Battery charge state prediction method and device based on multi-scale interactive fusion

Abstract

The invention provides a battery charge state prediction method and device based on multi-scale interactive fusion. The method comprises the following steps: carrying out convolution operation on voltage data of each time period by utilizing a plurality of convolution kernels of different scales to obtain a plurality of feature sets of different scales, constructing a feature graph structure, processing the feature graph structure to obtain spatial features of different scales, carrying out feature interaction fusion on the spatial features, and obtaining a feature fusion result; multi-scale space interactive fusion features are obtained; according to the node features of different scales corresponding to each time period, constructing a time graph structure of each time period, processing the time graph structure to obtain a time feature corresponding to each time period, and generating an attention coefficient of each time period; and carrying out weighted enhancement on the multi-scale space interaction fusion features by using the attention coefficient, and carrying out real-time prediction on the state of charge of the battery based on the enhanced features. The prediction accuracy and the real-time prediction efficiency of the state of charge of the battery are improved, so that the energy management efficiency of the vehicle is improved.

Description

Technical Field

[0001] The present application relates to the field of new energy vehicle technology, and in particular to a battery state of charge prediction method and device based on multi-scale interactive fusion. Background Art

[0002] The key to the energy management strategy of a programmable car is to intelligently allocate the energy flow between the battery and the engine to optimize the endurance of the entire system. This strategy not only extends the car's driving distance but also improves endurance by balancing the use of the battery with the operation of the engine. In this process, accurately estimating the battery's state of charge (SOC) is crucial because it directly affects the management and allocation strategy of the energy flow, ensuring that the battery can provide enough energy when needed and avoiding unnecessary charging or discharging.

[0003] At present, the prediction of SOC mainly relies on model-based methods and data-driven methods. Model-based methods directly use physical or chemical models to estimate SOC, which requires high accuracy of model parameters; while data-driven methods, especially deep learning techniques such as one-dimensional convolutional neural networks (1DCNN), estimate SOC by learning the local features of battery parameters such as current, voltage and temperature that change over time. 1D CNN captures local changes in battery parameters through the sliding window operation of the convolution kernel, effectively extracting local patterns and structural features in the signal.

[0004] Although deep learning methods have made progress in SOC estimation, these methods often require complex models and algorithms, which increase computational costs and processing delays. In addition, 1D CNN uses fixed-size convolution kernels, which limits its ability to extract features at different scales. Since the size of the convolution kernel determines the scale range of the observed signal, a single-scale convolution kernel is difficult to capture multi-scale feature information. More importantly, 1D CNN cannot effectively process signals with non-Euclidean distances, making it difficult to mine the potential spatial relationship between signals, which greatly limits the prediction performance of the model. Summary of the invention

[0005] In view of this, an embodiment of the present application provides a battery state of charge prediction method and device based on multi-scale interactive fusion to solve the problems of low computational efficiency and low accuracy of state of charge prediction in the prior art, which leads to reduced vehicle energy management efficiency.

[0006] In a first aspect of an embodiment of the present application, a battery state of charge prediction method based on multi-scale interactive fusion is provided, including: obtaining voltage data of a vehicle battery, and segmenting the voltage data according to a preset time window; performing convolution operations on the voltage data of each time period using convolution kernels of multiple different scales to obtain feature sets of multiple different scales; constructing a feature graph structure corresponding to each scale according to the feature set, processing the feature graph structure using a first graph convolution network to obtain spatial features between nodes at different scales, and interactively fusing the spatial features of different scales to obtain multi-scale spatial interactive fusion features; constructing a time graph structure corresponding to each time period according to node features of different scales corresponding to each time period, processing the time graph structure using a second graph convolution network to obtain time features corresponding to each time period, generating an attention coefficient for each time period according to the time features, wherein the node features are features obtained after convolution of the voltage data; weightedly enhancing the multi-scale spatial interactive fusion features using the attention coefficients to obtain enhanced features, and performing real-time prediction of the battery state of charge of the vehicle based on the enhanced features.

[0007] According to a second aspect of an embodiment of the present application, a battery state of charge prediction device based on multi-scale interactive fusion is provided, comprising: an acquisition module, configured to acquire voltage data of a vehicle battery, and segment the voltage data according to a preset time window; a convolution module, configured to perform a convolution operation on the voltage data of each time period using convolution kernels of multiple different scales to obtain a feature set of multiple different scales; a fusion module, configured to construct a feature graph structure corresponding to each scale according to the feature set, process the feature graph structure using a first graph convolution network, obtain spatial features between nodes at different scales, and perform feature interactive fusion on the spatial features of different scales to obtain multi-scale spatial interactive fusion features; a generation module, configured to construct a time graph structure corresponding to each time period according to node features of different scales corresponding to each time period, process the time graph structure using a second graph convolution network, obtain time features corresponding to each time period, and generate an attention coefficient for each time period according to the time features, wherein the node features are features obtained after convolution of the voltage data; a prediction module, configured to perform weighted enhancement on the multi-scale spatial interactive fusion features using the attention coefficients to obtain enhanced features, and perform real-time prediction of the battery state of charge of the vehicle based on the enhanced features.

