Fusion feature-based lithium battery SOH prediction method, apparatus and device, and medium

Abstract

The invention discloses a lithium battery SOH prediction method and device based on fusion features, equipment and a medium, and relates to the technical field of lithium battery SOH prediction. The lithium battery SOH prediction method comprises the following steps: A1, acquiring a first data set of a target domain and a second data set of a source domain; a2, performing data preprocessing and feature extraction on the first data set and the second data set to obtain a source domain feature data set and a target domain feature data set; and A3, constructing a cross-domain prediction model. And A4, according to the source domain feature data sets of different source domains and the cross-domain prediction model, training a source domain prediction model for each source domain. And A5, fusing the source domain prediction model of each source domain to obtain a fused prediction model. S1, acquiring to-be-predicted data of a target domain; s2, performing preprocessing and feature extraction on to-be-predicted data to obtain to-be-predicted features; and S3, inputting to-be-predicted features into the fusion prediction model to obtain an SOH prediction value.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery SOH prediction, and in particular to a lithium battery SOH prediction method, device, equipment and medium based on fusion features. Background Art

[0002] As an important indicator for evaluating battery performance degradation, the state of health (SOH) of lithium batteries plays a key role in fault warning and safety assessment. Accurately obtaining the battery health status can provide a real-time monitoring basis for the battery management system, help to timely identify low-life batteries and formulate maintenance strategies, thereby reducing the risk of equipment failure and extending battery life. Especially in the scenario of cascade utilization of retired power lithium batteries, accurate SOH prediction is of great significance to ensuring the safe operation of battery packs and improving resource utilization.

[0003] In the existing technology, the SOH prediction method based on machine learning has made some progress. This method constructs health characteristics by analyzing the voltage, current and other time series data during the battery charging and discharging process, and uses data-driven models to achieve capacity decay trend modeling. Compared with the method based on electrochemical model, the data-driven solution does not need to rely on the knowledge of the internal aging mechanism of the battery, and can mine the battery degradation law through large-scale historical data, which has gradually become the mainstream research direction in this field.

[0004] However, such methods have significant limitations in cross-domain application scenarios, especially when the model is migrated from laboratory standard data sets to actual retired battery data, the problem of feature distribution offset caused by differences in operating conditions is difficult to eliminate.

[0005] Current research generally faces the challenge of inconsistent label distribution between laboratory data and retired battery data. In a laboratory environment, accurate capacity labels for the entire life cycle of the battery can be obtained through complete charge and discharge tests, but actual retired batteries are limited by the complexity of the operating environment and the cost of testing, and only a small amount of anchor label data can be obtained. This asymmetry in label acquisition makes it difficult for existing models to effectively integrate multi-source domain data features, and prediction bias will occur when directly migrating laboratory models. In addition, the differences in feature distribution caused by different battery types and charge and discharge strategies further aggravate the instability of cross-domain predictions, restricting the actual application effect of data-driven methods in retired battery scenarios. Summary of the invention

[0006] The present invention provides a lithium battery SOH prediction method, device, equipment and medium based on fusion features to improve at least one of the above technical problems.

[0007] In a first aspect, the present invention provides a lithium battery SOH prediction method based on fusion features, which comprises steps A1 to A5, and steps S1 to S3.

[0008] A1. Obtain a first data set of a target domain and a second data set of a source domain. There are at least two source domains. The data of each source domain is constructed as a second data set.

[0009] A2. Perform data preprocessing and feature extraction on the first data set and the second data set to obtain a source domain feature data set and a target domain feature data set.

[0010] A3. Construct a cross-domain prediction model. The cross-domain prediction model includes a predictor. In the training phase, the cross-domain prediction model also includes a generator and a discriminator.

[0011] A4. According to source domain feature datasets of different source domains and the cross-domain prediction model, a source domain prediction model is trained for each source domain.

[0012] A5. Fuse the source domain prediction models of each source domain to obtain a fusion prediction model.

[0013] S1. Obtain the data to be predicted in the target domain.

[0014] S2. Preprocessing and feature extraction are performed on the data to be predicted to obtain features to be predicted.

[0015] S3. Input the feature to be predicted into the fusion prediction model to obtain the SOH prediction value.

[0016] As a further solution of the present invention, the predictor is used to predict SOH based on input features. The generator is used to convert source domain features into alignment features of the target domain. The discriminator is suitable for discriminating whether the feature belongs to the source domain or the target domain, and is used for adversarial training with the generator.

[0017] Predictor The network structure includes: a third input layer of n neurons, a first fully connected layer of 512 neurons, a third ReLU activation function layer, a third LSTM layer of 128 neurons, a second self-attention module, a second fully connected layer of 256 neurons, a fourth ReLU activation function layer, a third output layer of 1 neuron, and a second Sigmoid activation function. The second Sigmoid activation function is used to predict SOH. During model training, a third random inactivation layer with a drop rate of 0.3 is also set after the self-attention module.

[0018] Generator The network structure includes the following: a first input layer of n neurons, a first one-dimensional convolutional layer with 64 neurons, 3 convolution kernels, and a convolution step size of 1, a first batch normalization layer, a first ReLU activation function layer, a first hidden layer with 64 neurons and provided with a LeakyReLU activation function layer and a batch normalization layer, a first residual module with 64 neurons, a first convolutional layer with 128 neurons, 3 convolution kernels, and a convolution step size of 2, a second batch normalization layer, a second ReLU activation function layer, a second residual module with 128 neurons, a first self-attention module, a first LSTM layer with 128 neurons, and a first output layer of n neurons. The first output layer is used to output the converted target domain features. The first residual module is used to splice the features of the first one-dimensional convolutional layer input and the features output by the first hidden layer. The second residual module is used to splice the features of the first convolutional layer input and the features output by the second ReLU activation function layer. During model training, a first random dropout layer with a post-drop rate of 0.3 is also connected after the first LSTM layer.

