A battery state prediction method and device based on a battery timing model

Abstract

The application discloses a battery state prediction method and device based on a battery time sequence model, and belongs to the technical fields of deep learning and battery management. The method comprises the following steps: acquiring a battery multivariate time sequence dataset, and preprocessing the battery multivariate time sequence dataset; constructing a battery time sequence model, which comprises a time-frequency feature extraction module, a multi-scale adaptive graph convolution network and a multi-scale multi-frequency fusion module; the time-frequency feature extraction module extracts time-frequency mixed features in the battery multivariate time sequence dataset, the multi-scale adaptive graph convolution network captures multi-scale features in the time-frequency mixed features, and multi-scale tensors are obtained by processing the multi-scale features, and the multi-scale multi-frequency fusion module fuses the multi-scale tensors in the frequency scale and the time scale to obtain fused features; and the battery state is predicted through the fused features. The application can effectively capture periodic patterns and non-periodic change characteristics under different time scales, and significantly improves the accuracy and robustness of battery state prediction.

Description

Technical Field

[0001] This invention relates to the fields of deep learning and battery management technology, and more specifically, to a method and apparatus for predicting battery state based on a battery time series model. Background Technology

[0002] In recent years, new energy vehicles have become a new trend in the automotive industry, with a continuously expanding market size and constant technological innovation. The core component of new energy vehicles is the power battery, whose performance directly affects the vehicle's range, charging and discharging efficiency, and lifespan. While new energy vehicles are rapidly gaining popularity, power battery condition prediction has also become a focus of industry attention. However, with the increasing complexity of battery systems, existing technologies still have many shortcomings in handling the multidimensional characteristics and complex operating conditions of batteries.

[0003] Currently, existing technologies for predicting the state of power batteries mainly include: (1) physical model-based methods, such as equivalent circuit models and thermodynamic electrochemical models. These technologies require a large amount of experimental data and complex modeling processes, making them difficult to adapt to different battery types and dynamic operating conditions. They are also highly sensitive to environmental conditions, and the prediction accuracy decreases under extreme conditions such as high or low temperatures. (2) data-driven machine learning methods, such as support vector machines and neural networks. Although these methods can capture complex relationships, they are insufficient in processing multi-scale correlations in battery data.

[0004] Power battery data typically exhibits multi-scale characteristics, such as long-term battery aging trends and short-term current fluctuation patterns. However, existing methods struggle to capture these multi-scale features simultaneously. Power battery state parameters exhibit significant periodicity, and existing state prediction methods fall short in discovering and utilizing these periodic characteristics. The time and frequency dimensions of time series data have important interactions, but existing methods often process time or frequency features individually, making it difficult to construct a synergistic effect between the two.

[0005] Chinese patent application No. 202411567374.1, published on December 27, 2024, discloses a method and system for predicting the state of charge (SOC) of a battery based on a convolutional neural network (CNN). The method includes: acquiring battery power consumption data; preprocessing the power consumption data; obtaining a first predicted SOC for the battery at each sampling time based on a dual Kalman filter algorithm and the power consumption data; combining the power consumption data and the first predicted SOC into a dataset; dividing the dataset into a training set, a validation set, and a test set; constructing a CNN model; training the CNN model using the training set to obtain a trained CNN model; validating the trained CNN model using the validation set to obtain a validated CNN model; and testing the validated CNN model using the test set to obtain a second predicted SOC for the battery at the next time step. This method can only improve the quality of the battery dataset to reduce error introduction, but it does not consider the multi-scale characteristics of battery data, thus exhibiting significant shortcomings in discovering and utilizing these multi-scale characteristics. Summary of the Invention

[0006] 1. Technical problems to be solved

[0007] To address the shortcomings in the discovery of periodic features and multi-scale characteristic modeling of battery state parameters in existing technologies, this invention provides a battery state prediction method and apparatus based on a battery time series model. By using multi-scale adaptive graph convolutional networks, discrete Fourier transforms, and orthogonal basis functions, it can better capture the multi-scale features of the battery and the periodic pattern characteristics at different time scales, effectively improving the accuracy and robustness of battery state prediction.

[0008] 2. Technical Solution

[0009] The objective of this invention is achieved through the following technical solutions.

[0010] A battery state prediction method based on a battery time-series model includes the following steps:

[0011] Obtain the battery multivariate time-series dataset and preprocess it.