[0008] According to a third aspect of an embodiment of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0009] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0010] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects:

[0011] By obtaining the voltage data of the vehicle battery, and processing the voltage data in segments according to the preset time window; using multiple convolution kernels of different scales to perform convolution operations on the voltage data of each time period, obtaining multiple feature sets of different scales; according to the feature set, constructing the feature graph structure corresponding to each scale, using the first graph convolution network to process the feature graph structure, obtaining the spatial features between nodes at different scales, and interactively fusing the spatial features of different scales to obtain multi-scale spatial interactive fusion features; according to the node features of different scales corresponding to each time period, constructing the time graph structure corresponding to each time period, using the second graph convolution network to process the time graph structure, obtaining the time features corresponding to each time period, generating the attention coefficient of each time period according to the time features, wherein the node features are the features obtained after convolution of the voltage data; using the attention coefficient to weighted enhance the multi-scale spatial interactive fusion features, obtaining enhanced features, and performing real-time prediction of the battery state of charge of the vehicle based on the enhanced features. This application improves the prediction accuracy and real-time prediction efficiency of the battery state of charge, thereby improving the energy management efficiency of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0013] Figure 1 It is a flowchart of a battery state of charge prediction method based on multi-scale interactive fusion provided in an embodiment of the present application;

[0014] Figure 2 is a schematic diagram of a multi-scale feature extraction process provided in an embodiment of the present application;

[0015] Figure 3 is a schematic diagram of a multi-scale spatial feature aggregation process provided in an embodiment of the present application;

[0016] Figure 4 is a schematic diagram of a multi-scale temporal attention generation process provided by an embodiment of the present application;

[0017] Figure 5It is a structural schematic diagram of a battery state of charge prediction device based on multi-scale interactive fusion provided in an embodiment of the present application;

[0018] Figure 6 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0020] The energy management strategy of an extended-range vehicle can intelligently manage the energy flow between the battery and the engine to maximize the range of the entire system. By properly allocating energy and balancing the use of the battery and the engine, the driving distance of the extended-range vehicle can be extended and a longer range can be provided. Accurately estimating the state of charge (SOC) of the battery is critical to optimizing the energy management system. By accurately estimating the difference between the current SOC and the predicted SOC, the energy management system can optimize the energy flow and allocation strategy to ensure that the battery provides sufficient energy when needed and avoid unnecessary charging or discharging. The prediction of SOC is usually modeled and predicted using parameters such as current, voltage, and temperature.

[0021] At present, SOC estimation methods are mainly divided into two types: model-based and data-driven. Model-based methods directly use models to estimate SOC, but have high requirements on model parameters. Data-driven methods have shorter design time than model-based methods. Although deep learning-based methods have made progress in SOC estimation, accurate prediction of SOC usually requires more complex models and algorithms, which increases computational costs and delays.

[0022] One-dimensional convolutional neural network (1D CNN) can effectively capture the local features of battery parameters changing over time through the sliding window operation of the convolution kernel. The convolution kernel performs convolution operation with the input signal at each position to extract the local features corresponding to the position. This local perception ability enables 1DCNN to extract features sensitive to local patterns, changes and structures in the signal. However, the size of the convolution kernel determines the scale range that the model observes in the signal. Using only one size of convolution kernel cannot obtain feature information at different scales.

[0023] In addition, 1DCNN cannot process non-Euclidean distance signals, resulting in the inability to extract the hidden spatial relationship between signals, which greatly limits the prediction performance of the model. Graph Convolutional Neural Network (GCN) is a promising method for time series prediction with spatiotemporal feature data. By deeply mining the interdependencies between feature nodes in GCN to reflect the spatial features between signals, hidden information between different scales can be well mined.

[0024] In view of the problems existing in the prior art, this application proposes a SOC prediction method based on multi-scale interactive fusion to address the problem that the existing SOC prediction model cannot obtain feature information at different scales and ignores the relationship between multi-scale features, resulting in limited prediction accuracy. The main contents of the technical solution of this application include: first, the battery voltage parameters of the extended-range electric vehicle under different driving conditions are selected, and then the voltage signal is divided into n segments according to a time window of 10s. Secondly, three convolution kernels of different sizes are used to extract multi-scale features for each voltage signal, and then each voltage signal at each scale is used as a node feature, and the spatial features of all time periods at each scale are extracted through GCN, and then the three spatial features after aggregation of the three scales are obtained, and then the products are interactively multiplied and added to obtain the final multi-scale spatial interactive fusion. Next, a multi-scale temporal attention enhancement module was designed. The features of different scales in the same time period were used as node features. The features between different scales in the same time period were mined, and then the attention coefficient was generated. The multi-scale spatial interactive fusion features obtained above were enhanced on the time channel. The enhanced features were finally used to output the predicted SOC, thereby improving the accuracy of SOC prediction and laying the foundation for optimizing the energy flow and distribution strategy of the extended-range vehicle energy management strategy.

[0025] The contents of the technical solution of the present application are described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Figure 1 It is a flow chart of a battery state of charge prediction method based on multi-scale interactive fusion provided in an embodiment of the present application. Figure 1 The battery state of charge prediction method based on multi-scale interactive fusion can be executed by the vehicle battery management system. Figure 1 As shown, the battery state of charge prediction method based on multi-scale interactive fusion may specifically include:

[0027] S101, obtaining voltage data of a vehicle battery, and processing the voltage data in segments according to a preset time window;

[0028] S102, using a plurality of convolution kernels of different scales to perform a convolution operation on the voltage data of each time period to obtain a plurality of feature sets of different scales;

[0029] S103, constructing a feature graph structure corresponding to each scale according to the feature set, processing the feature graph structure using the first graph convolutional network to obtain spatial features between nodes at different scales, and interactively fusing the spatial features of different scales to obtain multi-scale spatial interactive fusion features;

[0030] S104, constructing a time graph structure corresponding to each time period according to node features of different scales corresponding to each time period, processing the time graph structure using a second graph convolutional network to obtain a time feature corresponding to each time period, and generating an attention coefficient for each time period according to the time feature, wherein the node feature is a feature obtained after convolution of the voltage data;

[0031] S105, using the attention coefficient to perform weighted enhancement on the multi-scale spatial interactive fusion features to obtain enhanced features, and based on the enhanced features, perform real-time prediction of the battery state of charge of the vehicle.

[0032] The embodiment of the present application provides a method for predicting the state of charge of an extended-range electric vehicle. The method focuses on extracting key information from the voltage data of the vehicle battery and predicting the state of charge (hereinafter referred to as SOC) of the battery through an intelligent algorithm. Implementing this method can not only improve the efficiency of energy management, but also optimize the overall performance of the vehicle.