[0019] Discriminator network The network structure includes: a second input layer of n neurons, a second convolutional layer with 128 neurons, 3 convolution kernels, and a convolution step size of 1, a third batch normalization layer, a first LeakyReLU layer with a negative slope of 0.2, a second LSTM layer with 128 neurons, a third convolutional layer with 256 neurons, 3 convolution kernels, and a convolution step size of 1, a fourth batch normalization layer, a second LeakyReLU layer with a negative slope of 0.2, a second output layer of 1 neuron, and a first Sigmoid activation function. The first Sigmoid activation function is used to predict whether the features of the input discriminator belong to the source domain or the target domain. During model training, a second random dropout layer with a drop rate of 0.3 is also set after the second LSTM layer.

[0020] As a further solution of the present invention, according to source domain feature data sets of different source domains and the cross-domain prediction model, a source domain prediction model is trained for each source domain, specifically including: The following training steps are repeatedly performed based on the source domain feature data sets of each source domain until the training is completed, and the source domain prediction model of each source domain is obtained.

[0021] A source domain feature dataset of a preset proportion is obtained as a training set.

[0022] The features of the training set are input into the generator to obtain the aligned features of the target domain.

[0023] According to the alignment features, the source domain labels of the input features are mapped to obtain source domain mapping labels. The mapping model is: . In the formula, For the The mapped labels of the iterations, is the health status extracted for the first time from the source domain data, For the The iteration Dynamic mapping weights of samples, is the health status set of source domain data, Indicates the minimum value, Indicates the maximum value, The source domain data The health status of the extraction, The feature vector set of the features in brackets relative to the target domain data The kernel density (KDE) of The feature vector set of the features in brackets relative to the source domain data The kernel density (KDE) of For the Source domain samples, For the The iteration The features of samples after being transformed by the generator, To avoid extremely small amounts with a denominator of zero.

[0024] The alignment features and the first data and the last data of the target domain feature data set are input into the predictor to obtain the SOH prediction result.

[0025] The prediction model is updated through a loss function according to the alignment features, the source domain mapping labels, and the SOH prediction results.

[0026] As a further solution of the present invention, the generator and the discriminator are combined into a feature distribution alignment module. The loss function of the domain alignment module is: In the formula, is the adversarial training loss between the generator and the discriminator, is the MMD loss between the features generated by the generator and the feature vector set of the target domain data, To reconstruct the loss, is the trade-off coefficient of MMD loss, is the trade-off coefficient of reconstruction loss.

[0027] .

[0028] In the formula, Expressed as expectation, For the Source domain samples, is the source domain feature dataset, represents all source domain samples, For the target domain samples, is the target domain feature dataset, represents all target domain samples, is the feature vector set of source domain data, For the generator, For the discriminator, is the feature of the source domain after being transformed by the generator, is the discriminator’s judgment result on the converted source domain features, is the discriminator's discrimination result on the target domain features.

[0029] .

[0030] In the formula, is the number of samples in the source domain dataset, is the number of samples in the target domain dataset, is the RBF kernel function, which is used to calculate the distribution characteristics between two eigenvectors. For the Source domain samples, For the Source domain samples, For the target domain samples, For the target domain samples.

[0031] .

[0032] In the formula, is the feature of the source domain after being transformed by the generator, is the feature vector set of source domain data, is the square of the Euclidean distance.

[0033] The loss function of the predictor is: . In the formula, For the The source domain mapping label of the iteration, For the The SOH prediction results of the alignment features of the iterations, is the health status extracted for the first time from the source domain data, is the SOH prediction result of the first data of the target domain feature dataset, The source domain data The health status of the extraction, is the SOH prediction result of the last data of the target domain feature dataset, is the square of the Euclidean distance.

[0034] As a further solution of the present invention, the source domain prediction models of each source domain are fused to obtain a fused prediction model, which specifically includes: Calculate the mean absolute error of the source domain prediction model on the validation set for each source domain.

[0035] According to the mean absolute error of each source domain, the relative difference elimination strategy is adopted to introduce the relative difference threshold , remove the source domain prediction models whose relative error ratio exceeds the threshold, and obtain the available model set. for: . In the formula, is the number of source domain predictors, For the The mean absolute error of the source domain predictors, For the The mean absolute error of the source domain predictors.

[0036] Based on the mean absolute error, usable model weights are constructed. In the formula, For the The weights of the available models, For the The mean absolute error of available models, For the available model collection, For the The mean absolute error of the available models.

[0037] According to the predictor weights, the source domain prediction models of the respective source domains are fused to obtain a fused prediction model. In the formula, To integrate the prediction model, For the available model collection, For the The weights of the available models, For the Available models.

[0038] The fusion prediction model is tested using the non-anchor data of the target domain feature dataset. If the test is qualified, the training is completed and the fusion prediction model is obtained. Otherwise, training is performed again.

[0039] As a further solution of the present invention, data preprocessing and feature extraction are performed on the first data set and the second data set to obtain a source domain feature data set and a target domain feature data set, specifically including: The first data set and the second data set are subjected to data preprocessing, wherein the preprocessing includes removing outliers using the 3 sigma rule and performing normalization or standardization processing.