[0012] A battery time series model is constructed, which includes a time-frequency feature extraction module, a multi-scale adaptive graph convolutional network, and a multi-scale multi-frequency fusion module.

[0013] The time-frequency feature extraction module extracts the time-frequency mixed features from the preprocessed battery multivariate time-series dataset; the multi-scale adaptive graph convolutional network captures the multi-scale features in the time-frequency mixed features and processes the multi-scale features to obtain multi-scale tensors; the multi-scale multi-frequency fusion module fuses the multi-scale tensors on the frequency scale and the time scale to obtain fused features.

[0014] Battery status can be predicted by fusing features.

[0015] As a further improvement of the present invention, the step of extracting time-frequency hybrid features from the preprocessed battery multivariate time-series dataset using the time-frequency feature extraction module specifically includes:

[0016] A complex spectrum is obtained by performing a discrete Fourier transform on the preprocessed battery multivariate time-series dataset. The complex spectrum includes real components and imaginary components.

[0017] Multiplying the real component with the orthogonal cosine basis function yields the time-series characteristics; multiplying the imaginary component with the orthogonal sine basis function yields the frequency-domain characteristics; and adding the time-series characteristics to the frequency-domain characteristics yields the time-frequency hybrid characteristics.

[0018] As a further improvement of the present invention, the multi-scale adaptive graph convolutional network includes a multi-head attention mechanism, an adaptive adjacency matrix generation module, and a dynamically weighted multi-scale graph convolutional network.

[0019] As a further improvement of the present invention, the multi-head attention mechanism projects the time-frequency hybrid features into the multi-head attention space and concatenates the output features generated by each attention head to obtain a concatenated tensor.

[0020] As a further improvement of the present invention, a graph attention network is introduced into the adaptive adjacency matrix generation module to capture the multi-scale dependencies between spliced ​​tensors and generate an adaptive adjacency matrix.

[0021] As a further improvement of the present invention, in the dynamically weighted multi-scale graph convolutional network, learnable dynamic weight factors are introduced into the adaptive adjacency matrices of different orders. The contribution of each graph convolution scale in feature fusion is adjusted by the learnable dynamic weight factors to generate multi-scale features.

[0022] As a further improvement of the present invention, the multi-scale features are projected back into the original feature space to generate a multi-scale tensor.

[0023] As a further improvement of this invention, in the multi-scale multi-frequency fusion module, multi-scale tensors are fused at both the frequency and time scales to obtain fused features, calculated using the following formula:

[0024]

[0025] Among them, F fusion Let f represent the fused feature, k represent the ReLU nonlinear activation function, and K represent the set of all feature sources participating in feature fusion, K = {(m, l) | 1 ≤ m ≤ M, 1 ≤ l ≤ L}, where m represents the index of the time scale, l represents the index of the frequency component, and X represents the index of the frequency component.(k) The feature at index k, w k This indicates that each feature X is assigned a value. (k) The weights are given by X, where X represents the input feature vector.

[0026] As a further improvement of the present invention, the fused features are mapped to the battery life and health index space to predict the battery state.

[0027] A battery state prediction device based on a battery time-series model includes:

[0028] The data processing module acquires the battery multivariate time-series dataset and performs preprocessing on it.

[0029] The model building module constructs a battery time-series model, which includes a time-frequency feature extraction module, a multi-scale adaptive graph convolutional network, and a multi-scale multi-frequency fusion module.

[0030] The model processing module uses the time-frequency feature extraction module to extract the time-frequency mixed features from the preprocessed battery multivariate time-series dataset; the multi-scale adaptive graph convolutional network captures the multi-scale features in the time-frequency mixed features and processes the multi-scale features to obtain multi-scale tensors; the multi-scale multi-frequency fusion module fuses the multi-scale tensors on the frequency scale and time scale to obtain fused features.

[0031] The state prediction module predicts the battery state by fusing features.