[0033] First, data collection and processing includes collecting battery voltage data from extended-range electric vehicles. This data includes changes in battery voltage under different driving conditions (such as acceleration, deceleration, and constant speed driving). Data collection can be achieved through the vehicle's built-in sensor system, which can monitor and record the dynamic changes in battery voltage in real time.

[0034] Then, the collected voltage data is segmented according to the preset time window. In this embodiment, the selected time window is 10 seconds. That is to say, from the continuously recorded voltage data stream, data is extracted every 10 seconds to form an independent data segment. This segmentation method allows the system to process more detailed time series data, thereby capturing the voltage fluctuations of the battery under different working conditions.

[0035] In some embodiments, a convolution operation is performed on the voltage data of each time period using convolution kernels of multiple different scales to obtain feature sets of multiple different scales, including:

[0036] The first-scale convolution kernel, the second-scale convolution kernel and the third-scale convolution kernel are used to perform convolution operations on the voltage data of each time period, respectively, to obtain the voltage signal features corresponding to each time period at different scales, and the voltage signal features are used to generate the first-scale feature set, the second-scale feature set and the third-scale feature set. The feature sets of different scales are adaptively pooled to obtain the feature sets after adaptive pooling processing.

[0037] Specifically, this embodiment describes in detail a multi-scale spatial interactive fusion module for predicting the state of charge (SOC) of the battery of an extended-range electric vehicle. The module uses three different sizes of convolution kernels to extract features from voltage data and optimizes feature representation through adaptive pooling processing to enhance the performance and accuracy of the prediction model.

[0038] The multi-scale feature extraction process is described in detail below with reference to the accompanying drawings and specific embodiments. Figure 2 is a schematic diagram of the multi-scale feature extraction process provided in the embodiment of the present application, such as Figure 2 As shown, in the embodiment of the present application, the process of performing convolution operation on the voltage data of each time period using convolution kernels of different scales to obtain a multi-scale feature set may include the following:

[0039] First, continuous voltage data is collected from the extended-range electric vehicle. This data records the voltage changes of the vehicle under different driving conditions. The continuous voltage data is segmented according to the preset 10-second time window. This segmentation method can ensure the time sensitivity of data processing, so that feature extraction can reflect the dynamic changes of battery status in a short time.

[0040] Furthermore, this embodiment designs three scales of convolution kernels: small scale (10×1), medium scale (20×1), and large scale (30×1). Each scale of convolution kernel targets different feature granularity, ranging from details to global features. Small scale convolution kernels focus on capturing fast detail changes in voltage signals, medium scale convolution kernels balance local and global information, and large scale convolution kernels focus on the overall trend of voltage signals.

[0041] Furthermore, the three convolution kernels are applied to the voltage data of each time period. This process is implemented by a one-dimensional convolutional neural network (1dCNN), and the convolution operation of each scale is performed independently to ensure the diversity and richness of the features. The extracted features are divided into three groups according to the scale: the first scale feature set (X 1n ), the second scale feature set (X 2n ), the third scale feature set (X 3n ).

[0042] Furthermore, adaptive pooling is performed on the three scale feature sets for each period of time. This step aims to unify the dimensions of different convolution outputs so that the features of each period of time are consistent in dimension, which is convenient for subsequent fusion and analysis. The pooling operation adjusts the length of the feature vector to match the preset target dimension, ensuring the compatibility and consistency of the features in subsequent processing.

[0043] For example, in a specific example, Figure 2 As shown in FIG. 1 , this embodiment designs a multi-scale spatial interactive fusion module. First, three convolution kernels of different sizes are used to extract multi-scale features for each voltage signal. 1dCNN_1 uses a small convolution kernel (10×1), 1dCNN_2 uses a medium-sized convolution kernel (20×1), and 1dCNN_3 uses a large convolution kernel (30×1) to automatically extract signal features at different scales, thereby obtaining a small-scale feature denoted as X. 1n =[X 11 ,X 12 ,…,X 1n ], the medium-scale feature is recorded as X 2n = = [X 21 ,X 22 ,…,X 2n ], the large-scale feature is recorded as X 3n =[X 31 ,X 32 ,…,X 3n ], where n generation corresponds to the n-segment voltage signal divided as above.

[0044] Each segment of small-scale features will contain more detailed features, and each segment of large-scale features contains fewer features, but is rich in more global features, which can better capture the overall or global feature information. For the features at the medium scale, it will make up for the feature information ignored at the small and large scales. However, the above operation will cause the feature dimensions in each segment to be different. Therefore, this embodiment keeps the feature dimensions extracted at different scales consistent for each segment through adaptive pooling control, which is convenient for subsequent operations.

[0045] Through the method of the present embodiment described above, the multi-scale feature extraction and adaptive pooling processing of the present embodiment improves the applicability of the features and the accuracy of the prediction model. By combining information at different scales, the model can more comprehensively understand the dynamic changes in battery voltage, improve the efficiency of energy management and the performance of the vehicle. In summary, the present embodiment provides an efficient method for processing and analyzing the voltage data of the extended-range electric vehicle battery, and optimizes the SOC prediction process through multi-scale analysis and intelligent algorithms.

[0046] In some embodiments, a feature graph structure corresponding to each scale is constructed according to the feature set, including:

[0047] The voltage signal features corresponding to each time period in the feature set after adaptive pooling processing are taken as nodes, and the Euclidean distance matrix between nodes is taken as the adjacency matrix. The nodes, adjacency matrices and node feature sets corresponding to feature sets of different scales are used to construct the feature graph structure corresponding to each scale.

[0048] Specifically, this embodiment also describes in detail how to use a multi-scale feature graph structure to predict the state of charge (SOC) of a range-extended electric vehicle battery. This method achieves in-depth analysis of voltage data by constructing a graph structure based on features of different scales to improve the accuracy of SOC prediction.