[0040] The preprocessed data set is subjected to feature extraction to form a multi-dimensional feature vector, wherein the feature extraction includes extracting the mean, standard deviation, kurtosis, skewness, charging time, accumulated power, curve slope, and curve entropy according to the timing curves of voltage and current.

[0041] The multidimensional feature vector and the health status label constitute the source domain feature dataset and the target domain feature dataset.

[0042] In the second aspect, the present invention provides a lithium battery SOH prediction device based on fusion features, which includes a training data acquisition module, a training feature extraction module, a model building module, a training module, a fusion module, a to-be-predicted data acquisition module, a to-be-predicted feature extraction module and a prediction module.

[0043] The training data acquisition module is used to acquire a first data set of the target domain and a second data set of the source domain. There are at least two source domains. The data of each source domain is constructed as a second data set.

[0044] A training feature extraction module is used to perform data preprocessing and feature extraction on the first data set and the second data set to obtain a source domain feature data set and a target domain feature data set.

[0045] The model building module is used to build a cross-domain prediction model. The cross-domain prediction model includes a predictor. In the training stage, the cross-domain prediction model also includes a generator and a discriminator.

[0046] The training module is used to train a source domain prediction model for each source domain according to source domain feature data sets of different source domains and the cross-domain prediction model.

[0047] The fusion module is used to fuse the source domain prediction models of various source domains to obtain a fusion prediction model.

[0048] The module for obtaining data to be predicted is used to obtain the data to be predicted in the target domain.

[0049] The module for extracting features to be predicted is used to preprocess and extract features from the data to be predicted to obtain features to be predicted.

[0050] The prediction module is used to input the feature to be predicted into the fusion prediction model to obtain the SOH prediction value.

[0051] In a third aspect, the present invention provides a lithium battery SOH prediction device based on fusion features, comprising a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement a lithium battery SOH prediction method based on fusion features as described in any paragraph of the first aspect.

[0052] In a fourth aspect, the present invention provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a lithium battery SOH prediction method based on fusion features as described in any paragraph of the first aspect.

[0053] By adopting the above technical solution, the present invention can achieve the following technical effects: The lithium battery SOH prediction method based on fusion features of the present invention effectively solves the cross-domain deviation problem between laboratory data and retired battery data through multi-source domain feature distribution alignment and dynamic label mapping strategy. The feature alignment mechanism based on generative adversarial network and maximum mean difference significantly reduces the distribution difference under different working conditions. The dynamic label mapping method combined with kernel density estimation realizes the accurate conversion of source domain labels to target domain range, thereby improving the adaptability of cross-domain prediction. The error screening fusion mechanism of the multi-source domain predictor further enhances the generalization ability of the model. Through the verification set relative error threshold control and weighted average strategy, it effectively integrates the advantageous features of multiple source domains and ensures the stability and accuracy of the prediction results.

[0054] This solution achieves accurate modeling of the degradation trend of the target domain through limited supervision information of anchor labels, greatly reducing the cost of acquiring full-cycle labels for retired batteries. The root mean square error of the fusion prediction model on the test set is reduced by more than 60% compared with the traditional single-source domain method, and the mean absolute error is stable within 0.5%, which is significantly better than the existing cross-domain prediction method. This efficient feature engineering and model fusion strategy provides a reliable means of health status assessment for the cascade utilization of retired lithium batteries, and has good engineering practicality while ensuring prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the specific implementation methods of the present invention. It should be understood that the following drawings only show certain specific implementation methods of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 It is a flow chart of the lithium battery SOH prediction method.

[0057] Figure 2 It is a schematic diagram of the model framework of the lithium battery SOH prediction method.

[0058] Figure 3 It is a schematic diagram of the multi-source domain fusion prediction model strategy for lithium battery SOH prediction method. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0060] Example 1, please refer to Figures 1 to 3 The first embodiment of the present invention provides a lithium battery SOH prediction method based on fusion features, which can be performed by a lithium battery SOH prediction device based on fusion features (hereinafter referred to as: prediction device). In particular, it is performed by one or more processors in the prediction device to implement steps A1 to A5, and steps S1 to S3.

[0061] A1. Obtain a first data set of a target domain and a second data set of a source domain. There are at least two source domains. The data of each source domain is constructed as a second data set.

[0062] In this embodiment, each cycle of the battery can be used as a sample to observe the change of battery aging data under different working conditions as the cycle progresses and extract effective features. Among them, one cycle of lithium battery is divided into charging stage and discharging stage. The charging stage is charged to the cut-off voltage with constant current, and then charged to the charging cut-off current with the constant voltage. The discharging stage is discharged to the discharge cut-off voltage with constant current. Each sample records the time series data of current and voltage of one cycle.

[0063] In this embodiment, laboratory degradation datasets of lithium batteries of different types or working conditions are used as a second dataset in a source domain. The measured degradation dataset of retired lithium batteries is collected as the first dataset in the target domain. The first dataset and the second dataset contain not only samples, but also labels of the samples (i.e., health status). Among them, the health status of the source domain is higher than that of the target domain. For example: the health status of the battery in the source domain is generally 1-0.8. The health status of the battery in the target domain is below 0.9 or 0.8.

[0064] A2, performing data preprocessing and feature extraction on the first data set and the second data set to obtain a target domain feature data set and a source domain feature data set. Preferably, step A2 specifically includes steps A21 to A23.