[0032] 3. Beneficial effects

[0033] Compared with the prior art, the advantages of this invention are:

[0034] This invention discloses a battery state prediction method and apparatus based on a battery time series model. By combining discrete Fourier transform with orthogonal basis functions, the time-frequency features of battery data are extracted. An adaptive graph convolutional network is introduced in multiple dimensions to dynamically weight and fuse the complex dependencies between battery data. As a result, the degradation characteristics of batteries at multiple frequencies and time scales can be accurately captured, improving the accuracy and robustness of remaining life prediction and health status assessment. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the battery timing model structure according to an embodiment of the present invention;

[0036] Figure 2 This is a flowchart illustrating the workflow of the multi-scale adaptive graph convolutional network in an embodiment of the present invention. Detailed Implementation

[0037] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0038] Example

[0039] This embodiment provides a battery state prediction method based on a battery time-series model, comprising the following steps: acquiring a battery multivariate time-series dataset and preprocessing the dataset; constructing a battery time-series model, which includes a time-frequency feature extraction module, a multi-scale adaptive graph convolutional network, and a multi-scale multi-frequency fusion module; using the time-frequency feature extraction module to extract time-frequency mixed features from the preprocessed battery multivariate time-series dataset; using the multi-scale adaptive graph convolutional network to capture multi-scale features from the time-frequency mixed features and processing the multi-scale features to obtain multi-scale tensors; using the multi-scale multi-frequency fusion module to fuse the multi-scale tensors at both the frequency and time scales to obtain fused features; and predicting the battery state using the fused features.

[0040] Specifically, in this embodiment, a multivariate time-series dataset of the battery is obtained. This dataset includes parameters such as timestamps, battery voltage, current, temperature, and remaining battery life (RUL). Further, the multivariate time-series dataset is preprocessed. Specifically, it is cleaned, formatted, and standardized to improve data quality. In addition, existing interpolation and clustering methods are used to fill in missing values, outliers are removed, and time alignment and resampling are performed on data from different sensors. Existing feature engineering is used to generate statistical, physical, and time-series features, and normalization and standardization are applied to improve the performance of the battery time-series model. In the embedding process, a sliding window is used to extract time-series subsequences, and positional embedding is combined to capture time-series features, generating a unified and efficient multivariate input representation to ensure that the subsequent battery time-series model can fully learn the dynamic behavior of the battery.

[0041] Furthermore, such as Figure 1 As shown, a battery time-series model is constructed. In this embodiment, the battery time-series model includes a time-frequency feature extraction module, a multi-scale adaptive graph convolutional network, and a multi-scale multi-frequency fusion module. It is worth noting that the battery time-series model constructed in this embodiment can efficiently extract the potential patterns in the multi-time-scale and multi-frequency information of the power battery, ultimately achieving accurate modeling of battery remaining life (RUL) prediction, health status assessment, and capacity degradation trend.

[0042] In this embodiment, the time-frequency feature extraction module is used to extract the time-frequency mixed features from the preprocessed battery multi-dimensional time-series dataset. Specific steps include: performing a discrete Fourier transform on the preprocessed battery multi-dimensional time-series dataset to obtain a complex spectrum, which includes real and imaginary components. The real and imaginary components correspond to the amplitude and phase information of the frequency domain signal at each frequency component, respectively. It should be noted that in this embodiment, the frequency domain signal refers to the battery multi-dimensional time-series data used for analysis during the time-frequency mixed feature extraction process. After the battery multi-dimensional time-series data undergoes a Fourier transform, it is converted into a frequency domain signal, where the real and imaginary components respectively contain the amplitude and phase information of the frequency domain signal. In this embodiment, the preprocessed battery multi-dimensional time-series dataset is set as... Where N represents the number of channels and T represents the time step. Then, for the preprocessed battery multivariate time-series dataset X... (norm) Perform a discrete Fourier transform on each channel to obtain the complex spectrum z∈C 1×T The real and imaginary components of the complex spectrum represent z, respectively. R and z I The calculation formula is:

[0043]

[0044] Z R =Re(z), z I =Im(z)

[0045] Among them, z [k] This represents the k-th component of the input sequence in the frequency domain representation, where k is the frequency index, ranging from 0 to T-1, corresponding to different frequency components; n is the summation index, ranging from 0 to T-1, representing the accumulation over each sampling point in the time domain; T represents the time length of the frequency domain signal; and X... (norm) [n] represents the preprocessed multi-element time-series battery data. Let z represent a complex exponential function. R z represents the real component. I It represents the imaginary component.

