Multi-task learning battery health state estimation method and system and computer equipment

Abstract

The invention provides a multi-task learning battery health state estimation method and system and computer equipment, and relates to the field of battery health estimation, and the method comprises the steps: obtaining a battery data set containing different types of batteries and charging and discharging protocols, and carrying out the feature extraction of the battery data set, and obtaining a feature data set; the method specifically comprises the steps that a feature data set is input into a sharing layer, a feature extraction encoder of the sharing layer converts the feature data set into sharing features, a multi-head attention mechanism is used for converting sharing feature representation into task specific features, and the task specific features are used for completing a battery health degree estimation task and a prediction confidence degree evaluation task. According to the multi-task learning battery health state estimation method and system and the computer equipment, high-precision health degree prediction can be carried out on batteries of different types and charging and discharging protocols; and confidence reference is provided for prediction of unseen feature data, and the SOH prediction reliability is improved.

Description

Technical Field

[0001] The present invention relates to the field of battery health estimation, and in particular to a multi-task learning battery state of health estimation method, system and computer device. Background Art

[0002] With the rapid development of electric vehicles and renewable energy systems, the performance requirements for battery management systems are becoming increasingly stringent. The performance of the battery management system directly affects the vehicle's endurance, charging efficiency and safety. Therefore, accurately estimating the state of health (SOH) of the battery is of great significance for improving the performance of the battery management system.

[0003] Traditional battery state of health estimation methods mainly rely on single measurement indicators, such as voltage, capacity and internal resistance. This method ignores the comprehensive influence of multiple factors on battery health and has high requirements for equipment accuracy, resulting in low estimation accuracy; constructing an equivalent circuit model using partial differential equations with physical constraints can also be used for lithium battery state of health assessment. This method is simple, practical and has low complexity, but due to the influence of many uncertain factors on this model, the reliability and accuracy of the model need to be improved.

[0004] With the continuous development of artificial intelligence, some data-driven algorithms have also been applied in battery state of health assessment. Multi-task learning (MTL), as an emerging machine learning technology, has the ability to share knowledge between different tasks and improve the performance of the model on each task. Therefore, the application of multi-task learning in battery state of health assessment is feasible. Summary of the Invention

[0005] The purpose of the present invention is to provide a multi-task learning battery state of health estimation method, system and computer device. Through the multi-task learning framework, using the data sharing ability between tasks, the number of training iterations and computing power costs are reduced, and the prediction accuracy and transfer learning ability are improved; through the battery state of health estimation task of multi-task learning, the dynamic process of battery degradation is simulated and the battery SOH is estimated, with good transfer learning ability and generalization, realizing the prediction of the state of health of different types and charge-discharge protocols of batteries; through the prediction confidence evaluation task, using the deep autoencoding Gaussian mixture model, the abnormality of feature data is identified when estimating the state of health of the battery, providing a confidence reference for the prediction of unseen feature data, and improving the prediction reliability of SOH.

[0006] To achieve the above object, the present invention provides a multi-task learning battery state of health estimation method, including the following steps:

[0007] Obtain a battery dataset covering different types of batteries and charge-discharge protocols, input the battery dataset into a deep neural network for feature extraction to obtain a feature dataset, and input the feature dataset into a multi-task learning model to complete the battery health estimation task and the prediction confidence evaluation task. The multi-task learning model includes a shared layer, a physics-informed neural network, and a deep autoencoder Gaussian mixture model, specifically including

[0008] The feature dataset is input into the shared layer. The feature extraction encoder of the shared layer converts the feature dataset into shared features, and uses the multi-head attention mechanism to convert the shared feature representation into task-specific features, and uses the task-specific features to complete the battery health estimation task and the prediction confidence evaluation task;

[0009] In the battery health estimation task, a physics neural network is used to simulate the battery degradation dynamic process to estimate the battery SOH; in the prediction confidence evaluation task, a deep autoencoder Gaussian mixture model is used to output the confidence of the SOH prediction.

[0010] Preferably, the feature dataset includes battery health features:

[0011] The number of battery cycles and the mean, standard deviation, kurtosis, skewness, charging time, cumulative charge, curve slope, and curve entropy extracted from the current and voltage curves.