[0065] A21. Perform data preprocessing on the first data set and the second data set, wherein the preprocessing includes removing outliers using the 3 sigma rule and performing normalization or standardization.

[0066] A22. Extract features from the preprocessed data set to form a multidimensional feature vector. Feature extraction includes extracting features such as mean, standard deviation, kurtosis, skewness, charging time, cumulative power, curve slope, curve entropy, etc. from the timing curves of voltage and current, and forming these features into a multidimensional feature vector. It should be noted that a variety of feature combinations and extraction methods can also be considered, including deep extraction through 1DCNN or extraction based on timing features.

[0067] A23. The multidimensional feature vector and the health status label constitute the target domain feature data set and the source domain feature data set.

[0068] The source domain feature dataset is represented as: .

[0069] in, is the source domain feature dataset, is a set of feature vectors, For health status collection, is the feature vector set of source domain data, is the health status set of source domain data, is the feature vector extracted once, For a single extraction of health status, is the number of samples in the source domain dataset, The source domain data The extracted feature vectors The source domain data The health status of the extraction.

[0070] The target domain feature dataset is expressed as: .

[0071] in, is the target domain feature dataset, is the feature vector set of the target domain data, is the health status set of the target domain data, The source domain data The extracted feature vectors, The source domain data The health status of the extraction.

[0072] In this embodiment, the first data and the last data in the target domain feature data set are used for model training, and the remaining data are used for testing the prediction model.

[0073] A3. Construct a cross-domain prediction model. The cross-domain prediction model includes a predictor In the training phase, the cross-domain prediction model also includes a generator and the discriminator The predictor is used to predict the SOH according to the input features. The generator is used to convert the source domain features into the alignment features of the target domain. The discriminator is suitable for discriminating whether the features belong to the source domain or the target domain, and is used for adversarial training with the generator.

[0074] The predictor is used to predict the battery's SOH based on the input features. The network structure includes: a third input layer of n neurons, a first fully connected layer of 512 neurons, a third ReLU activation function layer, a third LSTM layer of 128 neurons, a second self-attention module, a second fully connected layer of 256 neurons, a fourth ReLU activation function layer, a third output layer of 1 neuron, and a second Sigmoid activation function. The second Sigmoid activation function is used to predict SOH. During model training, a third random inactivation layer with a drop rate of 0.3 is also set after the self-attention module.

[0075] The generator is used to convert the features of the source domain into the features of the target domain to increase the training data of the predictor and improve the training effect of the predictor. In order to enhance the feature generation capability, a one-dimensional convolutional neural network (Conv1D) is used to capture local features, combined with a residual network (ResNet) to prevent the gradient from disappearing in the deep network, and the self-attention mechanism is used to extract important feature information.

[0076] Generator The network structure includes the following: a first input layer of n neurons, a first one-dimensional convolutional layer with 64 neurons, 3 convolution kernels, and a convolution step size of 1, a first batch normalization layer, a first ReLU activation function layer, a first hidden layer with 64 neurons and provided with a LeakyReLU activation function layer and a batch normalization layer, a first residual module with 64 neurons, a first convolutional layer with 128 neurons, 3 convolution kernels, and a convolution step size of 2, a second batch normalization layer, a second ReLU activation function layer, a second residual module with 128 neurons, a first self-attention module, a first LSTM layer with 128 neurons, and a first output layer of n neurons. The first output layer is used to output the converted target domain features. The first residual module is used to splice the features of the first one-dimensional convolutional layer input and the features output by the first hidden layer. The second residual module is used to splice the features of the first convolutional layer input and the features output by the second ReLU activation function layer. During model training, a first random dropout layer with a post-drop rate of 0.3 is also connected after the first LSTM layer.

[0077] The discriminator is used to determine whether the input features belong to the source domain or the target domain. It combines one-dimensional convolution with LSTM to increase the ability to discriminate sequence data and local features.

[0078] Discriminator network The network structure includes: a second input layer of n neurons, a second convolutional layer with 128 neurons, 3 convolution kernels, and a convolution step size of 1, a third batch normalization layer, a first LeakyReLU layer with a negative slope of 0.2, a second LSTM layer with 128 neurons, a third convolutional layer with 256 neurons, 3 convolution kernels, and a convolution step size of 1, a fourth batch normalization layer, a second LeakyReLU layer with a negative slope of 0.2, a second output layer of 1 neuron, and a first Sigmoid activation function. The first Sigmoid activation function is used to predict whether the features of the input discriminator belong to the source domain or the target domain. During model training, a second random dropout layer with a drop rate of 0.3 is also set after the second LSTM layer.

[0079] The optimization of the generator, discriminator, and predictor of this embodiment is performed by using the Adam optimizer and the CosineAnnealingLR learning rate scheduler. Specifically, the learning rate of the generator is set to 0.0001, the learning rate of the discriminator is 0.0008, and the learning rate of the predictor is 0.0005, and L2 regularization (weight decay = 1e-5) is used to prevent overfitting. Each optimizer updates the network parameters quickly and stably by adaptively adjusting the gradient.

[0080] A LeakyReLU activation function with a slope of 0.2 was added, and Batch Normalization was applied to standardize the output of each layer to accelerate training and improve stability. In order to avoid overfitting, a random dropout layer (Dropout) was added, and the dropout rate was set to 20%. In addition, CosineAnnealingLR used Tmax=100 as a cycle to gradually reduce the learning rate, and finally made the learning rate approach 1e-6 (i.e.: ), to ensure smooth network convergence.

