Method for predicting RUL based on HS-MH-MoE model and current resistance impedance of battery

Abstract

The invention discloses a method for predicting RUL based on an HS-MH-MoE model and current resistance impedance of a battery. The method comprises the following steps: S1, testing and acquiring current, resistance, impedance and RUL data of the battery; s2, constructing an HS-MH-MoE model, wherein the HS-MH-MoE model comprises a multilayer multi-head sparse expert sub-model; a two-layer structure is used, each layer comprises three expert sub-models, and each sub-model is an MLP model; s3, the HS-MH-MoE model is trained and verified; and S4, predicting the RUL of the battery: after the model training is completed, deploying an HS-MH-MoE model, and predicting the RUL of the battery by using the HS-MH-MoE model. According to the method, the RUL of the battery is predicted by constructing the HS-MH-MoE model and using the current, resistance and impedance of the battery. The multi-level feature extraction capability, the efficient sparse activation mechanism, the parallel processing capability, the high prediction precision and the strong adaptability and robustness are realized.

Description

Technical Field

[0001] The invention belongs to the field of electrochemical technology, and in particular to a method for predicting RUL based on a HS-MH-MoE model and the current resistance impedance of a battery. Background Art

[0002] Battery RUL (Remaining Useful Life) is the remaining life of the battery, which is an important indicator for evaluating the health status of the battery. Common battery RULs include the number of remaining battery cycles. Battery RUL is an important indicator of the battery, which is mainly reflected in several aspects: evaluating the health status of the battery, guiding the use and maintenance of the battery, warning of battery failure, evaluating the battery capacity reduction, optimizing the battery management strategy and evaluating the secondary utilization value of the battery.

[0003] Predicting battery RUL has high application value.

[0004] The remaining useful life (RUL) of a battery is closely related to operating data such as current, resistance, and impedance, making it a hot research area to use these parameters to estimate the RUL of a battery. Researchers usually use data on current, resistance, and impedance to train machine learning algorithms, such as neural networks or support vector machines, to capture the changing trends of battery RUL and make predictions.

[0005] However, traditional machine learning models often rely on a single model to process all data, which is inadequate when faced with complex and highly heterogeneous data sets. The generalization ability of such a single model is limited, which can easily lead to overfitting. At the same time, when processing large amounts of data, the processing efficiency of a single model is relatively low.

[0006] Mixture of Experts (MOE) is a machine learning architecture that integrates multiple specialized models, where each sub-model exhibits strong predictive power on a different subset of the dataset.

[0007] The core concept of MOE is to use a gating network to distribute input data to various expert models, each of which learns a specific aspect of the data set. Subsequently, the predictions of these expert models are aggregated to generate the final prediction.

[0008] The Sparse Mixture of Experts (SMoE) model is designed to solve complex problems. It implements sparse activation of expert sub-models through a gating network and then performs weighted aggregation to reduce the computational burden.

[0009] Hierarchical Sparse Mixture of Experts (HS-MoE) is an advanced version of SMoE, which uses tree-like hierarchical experts and gating networks to handle more complex tasks.

[0010] As a neural network architecture, the Multi-Head Mixture of Experts (MH-MoE) model operates in parallel through multiple gating networks (heads), each activating a different expert network, improving the performance and computational efficiency of the model.

[0011] Combining HS-MoE and MH-MoE can form a new model, which integrates the advantages of multi-level feature extraction and multi-head mechanism and is more powerful. It can be named as multi-layer multi-head sparse hybrid expert model (HS-MH-MoE). Summary of the invention

[0012] In order to solve the above technical problems, the present invention proposes a method for predicting RUL based on the HS-MH-MoE model and the current resistance impedance of the battery, and uses the battery current, resistance and impedance to predict the battery RUL.

[0013] The technical solution adopted by the present invention to solve its technical problem is: A method for predicting RUL based on the HS-MH-MoE model and the current resistance impedance of a battery mainly includes the following steps: Step S1. Test and collect battery current, resistance, impedance and RUL data. When collecting data, it is necessary to record the battery current, resistance and impedance data, as well as the target value to be predicted, which is the corresponding battery RUL.