[0049] Using the method described previously, features of different scales are extracted from the voltage data. These features are obtained by convolution kernels of three different sizes, namely small scale (10×1), medium scale (20×1) and large scale (30×1). The features extracted in each time period are adaptively pooled to unify the length of the feature vector to ensure consistency and comparability when constructing the graph structure.

[0050] Furthermore, for each scale feature set, an independent graph structure is constructed. In each graph, the voltage feature of each time period is taken as a node, and the feature set of the node corresponds to the pooled voltage feature of the time period. The Euclidean distance between each node is calculated to form an adjacency matrix.

[0051] Furthermore, the calculated Euclidean distance matrix is ​​used as the adjacency matrix of the small-scale, medium-scale and large-scale graphs respectively. Define each graph structure, the small-scale graph structure G 1 =(V n ,A 1 ,X 1n ), medium-scale graph structure G 2 =(V n ,A 2 ,X 2n ), large-scale graph structure G 3 =(V n ,A 3 ,X 3n ), where V represents the node set, A represents the adjacency matrix, and X represents the node feature set.

[0052] For example, in a specific example, for the small-scale feature X 1n =[X 11 ,X 12 ,…,X 1n ], medium-scale feature X 2n = = [X 21 ,X 22 ,…,X 2n ], large-scale feature X3n =[X 31 ,X 32 ,…,X 3n ] The graph structure is constructed respectively, and the graph structure is represented by G = (V, A, X), where V represents the set of nodes in the graph, A represents the adjacency matrix of the graph G, and X represents the feature corresponding to the node. At each scale, each voltage signal is used as a node feature. With n voltage signals, n nodes and their corresponding node features can be obtained. Then, by calculating the Euclidean distance of each node feature vector at three scales, the small-scale Euclidean distance matrix E can be obtained. 1 , medium-scale Euclidean distance matrix E 2 , the large-scale Euclidean distance matrix E 3 Obviously, E 1 、E 2 and E 3 is a symmetric matrix, each element E of the matrix ij represents the Euclidean distance between nodes i and j.

[0053]

[0054] E 1 、E 2 and E 3 As the adjacency matrix A of small scale, medium scale and large scale respectively 1 , A 2 and A 3 Therefore, this embodiment can obtain the graph structure of all time periods at different scales: small-scale graph structure G 1 =(V n ,A 1 ,X 1n ), medium-scale graph structure G 2 =(V n ,A 2 ,X 2n ), large-scale graph structure G 3 =(V n ,A 3 ,X 3n ), where V n Represents a set of nodes. Since there are n time periods at each scale, the number of nodes is the same.

[0055] Through the method of the present embodiment described above, the constructed graph structure is utilized and a graph convolutional network (GCN) is applied to further extract and fuse features. This enables the model to deeply understand and utilize the spatial relationships in the voltage data and improve the expressiveness of the features. The SOC is predicted using the features processed by the graph convolutional network. This graph-structure-based method can provide more accurate prediction results than traditional methods because it integrates voltage change information at different times and scales. This embodiment optimizes the accuracy and efficiency of SOC prediction through refined feature processing and advanced graph convolution technology. In addition, the use of a multi-scale feature graph structure provides a comprehensive analysis framework for the battery management system, enhancing the adaptability of the prediction model to different driving conditions.

[0056] In some embodiments, the feature graph structure is processed using the first graph convolutional network to obtain spatial features between nodes at different scales, and the spatial features at different scales are interactively fused to obtain multi-scale spatial interactive fusion features, including:

[0057] The adjacency matrix in the feature graph structure of each scale is symmetrically normalized, and the adjacency matrix after symmetrical normalization is multiplied with the feature matrix and weight matrix of the node to obtain aggregated features. The aggregated features are processed using an activation function to obtain spatial features corresponding to different scales; the spatial features corresponding to different scales are multiplied with each other, and the multiplication results are added together to fuse the spatial features of different scales to obtain multi-scale spatial interactive fusion features.

[0058] Specifically, this embodiment also describes in detail how to use a graph convolutional network (GCN) to process the characteristic graph structure of voltage data to achieve interactive fusion of multi-scale spatial features in the SOC prediction of extended-range electric vehicles. Through this method, the spatial relationship between voltage features at various scales can be effectively mined and utilized, thereby improving the accuracy and efficiency of SOC prediction.

[0059] The process of multi-scale spatial feature aggregation is described in detail below with reference to the accompanying drawings and specific embodiments. Figure 3 is a schematic diagram of the multi-scale spatial feature aggregation process provided by the embodiment of the present application, such as Figure 3 As shown, the process of aggregating spatial features of different scales in the embodiment of the present application may include the following contents:

[0060] First, the adjacency matrix in each scale of the graph structure is symmetrically normalized to optimize the effect of graph convolution. Normalization is performed by calculating Completed, including is the adjacency matrix A i This step ensures the balance of information propagation during graph convolution.

[0061] Furthermore, the first graph convolutional network is applied to process the normalized adjacency matrix and node feature matrix. In this process, the node features are multiplied by the weight matrix and nonlinearly transformed by the activation function (such as ReLU) to obtain the aggregated spatial features at each scale.

[0062] Furthermore, the spatial features obtained from different scales are interactively fused. This includes interacting these features through product operations. This interactive processing can capture and strengthen the nonlinear relationship between features of different scales. The above product results are added to obtain a comprehensive multi-scale spatial interactive fusion feature. In this way, features of different scales are effectively integrated, which improves the expressiveness of features and the performance of prediction models.