[0046] In this embodiment, the real component z R The amplitude information corresponds to different frequency components, while the imaginary component corresponds to the phase information of different frequency components. Thus, through the Discrete Fourier Transform, the preprocessed multivariate time-series data of the battery can be transformed from the time domain to the frequency domain, explicitly separating the periodic components in the sequence.

[0047] Furthermore, the time-series features are obtained by multiplying the real components by the orthogonal cosine basis functions, and the frequency-domain features are obtained by multiplying the imaginary components by the orthogonal sine basis functions. The time-series features and frequency-domain features are then added together to obtain the time-frequency hybrid features. Specifically, to preserve phase information in the time dimension while extracting both time-series and frequency-domain information, this embodiment introduces orthogonal cosine basis functions and orthogonal sine basis functions of time length T for the frequency-domain signal. Each frequency corresponds to a set of basis functions, expressed as follows:

[0048]

[0049] Among them, C (k,n) Let S(k,n) represent the orthogonal cosine basis functions, T represent the time length of the frequency domain signal, k represent the frequency index (k = 0, 1, 2, ..., T / 2), and n represent the time index (n = 0, 1, 2, ..., T-1).

[0050] Therefore, in this embodiment, the time-frequency mixing characteristics are obtained by multiplying the real component with the orthogonal cosine basis function and the imaginary component with the orthogonal sine basis function and then adding them together. The calculation formula is as follows:

[0051]

[0052] Among them, G [n] This represents the time-frequency hybrid characteristic, where k represents the frequency index, T represents the time step of the signal, and z represents the frequency index. R [k] represents the real component of the complex spectrum, z I [k] represents the imaginary part of the complex spectrum.

[0053] The amplitude and phase information of time series data at different frequencies are explicitly incorporated into the time series dimension. This not only preserves the original time resolution but also avoids ambiguity caused by phase inconsistency or sequence length mismatch. Therefore, in this embodiment, orthogonal cosine basis functions and orthogonal sine basis functions are used to expand the real and imaginary components of the complex spectrum, respectively. The complex spectrum obtained by the discrete Fourier transform is mapped onto the orthogonal cosine and sine bases, thereby explicitly preserving and expressing the phase information of the frequency domain signal in the time domain, while also capturing the periodicity and global trend in the frequency domain. Finally, the two are added together to obtain the time-frequency hybrid feature. The time-frequency hybrid feature carries the phase and amplitude information of different frequency components in the time dimension, which can better reveal the periodic changes and potential patterns in multivariate time series.

[0054] Furthermore, multi-scale adaptive graph convolutional networks capture multi-scale features in time-frequency mixed features and process these features to obtain multi-scale tensors. Specifically, such as... Figure 2As shown, the multi-scale adaptive graph convolutional network includes a multi-head attention mechanism, an adaptive adjacency matrix generation module, and a dynamically weighted multi-scale graph convolutional network.

[0055] It should be noted that the Discrete Fourier Transform and orthogonal basis function expansion yield time-frequency mixture features, which comprehensively preserve the information of the frequency domain signal in both the time and frequency domains in tensor form. To enhance the ability to capture cross-time series dependencies, this embodiment introduces a multi-head attention mechanism. The time-frequency mixture feature tensor is used as the input tensor, which is then projected into the multi-head attention space. The output tensors generated by each attention head are then concatenated to form a concatenated tensor, the calculation formula of which is expressed as:

[0056] Y i =Concat(head1,head2,...,head k W o

[0057]

[0058] Among them, Y i This indicates concatenating tensors; `Concat` represents the concatenation operation; `head` k W represents the output of the k-th attention head, where k represents the number of attention heads. o The output linear transformation matrix is ​​represented by Q, where Softmax represents the Softmax activation function. k Represents the query matrix, K k V represents the bond matrix. k Let d represent a value matrix, T denote the transpose operation, and d k This indicates the dimension size for each query, key, and value.