[0014] The specific steps include: S11 Prepare equipment and materials: Prepare data acquisition equipment, such as current sensor, AC impedance tester, signal recorder, etc. Prepare data storage media and battery samples.

[0015] S12 current data acquisition: insert a collection resistor into the battery's charge and discharge circuit, and use a current sensor to measure and record the current value in real time.

[0016] S13 resistance data collection: Under different charge and discharge conditions, measure the voltage across the battery and the current passing through, and calculate the resistance according to the resistance formula R = U / I.

[0017] S14 Impedance data acquisition: Use an impedance tester to test the impedance value of the battery at different frequencies to obtain an impedance frequency spectrum.

[0018] S15RUL data collection: statistics of historical battery cycle data.

[0019] Step S2. Construct HS-MH-MoE model: The HS-MH-MoE model includes a multi-layer multi-head sparse expert sub-model. A 2-layer structure is used, each layer contains 3 expert sub-models, each sub-model is an MLP model, and each sub-model is a 3-layer network structure. The input layer receives input data, the middle layer is a 32-dimensional vector, and the output layer is the feature representation of the input data or the RUL prediction value.

[0020] Furthermore, each level includes three gated network models, all of which adopt multi-layer perceptron models. Each gated network model 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 weighted sum is performed. The sum of the three weights is 1.

[0021] Furthermore, the weighted summation implementation process is as follows: The outputs of the three expert sub-models of the first level are respectively recorded as output1, output2 and output3. These three outputs are all high-dimensional vectors. The present invention takes 32 dimensions as an example to represent the feature representation of each sub-expert model for the input data.

[0022] Each gating network model at the first level selects the outputs of the two experts with the highest probability, multiplies them with the corresponding weights and sums them up to obtain the weighted sum of the outputs of each gating network model and sub-expert model at the first level.

[0023] Then, the weighted sum values ​​output by each gated network model and sub-expert model of the first level are averaged to obtain the weighted sum average value of all sub-expert models of the first level.

[0024] Afterwards, the weighted sum average of all sub-expert models in the first layer is used as the input data of the second layer, and is input into each expert sub-model and each gating network model in the second layer respectively.

[0025] The outputs of the three expert sub-models at the second level are denoted as RUL_1, RUL_2 and RUL_3 respectively.

[0026] The three gating networks in the second layer independently calculate the weight of each expert sub-model in the second layer. The output of each gating network is a three-dimensional vector, which corresponds to the weights of the three expert sub-models, and the sum of the three weights is 1.

[0027] Then, each gating network model at the second level selects the outputs of the two experts with the highest probability, multiplies them with the corresponding weights and sums them up to obtain the weighted sum of the outputs of each gating network model at the second level and all sub-expert models. Here, a total of three RUL weighted sum prediction values ​​are obtained.

[0028] Step S3. Training and verifying the HS-MH-MoE model: Implement the HS-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.

[0029] Furthermore, the number of training times is 100 and the number of batch samples is 32. The optimizer for model training is the adam algorithm, and the loss function is the mean square error.

[0030] Step S4. Predicting the battery RUL: After the HS-MH-MoE model is trained, if the model effect is good, deploy the HS-MH-MoE model and use HS-MH-MoE to predict the battery RUL. Connect the three RUL weighted sum prediction values ​​to a linear regression layer and output the final RUL prediction value.

[0031] Furthermore, the loss function of the model is mean square error; the optimizer is adam; and the evaluation indicator is mean absolute error, which is the sum of the absolute values ​​of the differences between the target value and the predicted value.

[0032] The beneficial effects of the present invention are The highlights of the HS-MH-MoE model in predicting battery RUL are its multi-level feature extraction capability, efficient sparse activation mechanism, parallel processing capability, high prediction accuracy, and strong adaptability and robustness. These features make it a potential tool for the development and application of battery health management systems.