[0063] For example, in a specific example, the three scale graph structures obtained by the above embodiment are input into the graph convolution, and the spatial features between all time periods at each scale are extracted by GCN, and then the three spatial features after the three scales are aggregated are obtained. The spatial relationship information between different time periods at each scale can be effectively utilized to further improve the representation ability of the features. The multi-scale spatial feature aggregation process is as follows: Figure 3 As shown, the calculation process is as follows:

[0064]

[0065] Where i = 1, 2, 3, corresponding to small scale, medium scale and large scale respectively. i Representative A 1 , A 2 and A 3 , Represents the adjacency matrix A i Perform symmetric normalization operation. i represents the weight matrix, X in Represents X 1n , X 2n and X 3n , H i Represents the three aggregated features generated after the three scales are aggregated by graph convolution, and σ is the Relu activation function.

[0066] Then the three spatial features H 1 , H 2 and H 3 After product interaction, the addition fusion is performed. Through the product interaction and addition fusion operations, the nonlinear relationship between them can be captured to highlight important related features and further improve the feature expression ability. The multi-scale spatial interaction fusion H S The calculation process is as follows:

[0067] HS =H 1 ·H 2 +H 1 ·H 3 +H 2 ·H 3

[0068] Through the method of the above-mentioned embodiment, by combining multi-scale features and graph convolution network technology, this embodiment not only enhances the utilization efficiency of data, but also improves the accuracy of prediction. In addition, through the application of graph structure, the model can better understand and utilize the complex spatial relationships in voltage data. Therefore, this embodiment effectively improves the performance of SOC prediction of extended-range electric vehicle batteries through the interactive fusion of advanced graph convolution technology and multi-scale features, and provides strong technical support for battery management and energy optimization.

[0069] In some embodiments, according to the node features of different scales corresponding to each time period, a time graph structure corresponding to each time period is constructed, including:

[0070] Different scales in each time period are taken as nodes, and different scale features in the same time period are taken as node features. An adjacency matrix is ​​established according to the relationship between nodes of different scales. The time graph structure corresponding to each time period is constructed using the nodes, adjacency matrix and node feature set in each time period.

[0071] Specifically, this embodiment also describes in detail a multi-scale time attention enhancement module, which is an advanced technology for predicting the state of charge (SOC) of the battery of an extended-range electric vehicle. This embodiment focuses on mining and utilizing the mutual relationship between different scale features in the same time period by constructing a time graph structure to improve the accuracy and efficiency of SOC prediction.

[0072] The following describes in detail the process of constructing a time graph structure based on node features of different scales corresponding to different time periods in conjunction with the accompanying drawings and specific embodiments. Figure 4 is a schematic diagram of the multi-scale temporal attention generation process provided by the embodiment of the present application, such as Figure 4 As shown, the multi-scale temporal attention generation process of the embodiment of the present application may include the following contents:

[0073] First, in each time period, three different scale features are defined as independent nodes. Specifically, small-scale features are used as the first node, medium-scale features are used as the second node, and large-scale features are used as the third node. This classification reflects the changes in voltage data at different levels of detail. The feature set of each node is composed of voltage signal features of the corresponding scale, which are obtained through the multi-scale convolution processing described previously, and the feature length is unified through adaptive pooling to ensure the consistency and comparability of the features.

[0074] Furthermore, in each time period, an adjacency matrix is ​​established by evaluating the relationship between features of different scales. Specifically, this adjacency matrix is ​​a 3x3 matrix that reflects the mutual influence and connection strength between nodes of three different scales. Each element in the matrix represents the similarity or correlation between the features of two nodes, which can be obtained by calculating the correlation between the features or other statistical methods.

[0075] Furthermore, a graph convolutional network (GCN) is applied to the temporal graph structure constructed for each time period. This step enables the model to effectively extract and integrate the relationship information between features of different scales in the same time period, thereby capturing the dynamic interaction between details and the global perspective.

[0076] Furthermore, the results of graph convolutional network processing are used to perform feature interaction fusion on the spatial features of different scales in each time period. This includes multiplying these features and adding the results to form a comprehensive multi-scale spatial interaction fusion feature. This fusion feature captures the nonlinear relationship between different scales and strengthens the model's understanding of key features.

[0077] For example, in a specific example, Figure 4 As shown in FIG. 1 , this embodiment also designs a multi-scale time attention enhancement module. The operation of the previous embodiment is to treat all time periods of the same scale as a graph, and to mine the spatial relationship between all time periods at the same scale. However, this operation lacks the extraction of the relationship between different scales in the same time period. Therefore, this embodiment regards the three different scales in each time period as three nodes, and the different scale features of the same time period as node features, and mines the relationship between different scales in the same time period. In this way, n time graph structures corresponding to n time periods can be obtained. This embodiment uses G N Indicates that G N =(V 3 ,A,X iN ), where V 3 represents that there are only three nodes, the first node corresponds to the first scale, the second node corresponds to the second scale, and the third node corresponds to the third scale. The adjacency matrix A is represented by a 3×3 matrix, which is the matrix obtained by connecting the three nodes in pairs. iN Represents the feature set corresponding to the three scales in N time periods.

[0078] Through the method of the present embodiment described above, through this multi-scale time graph structure and attention enhancement module, the model can more deeply analyze and understand the complex relationship between different scale features in the same time period, and provide a more accurate and comprehensive SOC prediction. The comprehensive use of multi-scale features and time relationship analysis can significantly improve the accuracy of SOC prediction, thereby optimizing the energy management system of electric vehicles.

[0079] In some embodiments, generating an attention coefficient for each time period according to the time feature includes:

[0080] A multi-layer perceptron is used to process the time features so as to perform adaptive feature extraction on the time features of each time period. The features obtained by the adaptive feature extraction are processed by activation function and normalization function in turn to obtain the attention coefficient corresponding to each time period.

[0081] Specifically, this embodiment also introduces in detail a time attention mechanism designed for the prediction of the state of charge (SOC) of the battery of an extended-range electric vehicle. This mechanism optimizes the utilization of features by analyzing the multi-scale feature relationship in each time period and generating corresponding attention coefficients to improve the accuracy and efficiency of prediction.