[0012] Preferably, the feature extraction encoder of the shared layer converts the feature dataset into shared features, and the feature extraction function:

[0013] z = f s (p; θ s )

[0014] where: z represents the shared representation; p represents the data features in the battery dataset; θ s represents the structural parameters of the multi-layer perceptron.

[0015] Preferably, the shared feature representation is converted into task-specific features through the multi-head attention mechanism:

[0016] The multi-head attention mechanism includes four heads. Among them, the first head and the second head focus on the voltage curve segment and the current response pattern to capture the electrochemical characteristics; the third head processes the cycle time dependence; the fourth head focuses on the long-term degradation trend through the historical capacity change.

[0017] Preferably, in the battery health estimation task, a physics neural network is used to simulate the battery degradation dynamic process to estimate the battery SOH. The physics-informed neural network includes a feature-SOH prediction network and a battery degradation dynamic simulation network, including the following steps:

[0018] (1) Feature - The SOH prediction network, through a multi - layer neural network structure, is used to compress task - specific features to capture the periodic patterns and monotonic degradation trends in voltage - current features, and realizes the mapping estimation of SOH.

[0019] (2) The degradation dynamic simulation network, through a neural network with the same structure, uses the structural state - space equation with additional physical constraints to simulate the SOH degradation rate and constrain the accuracy of the mapping estimation of SOH.

[0020] Preferably, in the prediction confidence evaluation task, the deep auto - encoding Gaussian mixture model is used to output the confidence of the SOH prediction. The deep auto - encoding Gaussian mixture model includes a compression network and an estimation network, and the steps are as follows:

[0021] The compression network reduces the dimension of the input task - specific features through a deep auto - encoder, reconstructs the original data using the decoder, and extracts the low - dimensional latent representation of the task - specific features from the reduced - dimension space and the reconstruction error features.

[0022] The estimation network receives the low - dimensional latent representation, models the density distribution of the low - dimensional latent representation using the Gaussian mixture model, performs density estimation and energy calculation, and uses the energy to measure the confidence of the SOH prediction.

[0023] Preferably, the compression network reduces the dimension of the input task - specific features through a deep auto - encoder, reconstructs the original data using the decoder, and extracts the low - dimensional latent representation of the task - specific features from the reduced - dimension space and the reconstruction error features, including:

[0024] The deep auto - encoder uses multiple non - linear transformations to gradually reduce the attention weighting of the features, realizing the dimensionality reduction of the task - specific features.

[0025] The decoder reconstructs the original input space through a symmetric expansion layer, realizing the reconstruction of the original feature data.

[0026] Calculate the relative reconstruction error and the feature similarity index to obtain the reconstruction error features, and jointly form the low - dimensional latent representation of the complete task - specific features with the low - dimensional latent representation in the reduced - dimension space as the input to the estimation network.

[0027] Preferably, the estimation network receives the low - dimensional latent representation, models the density distribution of the low - dimensional latent representation using the Gaussian mixture model, performs density estimation and energy calculation, including:

[0028] Density estimation: The estimation network receives the low - dimensional latent representation output z new from the compression network, and models the density distribution of γ using a mixture Gaussian model with K components.

[0029] γ = softmax(MLP(z new ))

[0030] In the formula, Softmax(·) represents the normalization function;

[0031] Calculate the mixing parameters according to the soft assignment γ of each low-dimensional latent representation output by the estimation network;

[0032] Energy calculation: Construct a likelihood function according to the mixing parameters to calculate the sample energy value, which is used to measure the confidence of the prediction.

[0033] A multi-task learning battery state of health estimation system includes:

[0034] A dataset acquisition module, configured to acquire a battery dataset covering different types of batteries and charge-discharge protocols and input the battery dataset into a deep neural network for feature extraction to obtain a feature dataset;

[0035] A battery health estimation module, configured to input the feature dataset into a multi-task learning model to complete the battery health estimation task and the prediction confidence evaluation task.

[0036] A computer device includes a memory and a processor; the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the method are implemented.