[0033] The details are as follows: 1. Multi-level feature extraction: The multi-level structure of HS-MH-MoE is able to extract complex features of battery health status from different levels of abstraction, which helps to understand and predict the battery degradation trend more accurately.

[0034] 2. Sparse activation mechanism: Through sparse activation, the model is able to activate only the most relevant expert network when making predictions, which greatly reduces the demand for computing resources while improving the efficiency and accuracy of predictions.

[0035] 3. Multi-head parallel processing: The multi-head design of HS-MH-MoE allows the model to process different input features in parallel, and each head can focus on different aspects of the data, thereby improving the expressiveness and adaptability of the model.

[0036] 4. Flexibility and scalability: The structure of HS-MH-MoE enables it to adapt to different datasets and prediction tasks. By increasing or adjusting the levels and number of expert networks and gating networks, the model can be optimized according to specific needs.

[0037] 5. Robustness: By combining the opinions of multiple experts, the model is robust to noise and outliers, which is particularly important in practical applications.

[0038] 6. Use the multi-layer perceptron MLP expert model: The MLP model has low computational complexity and is conducive to model training. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a structural diagram of the battery RUL prediction model based on the HS-MH-MoE model of the present invention.

[0040] Figure 2 It is the overall flow chart of battery RUL prediction based on HS-MH-MoE model of the present invention.

[0041] Figure 3 This is the structure diagram of the first-level MLP sub-model.

[0042] Figure 4 This is a diagram of the gated network model structure.

[0043] Figure 5 This is the structure diagram of the second-level MLP sub-model.

[0044] Figure 6 This is a sample diagram of battery RUL prediction results based on the HS-MH-MoE model. DETAILED DESCRIPTION

[0045] refer to Figure 1 refer to Figure 1 , the input data are the current, resistance and impedance data of the battery, specifically the sensor branch current, the battery branch current, the ratio of the sensor branch current to the battery branch current, the battery impedance calculated from the raw data, the calibrated and smoothed battery impedance, the estimated electrolyte resistance and the estimated charge transfer resistance; Next is the multi-layer multi-head sparse expert layer and the gated network layer. The present invention takes a two-layer structure, each layer is 3 MLP expert sub-models, and each layer has 3 gated network models based on multi-layer perceptron (MLP) as an example; The input data is input into the three expert sub-models of the first level respectively, and the corresponding features are extracted from different angles. The outputs of the three expert sub-models are output1, output2 and output3 respectively. These three outputs are all high-dimensional vectors. The present invention takes 32 dimensions as an example to represent the feature representation of each sub-expert model for the input data; The input data is input into the three gating networks of the first level respectively, and the weight of each expert sub-model in the first level is calculated independently. The output of each gating network is a 3D vector, which corresponds to the weight of the three expert sub-models respectively, and the sum of the three weights is 1; Afterwards, each gating network model selects the outputs of the two experts with the highest probability, multiplies them with the corresponding weights and sums them up to obtain the weighted sum of the outputs of each gating network model and sub-expert model at the first level; Then, the weighted sum values ​​output by each gating network model and sub-expert model of the first level are averaged to obtain the weighted sum average value of all sub-expert models of the first level; Afterwards, the weighted sum average of all sub-expert models in the first layer is used as the input data of the second layer, and is input into each expert sub-model and each gating network model in the second layer respectively; The outputs of the three expert sub-models at the second level are RUL_1, RUL_2, and RUL_3. These three outputs are all continuous values, representing the predicted values ​​of RUL by each sub-expert model. The three gating networks in the second layer independently calculate the weight of each expert sub-model in the second layer. The output of each gating network is a 3D vector, which corresponds to the weight of the three expert sub-models, and the sum of the three weights is 1. Then, each gating network model at the second level selects the outputs of the two experts with the highest probability, multiplies them with the corresponding weights and sums them up to obtain the weighted summed RUL prediction value of the outputs of each gating network model at the second level and all sub-expert models. Here, a total of 3 RUL prediction values ​​are obtained; Finally, the 3 RUL prediction values ​​are connected to a linear regression layer to output the final RUL prediction value.