[0082] First, a graph convolutional network (GCN) is applied to the temporal graph structure constructed for each time period to extract feature relations between different scales. The feature graph structure of each time period includes small-scale, medium-scale, and large-scale nodes, which are connected by a common adjacency matrix.

[0083] Furthermore, through graph convolution processing, the feature matrix of the node is multiplied by the symmetrically normalized adjacency matrix and weight matrix, and the ReLU function is used as the activation function to obtain the aggregated time features. The time features output by the graph convolution network represent the comprehensive relationship between the features of different scales in each time period, providing a basis for the subsequent generation of attention coefficients.

[0084] Furthermore, a multi-layer perceptron (MLP) is used to process the aggregated temporal features of each time period. MLP further abstracts and refines the features through a series of fully connected layers, so that the model can better capture the key information of temporal features. The processed temporal features are first transformed by the tanh activation function to enhance the model's ability to capture nonlinear relationships in the features. Subsequently, the activated output is normalized using the Softmax function to generate the attention coefficients for each time period. These attention coefficients reflect the relative importance of features in different time periods.

[0085] Finally, the generated attention coefficients are used to weight the multi-scale spatial interaction fusion features to strengthen the influence of those more critical time periods. This step is achieved by performing a dot product operation on the multi-scale fusion features of each time period and its corresponding attention coefficient.

[0086] For example, in a specific example, Figure 4 As shown in the figure, the n time graph structures of the above embodiment are input into the graph convolution, and the relationship between the three scale features in each time period is extracted through GCN, and then the n time features are obtained and H N It means that the relationship information between different scales in each time period can be effectively mined. The calculation process is as follows:

[0087]

[0088] Then based on H N Generate the attention coefficient in each time period, and perform the enhancement operation on the time channel for the multi-scale spatial interaction fusion features obtained above. The specific operations are as follows:

[0089] The feature H generated after the above graph convolution N First, a multi-layer perceptron MLP is used to analyze the feature H N The features of each time period in the adaptive feature extraction are then used to obtain the attention coefficient a corresponding to each time period through the tanh activation function and the Softmax function. N As shown below:

[0090] a N =soft max(tanh(MLP(H N )))

[0091] Where N = 1, 2, ..., n, representing n time periods. N Represents the attention coefficient corresponding to n time periods.

[0092] Through the method of the above embodiment, by introducing the time attention mechanism, the embodiment not only strengthens the understanding of the relationship between features, but also enables the model to automatically adjust its influence according to the importance of the features of each time period when predicting SOC, thereby improving the accuracy and response speed of the prediction. The introduction of the attention coefficient allows the model to focus more on those time periods that have the greatest impact on the prediction results when processing dynamically changing voltage data, thereby effectively improving the accuracy and efficiency of the overall prediction.

[0093] In some embodiments, the multi-scale spatial interactive fusion features are weighted enhanced using the attention coefficient to obtain enhanced features, and the battery state of charge of the vehicle is predicted in real time based on the enhanced features, including:

[0094] The dot product operation is performed on the attention coefficient and the multi-scale spatial interaction fusion feature to perform weighted enhancement on the multi-scale spatial interaction fusion feature to obtain the enhanced feature; the enhanced feature is input into the fully connected layer, and the fully connected layer is used to predict and output the battery state of charge of the vehicle in real time.

[0095] Specifically, this embodiment describes in detail how to use the attention coefficient to weighted enhance the multi-scale spatial interaction fusion features, and perform real-time prediction of the battery state of charge (SOC) of the extended-range electric vehicle based on these enhanced features. This method utilizes the combination of deep learning and attention mechanism to improve prediction accuracy and optimize vehicle energy management.

[0096] First, the multi-scale features of each time period are extracted from the graph convolutional network, and the attention coefficients corresponding to each time period are generated through the aforementioned multi-layer perceptron (MLP) and activation function processing. These coefficients reflect the relative importance of features in different time periods.

[0097] Furthermore, the obtained attention coefficient is subjected to a dot product operation with the corresponding multi-scale spatial interaction fusion feature. This operation directly applies the attention coefficient to the corresponding feature, weights and adjusts the contribution of the feature, thereby highlighting the most critical feature for SOC prediction. The result of the dot product operation is H out =a N ⊙H S , where ⊙ represents the product between elements, ensuring that each eigenvector element is appropriately adjusted according to its importance.

[0098] Furthermore, the enhanced features are fed into a fully connected layer, which is a typical neural network structure, for the final SOC prediction. The fully connected layer can further learn the complex relationship between features and output the final SOC value. The design of the fully connected layer usually includes several neural network layers, which learn the high-level representation of the input features through nonlinear transformations and finally give the predicted SOC value through an output layer.

[0099] For example, in a specific example, Figure 4 As shown, the multi-scale spatial interactive fusion features obtained in the above embodiment are enhanced on the time channel. Since the multi-scale spatial interactive fusion features do not change the order of time periods during the acquisition process, the multi-scale spatial interactive fusion features and a N In the time dimension, the attention coefficient a is consistent. N Interactive fusion of features H with multi-scale space S Perform dot product operation to achieve weighted enhancement of multi-scale spatial interactive fusion features, and obtain the enhanced features as the final output feature H out , the calculation process is expressed as follows:

[0100] H out =a N ⊙H S

[0101] The enhanced feature H out Finally, the SOC prediction task is performed through a fully connected layer to improve the accuracy of SOC prediction and lay the foundation for optimizing the energy flow and distribution strategy of the extended-range vehicle energy management strategy.

[0102] Through the method of the present embodiment described above, by weighting and enhancing features, the model can focus more on those features that have a decisive influence on the prediction results. This method effectively improves the accuracy of the prediction, especially in a dynamically changing driving environment. Accurate SOC prediction is the key to optimizing the energy management strategy of extended-range electric vehicles. By accurately predicting the battery's state of charge, the energy management system can more effectively allocate energy flow between the battery and the engine, avoiding unnecessary energy waste.