[0059] Furthermore, in this embodiment, a Graph Attention Network (GAT) is introduced into the adaptive adjacency matrix generation module to capture the multi-scale dependencies between concatenated tensors and generate an adaptive adjacency matrix. It should be noted that in this embodiment, the GAT is a graph neural network structure based on an attention mechanism, capable of updating the adjacency matrix by learning the attention weights between nodes, thereby performing more flexible feature aggregation on the graph structure. Thus, through the GAT, the relationships between nodes in the preprocessed battery multivariate time-series dataset are dynamically modeled, allowing the adjacency matrix to adaptively reflect the correlation between different nodes, further improving the ability of subsequent multi-scale graph convolutions to capture complex dependencies. In this embodiment, the GAT captures the multi-scale dependencies between output tensors and generates an adaptive adjacency matrix, the calculation formula of which is:

[0060]

[0061] in, Let represent the adaptive adjacency matrix, used to reflect the correlation strength between time series, where i represents the source node in the graph, j represents the target node, LeakyReLU represents the LeakyReLU activation function, and α represents the learnable attention vector. T This represents the transpose of the attention weight vector. and Represents different trainable node embedding matrices, used to represent the low-dimensional feature representation of nodes, and || represents the feature concatenation operation.

[0062] It is worth noting that in this embodiment, to effectively fuse the multi-scale dependencies in the preprocessed battery multivariate time-series dataset, a dynamically weighted multi-scale graph convolution method is proposed. This method involves introducing learnable dynamic weight factors into adaptive adjacency matrices of different orders, which adaptively adjusts the contribution of each graph convolution scale to feature fusion, thereby balancing short-range local information and long-range global dependency information. In this process, the battery time-series model not only becomes more robust under noise interference and abnormal data distribution conditions, but also flexibly captures complex features such as battery periodicity, trends, and abrupt changes under various dynamic operating conditions, thus significantly improving the accuracy and stability of battery remaining life prediction and health status assessment. The dynamically weighted multi-scale graph convolution network effectively overcomes the information loss and dependency deficiencies caused by a single scale, enhancing the battery time-series model's ability to mine multi-scale features of multivariate battery time-series data. In this embodiment, learnable dynamic weight factors are introduced into the dynamically weighted multi-scale graph convolution network into adaptive adjacency matrices of different orders. These learnable dynamic weight factors adjust the contribution of each graph convolution scale to feature fusion, generating multi-scale features. Multi-scale features are represented by adaptive adjacency matrices of different orders, and their fusion process is expressed as follows:

[0063]

[0064] in, Let σ represent the output tensor, σ represent the activation function, j represent the index of the matrix node, p represent the set of orders of the adaptive adjacency matrix, and α represent the output tensor. j The dynamic weights are calculated using the attention mechanism, and their formula is as follows:

[0065]

[0066] e j =w T ReLU((A i ) j Y i )

[0067] Here, w represents a learnable parameter used to dynamically model the importance of multi-scale features.

[0068] After multi-scale graph convolution, the multi-scale features obtained by each network layer are high-dimensional embeddings with information at different scales. Therefore, it is necessary to project these multi-scale features back into the original feature space to generate a multi-scale tensor. In this embodiment, the original feature space is the feature representation space of the battery multivariate time-series dataset before processing by the deep learning model. These original features include multiple time-series data parameters such as battery voltage, current, temperature, and remaining lifetime (RUL). Projecting the multi-scale features back into the original feature space ensures that the dimensionality is consistent with the original battery multivariate time-series data. This helps ensure that the output features are consistent with the requirements of the battery state prediction task during the battery time-series model output stage. Especially when dealing with practical applications such as remaining lifetime prediction and health status assessment, it avoids misalignment problems caused by the inconsistency between the high-dimensional features of the battery time-series model and the dimensionality of the task objective.

[0069] Specifically, multi-scale features are projected back into the original feature space to ensure consistency between the final output and the target dimension of subsequent prediction tasks. This projection process preserves the key features extracted by multi-scale convolutions while ensuring the battery time-series model output matches the original input or target task requirements, avoiding misalignment between high-dimensional embeddings and real-world application scenarios. In this embodiment, the multi-scale features are projected back into the original feature space using the following formula:

[0070]

[0071] in, The i-th feature is projected back into the original feature space. MLP stands for Multilayer Perceptron, which contains one or more fully connected layers and incorporates nonlinear activation functions to enhance the expressive power of the output features.

[0072] Therefore, in this embodiment, the multi-scale adaptive graph convolutional network combines multi-head attention mechanism, adaptive adjacency matrix generation and dynamically weighted multi-scale graph convolution, which can effectively capture the spatiotemporal features of complex time series data and provide higher prediction accuracy and robustness.