[0081] A4. According to the cross-domain prediction model and source domain feature data sets of different source domains, a source domain prediction model is trained for each source domain. Preferably, step A4 specifically includes steps A41 to A46.

[0082] A41. Repeat the following training steps based on the source domain feature data set of each source domain until the training is completed to obtain the source domain prediction model of each source domain. Specifically, when the number of iterations reaches the set maximum number of iterations, the training of the source domain prediction model of a single source domain is completed.

[0083] A42. Obtain a source domain feature data set of a preset proportion as a training set, wherein the first preset proportion is 80%.

[0084] A43. Input the features of the training set into the generator to obtain the alignment features of the target domain.

[0085] A44. Map the source domain label of the input feature according to the alignment feature to obtain a source domain mapping label.

[0086] The label mapping model is: .

[0087] .

[0088] In the formula, For the The source domain mapping label of the iteration, is the health status extracted for the first time from the source domain data, For the The iteration Dynamic mapping weights of samples, is the health status set of source domain data, Indicates the minimum value, Indicates the maximum value, The source domain data The health status of the extraction, The feature vector set of the features in brackets relative to the target domain data The kernel density (KDE) of The feature vector set of the features in brackets relative to the source domain data The kernel density (KDE) of For the Source domain samples, For the The iteration The features of samples after being transformed by the generator, To avoid extremely small amounts with a denominator of zero.

[0089] In this embodiment, the dynamic mapping weight is obtained by calculating the ratio of the probability density of the alignment features output by the generator to the source domain features and the target domain features. The MMD optimized generator of the iteration index Output , the feature probability density distribution of the source domain and the target domain is calculated by kernel density estimation.

[0090] A45, input the alignment features and the first and last data of the target domain feature data set into the predictor to obtain the SOH prediction result. In this embodiment, the first and last data in the target domain feature data set are used for model training. The remaining data are used for testing the fusion prediction model.

[0091] A46. According to the alignment features, the source domain mapping labels, and the SOH prediction results, the prediction model is updated by a loss function. The generator and the discriminator are combined into a feature distribution alignment module.

[0092] The loss function of the domain alignment module is: .

[0093] In the formula, is the adversarial training loss between the generator and the discriminator, is the MMD loss between the features generated by the generator and the feature vector set of the target domain data, To reconstruct the loss, is the trade-off coefficient of MMD loss, is the trade-off coefficient of reconstruction loss.

[0094] .

[0095] In the formula, Expressed as expectation, For the Source domain samples, is the source domain feature dataset, represents all source domain samples, For the target domain samples, is the target domain feature dataset, represents all target domain samples, is the feature vector set of source domain data, For the generator, For the discriminator, is the feature of the source domain after being transformed by the generator, is the discriminator’s judgment result on the converted source domain features, is the discriminator's discrimination result on the target domain features.

[0096] .

[0097] In the formula, is the number of samples in the source domain dataset, is the number of samples in the target domain dataset, is the RBF kernel function, which is used to calculate the distribution characteristics between two eigenvectors. For the Source domain samples, For the Source domain samples, For the target domain samples, For the target domain samples.

[0098] .

[0099] In the formula, is the feature of the source domain after being transformed by the generator, is the feature vector set of source domain data, is the square of the Euclidean distance.

[0100] Specifically, according to the distribution difference of the alignment features generated by the generator, the data consistency difference, and the discriminator output determination label result, the corresponding loss function is calculated and back propagated. In this embodiment, adversarial training is performed on the features generated by the generator and the output of the discriminator, so an adversarial training loss is set. In addition, by adding the MMD loss between the generated features (i.e., the source domain features after the generator conversion) and the target domain features, the distribution difference between the generated features and the target domain features is calculated. In addition, the difference between the two is measured by the reconstruction loss to ensure the consistency between the generated alignment features and the source domain features. Preferably, a heuristic algorithm can be used to initialize the loss function of the domain alignment module, and finally determine the optimal coefficient combination.

[0101] The loss function of the generator is the source domain alignment feature prediction loss.

[0102] The loss function of the generator is: ; In the formula, For the The source domain mapping label of the iteration, For the The SOH prediction results of the alignment features of the iterations, is the health status extracted for the first time from the source domain data, is the SOH prediction result of the first data of the target domain feature dataset, The source domain data The health status of the extraction, is the SOH prediction result of the last data of the target domain feature dataset, is the square of the Euclidean distance.

[0103] .

[0104] In the formula, is the feature vector set of source domain data, For the The features after the generator transformation of the iteration, For the predictor, The source domain data The extracted feature vectors, The source domain data The extracted feature vector.

[0105] A5. Fuse the source domain prediction models of each source domain to obtain a fusion prediction model.

[0106] In this embodiment, each source domain feature data set is divided into a training set and a validation set in a ratio of 8:2. Based on steps A1 to A5, the labeled feature training data set of the source domain and the first data sample data and the last sample data of the target domain are simultaneously input to train the model, and the corresponding domain adaptation network is trained for each source domain data set to construct a source domain prediction model. Then, based on the validation set, the error screening strategy and weight allocation are adopted to fuse the various source domain prediction models to obtain a fused prediction model.

[0107] The source domain prediction model and loss function for each source domain are: .