[0103] According to the technical solution provided by the above embodiment of the present application, the present technical solution has at least the following advantages:

[0104] This technical solution uses CNN to extract multi-scale features of voltage data, including details and global features. These features are further processed by GCN to explore the spatial relationship between features of different scales. This method can integrate independent information of each scale and deeply analyze the mutual influence between different time periods through graph convolution, thereby effectively improving the representation ability of features.

[0105] Using graph convolutional networks to mine spatial relationships among multi-scale features enables the model to identify and exploit complex dependencies between these features. This approach improves the accuracy of SOC prediction while also enhancing the model’s adaptability to battery behavior under different operating conditions.

[0106] In the time dimension, this technical solution introduces attention coefficients to emphasize important time period features. These coefficients are obtained by examining the interaction of features of different scales in the same time period. In this way, the model not only pays attention to the features within each scale, but also evaluates their importance in the time series, thereby giving priority to the features that have the greatest impact on the prediction results.

[0107] The attention coefficient is applied to the multi-scale spatial interaction fusion features, and the dot product operation is performed to weighted enhance these features. This processing step ensures that the model can more accurately identify and utilize data at key moments when making SOC predictions, improving the accuracy and response speed of predictions.

[0108] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0109] Figure 5 Schematic diagram of the structure of a battery state of charge prediction device based on multi-scale interactive fusion provided in an embodiment of the present application. Figure 5 As shown, the battery state of charge prediction device based on multi-scale interactive fusion includes:

[0110] The acquisition module 501 is configured to acquire the voltage data of the vehicle battery and process the voltage data in segments according to a preset time window;

[0111] The convolution module 502 is configured to perform a convolution operation on the voltage data of each time period using a plurality of convolution kernels of different scales to obtain a plurality of feature sets of different scales;

[0112] The fusion module 503 is configured to construct a feature graph structure corresponding to each scale according to the feature set, process the feature graph structure using the first graph convolutional network to obtain spatial features between nodes at different scales, and perform feature interactive fusion on the spatial features of different scales to obtain multi-scale spatial interactive fusion features;

[0113] A generation module 504 is configured to construct a time graph structure corresponding to each time period according to node features of different scales corresponding to each time period, process the time graph structure using a second graph convolutional network to obtain a time feature corresponding to each time period, and generate an attention coefficient for each time period according to the time feature, wherein the node feature is a feature obtained after convolution of the voltage data;

[0114] The prediction module 505 is configured to perform weighted enhancement on the multi-scale spatial interaction fusion features using the attention coefficient to obtain enhanced features, and perform real-time prediction of the battery state of charge of the vehicle based on the enhanced features.

[0115] In some embodiments, Figure 5 The convolution module 502 uses the first scale convolution kernel, the second scale convolution kernel and the third scale convolution kernel to perform convolution operations on the voltage data of each time period respectively, obtains the voltage signal features corresponding to each time period at different scales, uses the voltage signal features to generate the first scale feature set, the second scale feature set and the third scale feature set, performs adaptive pooling processing on the feature sets of different scales, and obtains the feature set after adaptive pooling processing.

[0116] In some embodiments, Figure 5The fusion module 503 uses the voltage signal features corresponding to each time period in the feature set after adaptive pooling processing as nodes, uses the Euclidean distance matrix between nodes as an adjacency matrix, and uses the nodes, adjacency matrices and node feature sets corresponding to feature sets of different scales to construct a feature graph structure corresponding to each scale.

[0117] In some embodiments, Figure 5 The fusion module 503 performs symmetric normalization on the adjacency matrix in the feature graph structure of each scale, and multiplies the adjacency matrix after symmetric normalization with the feature matrix and weight matrix of the node to obtain aggregated features, and processes the aggregated features using an activation function to obtain spatial features corresponding to different scales; the spatial features corresponding to different scales are multiplied with each other, and the multiplication results are added to fuse the spatial features of different scales to obtain multi-scale spatial interactive fusion features.

[0118] In some embodiments, Figure 5 The generation module 504 takes different scales in each time period as nodes, takes different scale features in the same time period as node features, establishes an adjacency matrix according to the relationship between nodes of different scales, and constructs a time graph structure corresponding to each time period using the nodes, adjacency matrix and node feature set in each time period.

[0119] In some embodiments, Figure 5 The generation module 504 uses a multi-layer perceptron to process the time features so as to perform adaptive feature extraction on the time features of each time period, and the features obtained by the adaptive feature extraction are processed by activation functions and normalization functions in turn to obtain the attention coefficient corresponding to each time period.

[0120] In some embodiments, Figure 5 The prediction module 505 performs a dot product operation on the attention coefficient and the multi-scale spatial interaction fusion feature to perform weighted enhancement on the multi-scale spatial interaction fusion feature to obtain an enhanced feature; the enhanced feature is input into the fully connected layer, and the fully connected layer is used to predict and output the battery state of charge of the vehicle in real time.

[0121] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0122] Figure 6 Schematic diagram of the structure of the electronic device 6 provided in the embodiment of the present application. Figure 6As shown, the electronic device 6 of this embodiment includes: a processor 601, a memory 602, and a computer program 603 stored in the memory 602 and executable on the processor 601. When the processor 601 executes the computer program 603, the steps in the above-mentioned various method embodiments are implemented. Alternatively, when the processor 601 executes the computer program 603, the functions of each module / unit in the above-mentioned various device embodiments are implemented.

[0123] Exemplarily, the computer program 603 may be divided into one or more modules / units, which are stored in the memory 602 and executed by the processor 601 to complete the present application. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 603 in the electronic device 6.