[0037] Therefore, the present invention adopts the above-mentioned multi-task learning battery state of health estimation method, system and computer device. Through the multi-task learning framework, by utilizing the data sharing ability between tasks, the number of training iterations and the computing power cost are reduced, and the prediction accuracy and transfer learning ability are improved; through the battery health estimation task of multi-task learning, the dynamic process of battery degradation is simulated and the battery SOH is estimated, with good transfer learning ability and generalization, and the battery health prediction of different types and charge-discharge protocols is realized; through the prediction confidence evaluation task, using the deep auto-encoding Gaussian mixture model, the abnormality of feature data is identified during the battery health estimation, and a confidence reference is provided for the prediction of unseen feature data, improving the prediction reliability of the battery state of health (SOH). BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flowchart of a multi-task learning battery state of health estimation method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The technical solutions of the present invention will be further described below with reference to the drawings and embodiments.

[0040] Unless otherwise defined, the technical terms or scientific terms used in the present invention shall have the ordinary meanings understood by those of ordinary skill in the art to which the present invention belongs.

[0041] Embodiment 1

[0042] As shown Figure 1 in the figure, a multi-task learning method for estimating the state of health of a battery includes the following steps:

[0043] Step 1: Obtain a battery dataset covering different types of batteries and charge-discharge protocols, and input the battery dataset into a deep neural network for feature extraction to obtain a feature dataset;

[0044] The feature dataset includes battery health features:

[0045] The number of battery cycles and the average value, standard deviation, kurtosis, skewness, charging time, cumulative power, curve slope, and curve entropy extracted from the current and voltage curves.

[0046] Step 2: Input the feature dataset into a multi-task learning model to complete the battery health estimation task and the prediction confidence evaluation task. The multi-task learning model includes a shared layer, a physics-informed neural network, and a deep autoencoder Gaussian mixture model, specifically including

[0047] Input the feature dataset into the shared layer. The feature extraction encoder of the shared layer converts the feature dataset into shared features, uses the multi-head attention mechanism to convert the shared feature representation into task-specific features, and uses the task-specific features to complete the battery health estimation task and the prediction confidence evaluation task;

[0048] The feature extraction encoder of the shared layer converts the feature dataset into shared features. The feature extraction function:

[0049] z = f s (p; θ s )

[0050] where: z represents the shared representation; p represents the data features in the battery dataset; θ s represents the structural parameters of the multi-layer perceptron MLP, and p is 17-dimensional and z is 32-dimensional.

[0051] Convert the shared feature representation into task-specific features through the multi-head attention mechanism:

[0052] The multi-head attention mechanism includes four heads. Among them, the first head and the second head focus on the voltage curve segment and the current response mode to capture electrochemical characteristics; the third head processes the cycle time dependence; the fourth head focuses on the long-term degradation trend through the historical capacity change.

[0053] In the battery health estimation task, use a physical neural network to simulate the dynamic process of battery degradation to estimate the SOH of the battery, including the following steps:

[0054] (1) Feature - SOH Prediction Network, through a multi - layer neural network structure, is used to compress task - specific features for processing, in order to capture the periodic patterns and monotonic degradation trends in voltage - current features, and achieve the mapping estimation of SOH;

[0055] (2) Degradation Dynamics Simulation Network, through a neural network with the same structure, uses the structural state - space equation with additional physical constraints to simulate the SOH degradation rate and constrain the accuracy of the mapping estimation of SOH.

[0056] The modeling function of the Feature - SOH Mapping Network can be expressed as:

[0057] SOH = f(t, x);

[0058] Among them, t represents the number of cycles of the battery; x represents the task - specific features of the estimation task;

[0059] The modeling function of the Battery Degradation Dynamics Simulation Network can be expressed as:

[0060]

[0061] Among them, u represents the SOH of the battery; g(.) is a non - linear function describing the battery degradation dynamics; u t ,u x are the partial derivatives of u with respect to t and x respectively;

[0062] In the prediction confidence evaluation task, using the deep auto - encoding Gaussian mixture model to output the confidence of the SOH prediction, including the following steps:

[0063] The compression network reduces the dimension of the input task - specific features through a deep auto - encoder, uses the decoder to reconstruct the original data, and extracts the low - dimensional latent representation of the task - specific features from the reduced - dimension space and the reconstruction error features, including:

[0064] Through the deep auto - encoder, using multiple non - linear transformations to gradually reduce the attention weighting of the features, realizing the dimension reduction of the task - specific features;

[0065] The decoder reconstructs the original input space through a symmetric expansion layer to realize the reconstruction of the original feature data;

[0066] Calculate the relative reconstruction error and the feature similarity index to obtain the reconstruction error features, and jointly form the complete low - dimensional latent representation of the task - specific features with the low - dimensional latent representation in the reduced - dimension space as the input to the estimation network.