[0108] In the formula, For the SOH prediction results of source domains, For the The predictor of the source domain, For the Generators of source domains, is the source domain training dataset samples, The generator gets The source domain training dataset The alignment features of samples, is the loss function of the source domain prediction model, is the total number of samples in the source domain training dataset, For the The source domain training dataset The SOH prediction results of samples, For the The source domain training dataset The source domain mapping labels of samples.

[0109] In this embodiment, by minimizing the loss function , optimize each source domain predictor.

[0110] Preferably, step A5 specifically includes steps A51 to A55.

[0111] A51. Calculate the mean absolute error (MAE) of the source domain prediction model of each source domain on the validation set, that is, the loss function .

[0112] A52. According to the mean absolute error of each source domain, the relative difference elimination strategy is adopted to introduce the relative difference threshold , the source domain prediction models whose relative error ratio exceeds the threshold are eliminated to obtain the available model set.

[0113] The relative error ratio of all models for: .

[0114] In the formula, is the number of source domain predictors, For the The mean absolute error of the source domain predictors, For the The mean absolute error of the source domain predictors.

[0115] In this embodiment, The value is 0.25. A53. Construct available model weights based on the mean absolute error.

[0116] .

[0117] In the formula, For the The weights of the available models, For the The mean absolute error of available models, For the available model collection, For the The mean absolute error of the available models.

[0118] A54. According to the predictor weights, the source domain prediction models of the respective source domains are fused to obtain a fused prediction model.

[0119] .

[0120] In the formula, To integrate the prediction model, For the available model collection, For the The weights of the available models, For the Available models.

[0121] A55. Test the fusion prediction model using the non-anchor data of the target domain feature data set. If the test is qualified, the training is completed and the fusion prediction model is obtained. Otherwise, the training is performed again.

[0122] In the previous steps, the first and last data of the target domain feature data set are used as anchor data for training. In step A55, the remaining data of the target domain feature data set are used as non-anchor data to test the fusion prediction model. If the prediction accuracy of the fusion prediction model is greater than the preset value, the training is completed and the fusion prediction model is obtained. Otherwise, the training is repeated.

[0123] Through the previous steps, the present invention trains a lithium battery SOH prediction model based on fusion features (ie, fusion prediction model). The processing steps of the application phase of the fusion prediction model are described below. The processing steps of the model application phase include steps S1 to S3.

[0124] S1. Obtaining data to be predicted in the target domain. Specifically, the data to be predicted is time series data of current and voltage of a cycle of a battery.

[0125] S2, preprocessing and feature extraction are performed on the data to be predicted to obtain features to be predicted. Specifically, the steps of preprocessing and feature extraction are the same as step A2 in the training phase, and the present invention will not be repeated here.

[0126] S3. Input the feature to be predicted into the fusion prediction model to obtain the SOH prediction value.

[0127] The source domain data and the target domain data are collected from different battery types and different charging and discharging conditions, so they have different statistical distributions. The embodiment of the present invention reduces the difference in feature distribution between the source domain and the target domain through the training steps from step A1 to step A5, thereby using the public laboratory degradation data set to train the predictor obtained by fusion training to perform high-precision capacity estimation on the target domain data.

[0128] It should be noted that the multi-source domain fusion prediction method described in this patent is not limited to a specific combination strategy, but is used in all prediction methods based on model integration, including but not limited to the following variants: 1. Weighted average multi-model fusion prediction. 2. Stacked generalization multi-model fusion prediction. Bayesian model fusion method, etc.

[0129] The following is a specific case to illustrate the advantages and disadvantages of different SOH prediction methods.

[0130] As shown in Table 1, this example uses three laboratory public data sets of different battery types and different charging and discharging conditions, and the SOH range is between [1-0.8]. As shown in Table 2, the target domain data set uses two retired battery data, and the anchor label range is selected in SOH [0.8-0.7].

[0131] Table 1 Source domain dataset

[0132] Table 2 Target domain dataset

[0133] As shown in Table 3, the SOH prediction methods used for comparison include no domain adaptation, single source domain adaptation, and the lithium battery SOH prediction method based on fusion features (i.e., fusion domain adaptation) of an embodiment of the present invention. The evaluation indicators are root mean square error RMSE and mean absolute error MAE.

[0134] Table 3 Comparison of different prediction methods

[0135] The lithium battery SOH prediction method based on fusion features of the embodiment of the present invention integrates multiple source domain training to obtain a final fusion prediction model, which is better than the domain-free adaptive SOH prediction and single-source domain adaptive SOH prediction in most cases.

[0136] It is understandable that the prediction device can be an electronic device with computing performance, such as a portable notebook computer, a desktop computer, a server, a smart phone or a tablet computer. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0137] Embodiment 2: The second embodiment of the present invention provides a lithium battery SOH prediction device based on fusion features, which includes a training data acquisition module, a training feature extraction module, a model building module, a training module, a fusion module, a to-be-predicted data acquisition module, a to-be-predicted feature extraction module and a prediction module.

[0138] The training data acquisition module is used to acquire a first data set of the target domain and a second data set of the source domain. There are at least two source domains. The data of each source domain is constructed as a second data set.

[0139] A training feature extraction module is used to perform data preprocessing and feature extraction on the first data set and the second data set to obtain a source domain feature data set and a target domain feature data set.

[0140] The model building module is used to build a cross-domain prediction model. The cross-domain prediction model includes a predictor. In the training stage, the cross-domain prediction model also includes a generator and a discriminator.

[0141] The training module is used to train a source domain prediction model for each source domain according to source domain feature data sets of different source domains and the cross-domain prediction model.