[0046] refer to Figure 2 The implementation of the present invention comprises the following steps Step S1: Test and collect battery current, resistance, impedance and RUL data: S11 Prepare equipment and materials: Prepare data acquisition equipment, such as current sensor, AC impedance tester, signal recorder, etc. Prepare data storage media and battery samples.

[0047] S12 current data acquisition: insert a collection resistor into the battery's charge and discharge circuit, and use a current sensor to measure and record the current value in real time.

[0048] S13 resistance data collection: Under different charge and discharge conditions, measure the voltage across the battery and the current passing through, and calculate the resistance according to the resistance formula R = U / I.

[0049] S14 Impedance data acquisition: Use an impedance tester to test the impedance value of the battery at different frequencies to obtain an impedance frequency spectrum.

[0050] S15RUL data collection: statistics of historical battery cycle data.

[0051] Step S2. Constructing HS-MH-MoE model: HS-MH-MoE model includes multi-layer multi-head sparse expert sub-model. The present invention uses a 2-layer structure, each layer includes 3 expert sub-models, each sub-model is an MLP model, each sub-model is a 3-layer network structure, the input layer receives input data, the middle layer is a 32-dimensional vector, and the output layer is the feature representation of the input data or the RUL prediction value.

[0052] Each level includes three gated network models, all of which adopt multi-layer perceptron models. Each gated network model 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.

[0053] Step S3. Training and verifying the HS-MH-MoE model: The present invention implements the HS-MH-MoE model code and trains it on the Keras platform, with the number of training epochs = 100 and the number of batch samples batch_size = 32. The optimizer for model training is the adam algorithm, and the loss function is mean square error (mse).

[0054] 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.

[0055] Step S4. Predicting the battery RUL: After the HS-MH-MoE model is trained, if the model effect is good, the HS-MH-MoE model can be deployed and the HS-MH-MoE can be used to predict the battery RUL.

[0056] refer to Figure 2 The specific implementation of the present invention includes the following steps: Step S1. Test and collect battery current, resistance, impedance and RUL data. When collecting data, it is necessary to record the battery current, resistance and impedance data, as well as the target value to be predicted, which is the corresponding battery RUL.

[0057] The specific steps include: S11 Prepare equipment and materials: Prepare data acquisition equipment, such as current sensor, AC impedance tester, signal recorder, etc. Prepare data storage media and battery samples.

[0058] S12 current data acquisition: insert a collection resistor into the battery's charge and discharge circuit, and use a current sensor to measure and record the current value in real time.

[0059] S13 resistance data collection: Under different charge and discharge conditions, measure the voltage across the battery and the current passing through, and calculate the resistance according to the resistance formula R = U / I.

[0060] S14 Impedance data acquisition: Use an impedance tester to test the impedance value of the battery at different frequencies to obtain an impedance frequency spectrum.

[0061] S15RUL data collection: statistics of historical battery cycle data.

[0062] Step S2. Construct HS-MH-MoE model: The HS-MH-MoE model includes a multi-layer multi-head sparse expert sub-model. A 2-layer structure is used, each layer contains 3 expert sub-models, each sub-model is an MLP model, and each sub-model is a 3-layer network structure. The input layer receives input data, the middle layer is a 32-dimensional vector, and the output layer is the feature representation of the input data or the RUL prediction value.

[0063] Each level includes three gated network models, all of which adopt multi-layer perceptron models. Each gated network model 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. The weighted sum is performed, and the sum of the three weights is 1.

[0064] Furthermore, the weighted summation implementation process is as follows: The outputs of the three expert sub-models of the first level are respectively recorded as output1, output2 and output3. These three outputs are all high-dimensional vectors. The present invention takes 32 dimensions as an example to represent the feature representation of each sub-expert model for the input data.

[0065] Each gating network model at the first level selects the outputs of the two experts with the highest probability, multiplies them with the corresponding weights and sums them up to obtain the weighted sum of the outputs of each gating network model and sub-expert model at the first level.