[0067] The estimation network receives the low - dimensional latent representation, uses the Gaussian mixture model to model the density distribution of the low - dimensional latent representation, and conducts density estimation and energy calculation, including:

[0068] Density Estimation: The estimation network receives the low - dimensional latent representation output z from the compression networknew , a Gaussian mixture model with K components is used to model the density distribution of γ;

[0069] γ = softmax(MLP(z new ));

[0070] In the formula, Softmax(·) represents the normalization function.

[0071] Calculate the mixture parameters according to the soft assignment γ of each low-dimensional latent representation output by the estimation network;

[0072]

[0073] Among them, N represents the number of mini-batches, φ k represents the mixture weight, μ k , ∑ k respectively represent the mean and covariance matrix of the k-th component, γ i represents the soft assignment of the i-th low-dimensional latent representation of γ, represents the i-th low-dimensional latent representation, and T represents the transpose;

[0074] Energy calculation: Construct a likelihood function according to the mixture parameters to calculate the sample energy value, which is used to judge the fitting degree of the new sample to the learned density distribution and measure the confidence of the SOH prediction. If the energy is high, the prediction deviates from the normal mode and the confidence is low.

[0075]

[0076] A multi-task learning battery state of health estimation system, including:

[0077] A data set acquisition module, configured to acquire a battery data set covering different types of batteries and charge and discharge protocols, and input the battery data set into a deep neural network for feature extraction to obtain a feature data set;

[0078] A battery health degree estimation module, configured to input the feature data set into a multi-task learning model to complete the battery health degree estimation task and the prediction confidence evaluation task.

[0079] The terminal device provided by an embodiment of the present invention. The terminal device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned various method embodiments are implemented. Or, when the processor executes the computer program, the functions of each module / unit in the above-mentioned various device embodiments are implemented.

[0080] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention.

[0081] The terminal device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor and a memory.

[0082] The processor can be a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), or other general-purpose processors.

[0083] The memory can be used to store the computer program and / or modules. The processor realizes various functions of the terminal device by running or executing the computer program and / or modules stored in the memory, and by calling the data stored in the memory.

[0084] If the modules / units integrated in the terminal device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be realized. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0085] Therefore, the present invention adopts the above-mentioned multi-task learning battery state of health estimation method, system and computer device. Through the multi-task learning framework, by utilizing the data sharing ability between tasks, the number of training iterations and computing power cost are reduced, and the prediction accuracy and transfer learning ability are improved; through the battery state of health estimation task of multi-task learning, the dynamic process of battery degradation is simulated and the battery SOH is estimated, which has good transfer learning ability and generalization, and realizes the prediction of the state of health of batteries of different types and charge-discharge protocols; through the prediction confidence evaluation task, by using the deep auto-encoding Gaussian mixture model, the abnormality of feature data is identified when estimating the battery state of health, and a confidence reference is provided for the prediction of unseen feature data, thereby improving the prediction reliability of SOH.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-task learning battery health state estimation method, characterized in that: The following steps are involved: Obtain a battery data set covering different types of batteries and charging and discharging protocols and input the battery data set into a deep neural network for feature extraction to obtain a feature data set. Input the feature data set into a multi-task learning model to complete the battery health estimation task and prediction confidence assessment task. The multi-task learning model includes a shared layer, a physical information neural network, and a deep autoencoder Gaussian mixture model. Specifically, The feature dataset is input into the shared layer. The feature extraction encoder of the shared layer converts the feature dataset into shared features. The multi-head attention mechanism is used to convert the shared feature representation into task-specific features. The task-specific features are used to complete the battery health estimation task and the prediction confidence assessment task. In the battery health estimation task, a physical neural network is used to simulate the dynamic process of battery degradation to estimate the battery SOH; in the prediction confidence assessment task, a deep autoencoder Gaussian mixture model is used to output the confidence of the SOH prediction.