[0142] The fusion module is used to fuse the source domain prediction models of various source domains to obtain a fusion prediction model.

[0143] The module for obtaining data to be predicted is used to obtain the data to be predicted in the target domain.

[0144] The module for extracting features to be predicted is used to preprocess and extract features from the data to be predicted to obtain features to be predicted.

[0145] The prediction module is used to input the feature to be predicted into the fusion prediction model to obtain the SOH prediction value.

[0146] Embodiment 3: The third embodiment of the present invention provides a lithium battery SOH prediction device based on fusion features, which includes a processor, a memory, and a computer program stored in the memory. The computer program can be executed by the processor to implement a lithium battery SOH prediction method based on fusion features as described in any paragraph of Embodiment 1.

[0147] Embodiment 4. The fourth embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a lithium battery SOH prediction method based on fusion features as described in any paragraph of Embodiment 1.

[0148] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method can also be implemented in other ways. The apparatus and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the apparatus, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0149] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0150] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, electronic device, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk. It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements inherent to such a process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0151] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0152] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0153] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0154] The "first\second" mentioned in the embodiments is only to distinguish similar objects, and does not represent a specific order for the objects. It is understandable that the "first\second" can be interchanged with the specific order or sequence where permitted. It should be understood that the objects distinguished by "first\second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0155] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A lithium battery SOH prediction method based on fusion features, characterized in that: Include: Obtain a first data set of a target domain and a second data set of a source domain; wherein there are at least two source domains; and data of each source domain is constructed as a second data set; Performing data preprocessing and feature extraction on the first data set and the second data set to obtain a source domain feature data set and a target domain feature data set; Constructing a cross-domain prediction model; wherein the cross-domain prediction model includes a predictor; in the training stage, the cross-domain prediction model also includes a generator and a discriminator; According to source domain feature data sets of different source domains and the cross-domain prediction model, a source domain prediction model is trained for each source domain; Fusion of source domain prediction models of various source domains to obtain a fusion prediction model; Obtain the data to be predicted in the target domain; Preprocessing and feature extraction are performed on the data to be predicted to obtain features to be predicted; The feature to be predicted is input into the fusion prediction model to obtain the SOH prediction value.

2. A lithium battery SOH prediction method based on fusion features according to claim 1, characterized in that: The predictor is used to predict SOH according to input features; Predictor The network structure includes: a third input layer of n neurons, a first fully connected layer of 512 neurons, a third ReLU activation function layer, a third LSTM layer of 128 neurons, a second self-attention module, a second fully connected layer of 256 neurons, a fourth ReLU activation function layer, a third output layer of 1 neuron, and a second Sigmoid activation function; wherein the second Sigmoid activation function is used to predict SOH; during model training, a third random inactivation layer with a dropout rate of 0.3 is also set after the self-attention module.

3. A lithium battery SOH prediction method based on fusion features according to claim 2, characterized in that: The generator is used to convert source domain features into alignment features of the target domain; The discriminator is suitable for discriminating whether a feature belongs to a source domain or a target domain, and is used for adversarial training with the generator; Generator The network structure includes: a first input layer of n neurons, a first one-dimensional convolution layer with 64 neurons, 3 convolution kernels, and a convolution step size of 1, a first batch normalization layer, a first ReLU activation function layer, a first hidden layer with 64 neurons and provided with a LeakyReLU activation function layer and a batch normalization layer, a first residual module with 64 neurons, a first convolution layer with 128 neurons, 3 convolution kernels, and a convolution step size of 2, a second batch normalization layer, a second ReLU activation function layer, a second residual module with 128 neurons, a first self-attention module, a first LSTM layer with 128 neurons, and a first output layer of n neurons, wherein the first output layer is used to output the converted target domain features; the first residual module is used to splice the features of the first one-dimensional convolution layer input and the features output by the first hidden layer; the second residual module is used to splice the features of the first convolution layer input and the features output by the second ReLU activation function layer; during model training, a first random dropout layer with a post-drop rate of 0.3 is also connected after the first LSTM layer; Discriminator network The network structure includes: a second input layer of n neurons, a second convolutional layer with 128 neurons, 3 convolution kernels, and a convolution step size of 1, a third batch normalization layer, a first LeakyReLU layer with a negative slope of 0.2, a second LSTM layer with 128 neurons, a third convolutional layer with 256 neurons, 3 convolution kernels, and a convolution step size of 1, a fourth batch normalization layer, a second LeakyReLU layer with a negative slope of 0.2, a second output layer of 1 neuron, and a first Sigmoid activation function; wherein the first Sigmoid activation function is used to predict whether the feature of the input discriminator belongs to the source domain or the target domain; during model training, a second random inactivation layer with a drop rate of 0.3 is also set after the second LSTM layer.

4. The lithium battery SOH prediction method based on fusion features according to claim 1 is characterized in that: According to source domain feature datasets of different source domains and the cross-domain prediction model, a source domain prediction model is trained for each source domain, specifically including: Repeat the following training steps based on the source domain feature data sets of each source domain until the training is completed, and obtain the source domain prediction model of each source domain; Obtain a source domain feature data set of a preset proportion as a training set; Input the features of the training set into the generator to obtain the alignment features of the target domain; According to the alignment features, the source domain labels of the input features are mapped to obtain source domain mapping labels; the mapping model is: ； In the formula, For the The mapped labels of the iterations, is the health status extracted for the first time from the source domain data, For the The iteration Dynamic mapping weights of samples, is the health status set of source domain data, Indicates the minimum value, Indicates the maximum value, The source domain data The health status of the extraction, The feature vector set of the features in brackets relative to the target domain data The kernel density (KDE) of The feature vector set of the features in brackets relative to the source domain data The kernel density (KDE) of For the Source domain samples, For the The iteration The features of samples after being transformed by the generator, To avoid extremely small quantities with zero denominators; Input the alignment feature and the first data and the last data of the target domain feature data set into the predictor to obtain the SOH prediction result; The prediction model is updated through a loss function according to the alignment features, the source domain mapping labels, and the SOH prediction results.