[0073] After completing the multi-scale adaptive graph convolutional network, a multi-scale tensor is obtained. This multi-scale tensor aggregates the embedding representations of different time scales and multi-dimensional feature relationships from the multi-layer network. To enable this multi-scale tensor to better highlight key features and suppress useless noise in subsequent prediction tasks, in this embodiment, the multi-scale tensor is fused at both the frequency and time scales to obtain fused features. The calculation formula is as follows:

[0074]

[0075] Among them, F fusion Let X represent the fused feature, f represent the ReLU nonlinear activation function, k represent the index used to traverse the different feature sources participating in the fusion, and K represent the set of all feature sources participating in feature fusion, K = {(m, l) | 1 ≤ m ≤ M, 1 ≤ l ≤ L}, where m represents the index of the time scale, l represents the index of the frequency component, and X represents the index of the frequency component. (k) The feature at index k, w k This indicates that each feature X is assigned a value. (k) The weights are given by X, where X represents the input feature vector.

[0076] Therefore, by mapping the fused features to the battery life and health indicator space for battery state prediction, this embodiment can directly generate remaining lifetime (RUL) predictions and capacity degradation trend results at the output layer of the battery time-series model. On the one hand, the deep features extracted by the time-frequency hybrid features and the multi-scale graph convolutional network are aligned with known lifetime, capacity, or health status labels and subjected to supervised training, thereby achieving accurate mapping from complex multivariate data under various operating conditions to degradation and lifetime indicators. On the other hand, the introduction of dynamic weighting and adaptive adjacency matrices effectively enhances the adaptability of the battery time-series model to the battery operating environment and multi-dimensional correlations, enabling it to quickly provide accurate and stable assessments of remaining lifetime, capacity degradation trends, and health status during actual inference, providing more targeted decision-making basis for the monitoring and early warning of the battery management system.

[0077] This embodiment provides a battery state prediction method based on a battery time-series model. It combines the traditional Discrete Fourier Transform with orthogonal basis functions to explicitly preserve phase information in the time domain and generate a hybrid representation that incorporates both time and frequency characteristics. The time-domain representation preserves the signal's temporal order and local dynamic information, while the frequency-domain representation is better at capturing periodicity and global trends. This hybrid representation method comprehensively reflects the signal's information in both time and frequency dimensions within a unified feature space, thus revealing more comprehensively the multi-scale dynamic changes and periodic characteristics of battery data. Furthermore, the use of a multi-scale adaptive graph convolutional network effectively captures the multi-scale dependencies and spatiotemporal features in the power battery time-series data, providing higher prediction accuracy and robustness.

[0078] This embodiment also provides a battery state prediction device based on a battery time-series model, including a data processing module, a model building module, a model processing module, and a state prediction module. The data processing module acquires and inputs a battery multivariate time-series dataset and preprocesses it. The model building module constructs a battery time-series model, which includes a time-frequency feature extraction module, a multi-scale adaptive graph convolutional network, and a multi-scale multi-frequency fusion module. The model processing module uses the time-frequency feature extraction module to extract time-frequency mixed features from the preprocessed battery multivariate time-series dataset; the multi-scale adaptive graph convolutional network captures multi-scale features from the time-frequency mixed features and processes these features to obtain multi-scale tensors; the multi-scale multi-frequency fusion module fuses the multi-scale tensors at both the frequency and time scales to obtain fused features. The state prediction module predicts the battery state using the obtained fused features. This embodiment provides a battery state prediction device based on a battery timing model, capable of implementing any of the battery state prediction methods based on battery timing models. The specific working process of this device can be referenced from the corresponding process in the embodiments of the battery state prediction methods based on battery timing models. The methods and devices provided in this embodiment can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for example, the division of a module is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the connections or communication links shown or discussed may be indirect couplings or communication links through some interfaces, devices, or units, or they may be electrical, mechanical, or other forms of connection.

[0079] This embodiment also provides a computer device. A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned battery state prediction method based on a battery timing model.