2. A multi-task learning battery health state estimation method according to claim 1, characterized in that: The feature dataset includes battery health features: Battery cycle number and mean, standard deviation, kurtosis, skewness, charging time, cumulative charge, curve slope, and curve entropy extracted from the current and voltage curves.

3. The multi-task learning battery health state estimation method according to claim 1, characterized in that: The feature extraction encoder of the shared layer converts the feature dataset into shared features. The feature extraction function is: z=f s (p;θ s ) Where: z represents the shared representation; p represents the data features in the battery dataset; θ s Represents the structural parameters of the multi-layer perceptron.

4. The multi-task learning battery health state estimation method according to claim 1, characterized in that: The shared feature representation is transformed into task-specific features through a multi-head attention mechanism: The multi-head attention mechanism consists of four heads, where the first and second heads focus on the voltage curve segments and current response patterns to capture the electrochemical characteristics; the third head handles the cycle time dependency; and the fourth head focuses on the long-term degradation trend through historical capacity changes.

5. The multi-task learning battery health state estimation method according to claim 1, characterized in that: In the battery health estimation task, a physical neural network is used to simulate the dynamic process of battery degradation to estimate the battery SOH. The physical information neural network includes a feature-SOH prediction network and a battery degradation dynamic simulation network, which includes the following steps: The feature-SOH prediction network uses a multi-layer neural network structure to compress and process task-specific features to capture the periodic patterns and monotonic degradation trends in the voltage-current features and achieve SOH mapping estimation; The degradation dynamic simulation network simulates the SOH degradation rate through a neural network with the same structure and uses the structural state space equation to attach physical constraints to constrain the mapping estimation accuracy of the SOH.

6. The multi-task learning battery health state estimation method according to claim 1, characterized in that: In the prediction confidence evaluation task, the deep autoencoder Gaussian mixture model is used to output the confidence of the SOH prediction. The deep autoencoder Gaussian mixture model includes a compression network and an estimation network, including the following steps: The compression network reduces the dimensionality of the input task-specific features through a deep autoencoder, reconstructs the original data using a decoder, and extracts a low-dimensional potential representation of the task-specific features from the reduced dimensionality space and the reconstructed error features; The estimation network receives low-dimensional potential representations, uses a Gaussian mixture model to model the density distribution of the low-dimensional potential representations, performs density estimation and energy calculation, and uses energy to measure the confidence of SOH predictions.

7. A multi-task learning battery health state estimation method according to claim 6, characterized in that: The compression network reduces the dimensionality of the input task-specific features through a deep autoencoder, reconstructs the original data using a decoder, and extracts low-dimensional potential representations of task-specific features from the reduced dimensional space and the reconstructed error features, including: The deep autoencoder uses multiple nonlinear transformations to gradually reduce the attention weight of features, thus achieving dimensionality reduction of task-specific features. The decoder reconstructs the original input space through the symmetric expansion layer to achieve the reconstruction of the original feature data; The relative reconstruction error and feature similarity index are calculated to obtain the reconstruction error feature, which together with the low-dimensional potential representation in the dimensionality reduced space constitutes a complete low-dimensional potential representation input estimation network for task-specific features.

8. The multi-task learning battery health state estimation method according to claim 6, characterized in that: The estimation network receives the low-dimensional potential representation, uses the Gaussian mixture model to model the density distribution of the low-dimensional potential representation, and performs density estimation and energy calculation, including: Density Estimation: The estimation network receives the low-dimensional latent representation output z from the compression network new , the density distribution of γ is modeled using a mixture Gaussian model with K components; γ=softmax(MLP(z new )); In the formula, Softmax(·) represents the normalization function; Calculate the mixing parameters based on the soft assignment γ of each low-dimensional latent representation of the estimated network output; Energy calculation: Construct a likelihood function based on the mixing parameters to calculate the sample energy value to measure the confidence of the prediction.

9. A multi-task learning battery health state estimation system, characterized in that: include: A data set acquisition module is used to acquire a battery data set covering different types of batteries and charging and discharging protocols and input the battery data set into a deep neural network for feature extraction to obtain a feature data set; The battery health estimation module is used to input the feature data set into the multi-task learning model to complete the battery health estimation task and prediction confidence assessment task.

10. A computer device, characterized in that: It comprises a memory and a processor; the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images