com / YunweiZhang / ML-identify-battery-degradation) and uses the impedance data therein to predict the battery RUL. There are 525 battery EIS and RUL data. Referring to Table 1, EIS has 120 features, and the feature numbers are 1 to 120. The first 60 features are the real part data of the impedance data corresponding to the frequency from high to low, and the last 60 features are the imaginary part data of the impedance data corresponding to the frequency from high to low.
[0060] Where f1 to f120 represent the EIS frequency characteristics, and RUL here represents the remaining number of battery cycles.
[0061] 2. Display of RUL prediction results The mean of the test set of battery cycle number is 149.2952, the standard deviation is 105.1953, and the mean of the prediction set is 150.1942, the standard deviation is 105.2969, which shows that the data distribution of the prediction results is basically consistent with the data distribution of the real data.
[0062] The root mean square error between the prediction set and the test set of battery RUL is 3.2861, which shows that the prediction effect of the model is acceptable and there is room for further optimization.
[0063] The prediction results are shown in Table 2 below:
[0064] The prediction results are as follows: Figure 5 As shown, Figure 5 The solid line in the upper middle represents the true value, the dotted line represents the predicted value, and the solid square represents the error value. The numbers represent the true values. It can be seen from the figure that the prediction effect is acceptable and can be further optimized.
Claims
1. A battery RUL prediction method based on MH-MoE model and EIS data, characterized in that: The steps include: Step S1. Test and collect battery EIS and RUL data: Use professional impedance spectrum test instrument to test and collect battery EIS data, and cover the frequency range of 10 -2 HZ to 10 5 HZ, select 5 to 60 frequency values in the middle of the frequency range, preferably 60 as samples; When collecting data, record the real and imaginary data of the impedance corresponding to each frequency, as well as the target value to be predicted, which is the corresponding battery RUL; Step S2. Constructing the MH-MoE model: The MH-MoE model includes three expert sub-models and three gating network models, and each sub-model is an MLP model; The adopted MLP sub-expert model consists of 1 output layer, 2 middle layers and 1 output layer, where the number of neurons in the middle layer is 32; The three gating network models all use a multi-layer perceptron model, which is a three-layer network structure. The input layer receives input data, and the middle layer is a 32-dimensional vector. Three gating network models independently calculate the weight of each expert sub-model. The output of each gating network is a 3D vector, corresponding to the weights of the three expert sub-models, and the sum of the three weights is 1. The RUL prediction values of the three expert sub-models are weighted and summed with the expert weights output by the three gating networks to obtain the weighted sum values of the three RULs; Perform linear regression prediction on the weighted sum of the three RULs to obtain the final predicted value RUL; Step S3. Training and verifying the MH-MoE model: Implement the MH-MoE model code and train it on the Keras platform. The training data is divided into a training set and a test set in a ratio of 8:
2. The model is trained on the training set and verified on the test set. Step S4. Predicting the battery RUL: After the MH-MoE model is trained, if the model effect is good, deploy the MH-MoE model and use MH-MoE to predict the RUL of the battery; perform linear regression on the weighted sum of the three RULs, and the predicted value obtained is the RUL prediction value.
2. A battery RUL prediction method based on MH-MoE model and EIS data according to claim 1, characterized in that: In step S3, the number of training times is 100 and the number of batch samples is 2; the optimizer for model training is the Adam algorithm, and the loss function is the mean square error.
3. A battery RUL prediction method based on MH-MoE model and EIS data according to claim 1, characterized in that: In step S4, the loss function of the model is mean square error; the optimizer is adam; and the evaluation index is mean absolute error, i.e., the sum of the absolute values of the difference between the target value and the predicted value.