[0080] This embodiment also provides a computer-readable storage medium. A computer-readable storage medium stores a computer program, which, when executed by a processor, performs the battery state prediction method based on a battery timing model as described in this embodiment. The computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0081] The present invention and its embodiments have been described above illustratively. This description is not restrictive, and the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. The accompanying drawings are only one embodiment of the present invention, and the actual structure is not limited thereto. No reference numerals in the claims should limit the scope of the claims. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the scope of protection of the present invention. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Multiple elements stated in the product claims may also be implemented by a single element through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A battery state prediction method based on a battery time-series model, comprising the following steps: Obtain the battery multivariate time-series dataset and preprocess it. A battery time series model is constructed, which includes a time-frequency feature extraction module, a multi-scale adaptive graph convolutional network, and a multi-scale multi-frequency fusion module. The time-frequency hybrid features in the preprocessed battery multivariate time-series dataset are extracted using the time-frequency feature extraction module. A multi-scale adaptive graph convolutional network is used to project time-frequency hybrid features into a multi-head attention space, and the output features generated by each attention head are concatenated to obtain a concatenated tensor. A graph attention network is introduced to capture the multi-scale dependencies between the concatenated tensors and generate an adaptive adjacency matrix. Learnable dynamic weight factors are introduced into the adaptive adjacency matrices of different orders, and the contribution of each graph convolution scale in feature fusion is adjusted by the learnable dynamic weight factors to generate multi-scale features. Finally, the multi-scale features are processed to obtain a multi-scale tensor. By using a multi-scale and multi-frequency fusion module, the multi-scale tensor is weighted and summed on both the frequency and time scales to obtain the fused features. Battery status can be predicted by fusing features.

2. The battery state prediction method based on a battery time-series model according to claim 1, characterized in that, The step of extracting time-frequency hybrid features from the preprocessed battery multivariate time-series dataset using the time-frequency feature extraction module includes: A complex spectrum is obtained by performing a discrete Fourier transform on the preprocessed battery multivariate time-series dataset. The complex spectrum includes real components and imaginary components. Multiplying the real component with the orthogonal cosine basis function yields the time-series characteristics; multiplying the imaginary component with the orthogonal sine basis function yields the frequency-domain characteristics; and adding the time-series characteristics to the frequency-domain characteristics yields the time-frequency hybrid characteristics.

3. The battery state prediction method based on a battery time-series model according to claim 2, characterized in that, The multi-scale adaptive graph convolutional network includes a multi-head attention mechanism, an adaptive adjacency matrix generation module, and a dynamically weighted multi-scale graph convolutional network.

4. The battery state prediction method based on a battery time-series model according to claim 3, characterized in that, The multi-scale features are projected back into the original feature space to generate a multi-scale tensor.

5. The battery state prediction method based on a battery time-series model according to claim 4, characterized in that, In the multi-scale, multi-frequency fusion module, the multi-scale tensors are weighted and summed at both the frequency and time scales to obtain the fused features. The calculation formula is as follows: Among them, F fusion Let f denote the fused features, k denote the ReLU nonlinear activation function, k denote the index, and K denote the set of all feature sources participating in feature fusion. m represents the index of the time scale, l represents the index of the frequency component, and X (k) The feature at index k, w k This indicates that each feature X is assigned a value. (k) The weights are given by X, where X represents the input feature vector.

6. The battery state prediction method based on a battery time-series model according to claim 5, characterized in that, The fused features are mapped to the battery life and health indicator space to predict battery status.

7. A battery state prediction device based on a battery time-series model, characterized in that, include: The data processing module acquires the battery multivariate time-series dataset and performs preprocessing on it. The model building module constructs a battery time-series model, which includes a time-frequency feature extraction module, a multi-scale adaptive graph convolutional network, and a multi-scale multi-frequency fusion module. The model processing module extracts time-frequency mixed features from the preprocessed battery multivariate time-series dataset using the time-frequency feature extraction module; it projects these mixed features onto a multi-head attention space using a multi-scale adaptive graph convolutional network and concatenates the output features generated by each attention head to obtain a concatenated tensor; it introduces a graph attention network to capture the multi-scale dependencies between the concatenated tensors and generates an adaptive adjacency matrix; it introduces learnable dynamic weight factors into adaptive adjacency matrices of different orders, adjusting the contribution of each graph convolution scale to feature fusion to generate multi-scale features; and it processes these multi-scale features to obtain a multi-scale tensor. By using a multi-scale and multi-frequency fusion module, the multi-scale tensor is weighted and summed on both the frequency and time scales to obtain the fused features. The state prediction module predicts the battery state by fusing features.

Images