[0066] Then, the weighted sum values ​​output by each gated network model and sub-expert model of the first level are averaged to obtain the weighted sum average value of all sub-expert models of the first level.

[0067] Afterwards, the weighted sum average of all sub-expert models in the first layer is used as the input data of the second layer, and is input into each expert sub-model and each gating network model in the second layer respectively.

[0068] The outputs of the three expert sub-models at the second level are denoted as RUL_1, RUL_2 and RUL_3 respectively.

[0069] The three gating networks in the second layer independently calculate the weight of each expert sub-model in the second layer. The output of each gating network is a three-dimensional vector, which corresponds to the weights of the three expert sub-models, and the sum of the three weights is 1.

[0070] Then, each gating network model at the second level selects the outputs of the two experts with the highest probability, multiplies them with the corresponding weights and sums them up to obtain the weighted sum of the outputs of each gating network model at the second level and all sub-expert models. Here, a total of three RUL weighted sum prediction values ​​are obtained.

[0071] refer to Figure 3 ,The first level of expert sub-model adopts a multi-layer perceptron model, which is a 3-layer network structure. The input layer receives the input data, the middle layer is a 32-dimensional vector, the activation function is the relu function, and the output layer is a 32-dimensional vector, the activation function is relu, and the output layer is the feature expression of the input data.

[0072] refer to Figure 4 The 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 sum of the three weights is 1.

[0073] refer to Figure 5 The second-level 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 middle layers, both of which are 32-dimensional vectors, and the activation function is the relu function. The output layer is a continuous variable, the activation function is linear, and the output layer is the predicted value of RUL.

[0074] The three gating networks of the second level have the same model structure as the gating networks of the first level. They independently calculate the weight of each expert sub-model in the second level. The output of each gating network is a three-dimensional vector, which corresponds to the weights of the three expert sub-models, and the sum of the three weights is 1.

[0075] Step S3. Training and verifying the HS-MH-MoE model: The present invention implements the HS-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.

[0076] Furthermore, during training, the number of epochs is 100, the number of batch samples is batch_size=32, the optimizer for model training is the Adam algorithm, and the loss function is the mean square error (MSE).

[0077] Step S4. Predict the battery RUL: After the HS-MH-MoE model is trained, if the model effect is good, deploy the HS-MH-MoE model and use HS-MH-MoE to predict the battery RUL. Connect the three RUL weighted sum prediction values ​​to a linear regression layer and output the final RUL prediction value. The activation function is a linear function.

[0078] Furthermore, the loss function of the model is Mean Square Error (MSE). The optimizer is adam. 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

[0079] 1. Prediction data display The present invention tests the effect on the data of the impedance state (type=impedance) of the B0018 battery in the NASA battery data set (https: / / www.nasa.gov / intelligent-systems-division#battery-data), and uses the battery current, resistance and impedance data therein to predict the battery RUL. The data set is divided into a training set and a test set in a ratio of 8:2, and the HS-MH-MoE model is trained on the training set, and the effect is tested on the test set. The test data has a total of 7 features, and the names and meanings of the features are shown in Table 1 below:

[0080] The specific data samples are shown in Table 2 below:

[0081] In Table 2, the features such as Sense_current, Battery_current, Current_ratio, Battery_impedance and Rectified_impedance are divided into real data and imaginary data.

[0082] The mean of the test set is 139.7838, the standard deviation is 96.9046, and the mean of the prediction set is 141.6108, the standard deviation is 96.9786, which shows that the data distribution of the prediction results is basically consistent with the data distribution of the real data.

[0083] The root mean square error between the prediction set and the test set is 7.3195, and there is still room for further optimization of the model.