5. The lithium battery SOH prediction method based on fusion features according to claim 1 is characterized in that: The generator and the discriminator are combined into a feature distribution alignment module; The loss function of the domain alignment module is: ； In the formula, is the adversarial training loss between the generator and the discriminator, is the MMD loss between the features generated by the generator and the feature vector set of the target domain data, To reconstruct the loss, is the trade-off coefficient of MMD loss, is the trade-off coefficient of reconstruction loss; ； In the formula, Expressed as expectation, For the Source domain samples, is the source domain feature dataset, represents all source domain samples, For the target domain samples, is the target domain feature dataset, represents all target domain samples, is the feature vector set of source domain data, For the generator, For the discriminator, is the feature of the source domain after being transformed by the generator, is the discriminator’s judgment result on the converted source domain features, is the discriminator's discriminant result on the target domain features; ； In the formula, is the number of samples in the source domain dataset, is the number of samples in the target domain dataset, is the RBF kernel function, which is used to calculate the distribution characteristics between two eigenvectors. For the Source domain samples, For the Source domain samples, For the target domain samples, For the target domain samples; ； In the formula, is the feature of the source domain after being transformed by the generator, is the feature vector set of source domain data, is the square of the Euclidean distance; The loss function of the predictor is: ； In the formula, For the The source domain mapping label of the iteration, For the The SOH prediction results of the alignment features of the iterations, is the health status extracted for the first time from the source domain data, is the SOH prediction result of the first data of the target domain feature dataset, The source domain data The health status of the extraction, is the SOH prediction result of the last data of the target domain feature dataset, is the square of the Euclidean distance.

6. A lithium battery SOH prediction method based on fusion features according to any one of claims 1 to 5, characterized in that: The source domain prediction models of each source domain are integrated to obtain a fusion prediction model, which specifically includes: Calculate the mean absolute error of the source domain prediction model on the validation set for each source domain; According to the mean absolute error of each source domain, the relative difference elimination strategy is adopted to introduce the relative difference threshold , the source domain prediction models whose relative error ratio exceeds the threshold are eliminated to obtain the available model set; the relative error ratio of all models is for: ； In the formula, is the number of source domain predictors, For the The mean absolute error of the source domain predictors, For the The mean absolute error of the source domain predictors; constructing available model weights based on the mean absolute error; ; In the formula, For the The weights of the available models, For the The mean absolute error of available models, For the available model collection, For the Mean absolute error of available models; According to the predictor weights, the source domain prediction models of the respective source domains are fused to obtain a fused prediction model; ; In the formula, To integrate the prediction model, For the available model collection, For the The weights of the available models, For the Available models; The fusion prediction model is tested using the non-anchor data of the target domain feature data set; if the test is qualified, the training is completed and the fusion prediction model is obtained, otherwise, training is performed again.

7. A lithium battery SOH prediction method based on fusion features according to any one of claims 1 to 5, characterized in that: Performing data preprocessing and feature extraction on the first data set and the second data set to obtain a source domain feature data set and a target domain feature data set specifically includes: Performing data preprocessing on the first data set and the second data set; wherein the preprocessing includes removing outliers using the 3sigma rule and performing normalization or standardization processing; Perform feature extraction on the preprocessed data set to form a multi-dimensional feature vector; wherein the feature extraction includes extracting the mean, standard deviation, kurtosis, skewness, charging time, cumulative power, curve slope, and curve entropy according to the timing curves of voltage and current; The multidimensional feature vector and the health status label constitute the source domain feature dataset and the target domain feature dataset.

8. A lithium battery SOH prediction device based on fusion features, characterized in that: Include: A training data acquisition module, used to acquire a first data set of a target domain and a second data set of a source domain; wherein there are at least two source domains; and the data of each source domain is constructed as a second data set; A training feature extraction module is used to perform data preprocessing and feature extraction on the first data set and the second data set to obtain a source domain feature data set and a target domain feature data set; A model building module, used to build a cross-domain prediction model; wherein the cross-domain prediction model includes a predictor; during the training phase, the cross-domain prediction model also includes a generator and a discriminator; A training module, used to train a source domain prediction model for each source domain according to source domain feature data sets of different source domains and the cross-domain prediction model; A fusion module, used to fuse the source domain prediction models of various source domains to obtain a fusion prediction model; A module for acquiring data to be predicted, used for acquiring data to be predicted in the target domain; A module for extracting features to be predicted, used for preprocessing and extracting features from the data to be predicted, and obtaining features to be predicted; The prediction module is used to input the feature to be predicted into the fusion prediction model to obtain the SOH prediction value.

9. A lithium battery SOH prediction device based on fusion features, characterized in that: It comprises a processor, a memory, and a computer program stored in the memory; the computer program can be executed by the processor to implement a lithium battery SOH prediction method based on fusion features as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute a lithium battery SOH prediction method based on fusion features as described in any one of claims 1 to 7.

Images