[0084] The prediction results are shown in Table 3 below:

[0085] The prediction results are as follows: Figure 6 As shown, reference Figure 6 The dotted line in the upper middle represents the true value, the solid 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 method for predicting RUL based on the HS-MH-MoE model and the current resistance impedance of a battery, characterized in that: The main steps are as follows: Step S1. Test and collect battery current, resistance, impedance and RUL data. When collecting data, it is necessary to record the battery current, resistance and impedance data, as well as the target value to be predicted, which is the corresponding battery RUL; Step S2. Construct HS-MH-MoE model: The HS-MH-MoE model includes a multi-layer multi-head sparse expert sub-model; a 2-layer structure is used, each layer contains 3 expert sub-models, each sub-model is an MLP model, and each sub-model is a 3-layer network structure, the input layer receives input data, the middle layer is a 32-dimensional vector, and the output layer is the feature representation of the input data or the RUL prediction value; Step S3. Training and verifying the HS-MH-MoE model: Implementing the HS-MH-MoE model code and training it on the Keras platform, dividing the training data into a training set and a test set in a ratio of 8:2, training the model on the training set, and verifying the model on the test set; Step S4. Predicting the battery RUL: After the HS-MH-MoE model is trained, if the model effect is good, deploy the HS-MH-MoE model and use HS-MH-MoE to predict the battery RUL; Connect the 3 RUL weighted sum prediction values ​​to a linear regression layer to output the final RUL prediction value.

2. A method for predicting RUL based on the HS-MH-MoE model and the current resistance impedance of a battery according to claim 1, characterized in that: The step S1 comprises the following steps: S11 Prepare equipment and materials: prepare data acquisition equipment, such as current sensor, AC impedance tester, signal recorder, etc.; prepare data storage media and battery samples; S12 current data acquisition: insert a collection resistor into the battery charging and discharging circuit, and use a current sensor to measure and record the current value in real time; S13 resistance data collection: Under different charge and discharge conditions, measure the voltage across the battery and the current passing through, and calculate the resistance according to the resistance formula R = U / I; S14 Impedance data acquisition: Use an impedance tester to test the impedance value of the battery at different frequencies to obtain an impedance frequency spectrum; S15RUL data collection: statistics of historical battery cycle data.

3. The method for predicting RUL based on the HS-MH-MoE model and the current resistance impedance of the battery according to claim 1, characterized in that: In step S2, each level includes three gated network models, all of which adopt multilayer perceptron models. Each gated network model 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 weighted sum is performed, and the sum of the three weights is 1.

4. A method for predicting RUL based on the HS-MH-MoE model and the current resistance impedance of a battery according to claim 3, characterized in that: The weighted summation process is as follows: The outputs of the three expert sub-models of the first level are respectively recorded as output1, output2 and output3. These three outputs are all high-dimensional vectors. The present invention takes 32 dimensions as an example to represent the feature representation of each sub-expert model for the input data; Each gating network model at the first level selects the outputs of the two experts with the highest probability, multiplies them by the corresponding weights and sums them up to obtain the weighted sum of the outputs of each gating network model and sub-expert model at the first level; Then, the weighted sum values ​​output by each gating network model and sub-expert model of the first level are averaged to obtain the weighted sum average value of all sub-expert models of the first level; Afterwards, the weighted sum average of all sub-expert models in the first layer is used as the input data of the second layer, and is input into each expert sub-model and each gating network model in the second layer respectively; The outputs of the three expert sub-models at the second level are denoted as RUL_1, RUL_2, and RUL_3; The three gating networks in the second layer independently calculate the weight of each expert sub-model in the second layer. The output of each gating network is a 3D vector, which corresponds to the weight of the three expert sub-models, and the sum of the three weights is 1. Then, each gating network model at the second level selects the outputs of the two experts with the highest probability, multiplies them with the corresponding weights and sums them up to obtain the weighted sum of the outputs of each gating network model at the second level and all sub-expert models. Here, a total of three RUL weighted sum prediction values ​​are obtained.

5. The method for predicting RUL based on the HS-MH-MoE model and the current resistance impedance of the battery according to claim 1, characterized in that: In step S3, the number of training times is 100 and the number of batch samples is 32; the optimizer for model training is the Adam algorithm, and the loss function is the mean square error.

6. The method for predicting RUL based on the HS-MH-MoE model and the current resistance impedance of the battery 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, that is, the sum of the absolute values ​​of the differences between the target value and the predicted value.

Images