[0124] The electronic device 6 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 6 may include, but is not limited to, a processor 601 and a memory 602. Those skilled in the art will appreciate that Figure 6 It is only an example of the electronic device 6 and does not constitute a limitation of the electronic device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0125] The processor 601 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0126] The memory 602 may be an internal storage unit of the electronic device 6, for example, a hard disk or memory of the electronic device 6. The memory 602 may also be an external storage device of the electronic device 6, for example, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 6. Further, the memory 602 may also include both an internal storage unit of the electronic device 6 and an external storage device. The memory 602 is used to store computer programs and other programs and data required by the electronic device. The memory 602 may also be used to temporarily store data that has been output or is to be output.

[0127] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0128] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0129] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0130] In the embodiments provided in the present application, it should be understood that the disclosed devices / computer equipment and methods can be implemented in other ways. For example, the device / computer equipment embodiments described above are only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. Multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0131] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0132] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0133] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. The computer program may include computer program code, which may be in source code form, object code form, executable file or some intermediate form. Computer-readable media may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0134] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A battery state of charge prediction method based on multi-scale interactive fusion, characterized in that: include: Acquire voltage data of a vehicle battery, and process the voltage data in segments according to a preset time window; The voltage data of each time period is convolved using convolution kernels of multiple different scales to obtain feature sets of multiple different scales. According to the feature set, a feature graph structure corresponding to each scale is constructed, the feature graph structure is processed by a first graph convolutional network to obtain spatial features between nodes at different scales, and spatial features of different scales are interactively fused to obtain multi-scale spatial interactive fusion features; According to the node features of different scales corresponding to each time period, a time graph structure corresponding to each time period is constructed, the time graph structure is processed by a second graph convolutional network to obtain the time features corresponding to each time period, and an attention coefficient of each time period is generated according to the time features, wherein the node features are features obtained after convolution of the voltage data; The multi-scale spatial interactive fusion feature is weightedly enhanced by using the attention coefficient to obtain an enhanced feature, and the battery state of charge of the vehicle is predicted in real time based on the enhanced feature.

2. The method according to claim 1, characterized in that The convolution operation is performed on the voltage data of each time period using convolution kernels of multiple different scales to obtain feature sets of multiple different scales, including: The first scale convolution kernel, the second scale convolution kernel and the third scale convolution kernel are used to perform convolution operations on the voltage data of each time period respectively, so as to obtain the voltage signal features corresponding to each time period at different scales, and the first scale feature set, the second scale feature set and the third scale feature set are generated by using the voltage signal features, and the feature sets of different scales are adaptively pooled to obtain the feature sets after adaptive pooling processing.

3. The method according to claim 2, characterized in that The step of constructing a feature graph structure corresponding to each scale according to the feature set includes: The voltage signal features corresponding to each time period in the feature set after the adaptive pooling process are taken as nodes, the Euclidean distance matrix between the nodes is taken as an adjacency matrix, and the nodes, adjacency matrices and node feature sets corresponding to feature sets of different scales are used to construct a feature graph structure corresponding to each scale.

4. The method according to claim 1, characterized in that: The first graph convolutional network is used to process the feature graph structure to obtain spatial features between nodes at different scales, and the spatial features at different scales are interactively fused to obtain multi-scale spatial interactive fusion features, including: The adjacency matrix in the feature graph structure of each scale is symmetrically normalized, and the adjacency matrix after symmetrical normalization is multiplied with the feature matrix and weight matrix of the node to obtain aggregated features, and the aggregated features are processed using an activation function to obtain spatial features corresponding to different scales; the spatial features corresponding to different scales are multiplied with each other, and the product results are added to fuse the spatial features of different scales to obtain the multi-scale spatial interactive fusion features.

5. The method according to claim 1, characterized in that The step of constructing a time graph structure corresponding to each time period according to node features of different scales corresponding to each time period includes: Different scales in each time period are taken as nodes, and different scale features in the same time period are taken as node features. An adjacency matrix is ​​established according to the relationship between nodes of different scales. The time graph structure corresponding to each time period is constructed using the nodes, adjacency matrix and node feature set in each time period.

6. The method according to claim 1, characterized in that Generating the attention coefficient of each time period according to the time feature includes: The time features are processed by a multi-layer perceptron so as to perform adaptive feature extraction on the time features of each time period. The features obtained by the adaptive feature extraction are processed by activation functions and normalization functions in turn to obtain the attention coefficient corresponding to each time period.

7. The method according to claim 1, characterized in that The method of using the attention coefficient to weighted enhance the multi-scale spatial interactive fusion feature to obtain an enhanced feature, and performing real-time prediction of the battery state of charge of the vehicle based on the enhanced feature, includes: A dot product operation is performed on the attention coefficient and the multi-scale spatial interaction fusion feature to perform weighted enhancement on the multi-scale spatial interaction fusion feature to obtain the enhanced feature; the enhanced feature is input into a fully connected layer, and the fully connected layer is used to predict and output the battery state of charge of the vehicle in real time.

8. A battery state of charge prediction device based on multi-scale interactive fusion, characterized in that: include: An acquisition module is configured to acquire voltage data of a vehicle battery and process the voltage data in segments according to a preset time window; A convolution module is configured to perform a convolution operation on the voltage data of each time period using convolution kernels of multiple different scales to obtain feature sets of multiple different scales; A fusion module is configured to construct a feature graph structure corresponding to each scale according to the feature set, process the feature graph structure using a first graph convolutional network to obtain spatial features between nodes at different scales, and interactively fuse the spatial features of different scales to obtain multi-scale spatial interactive fusion features; A generation module is configured to construct a time graph structure corresponding to each time period according to node features of different scales corresponding to each time period, process the time graph structure using a second graph convolutional network to obtain a time feature corresponding to each time period, and generate an attention coefficient for each time period according to the time feature, wherein the node feature is a feature obtained by convolution of the voltage data; The prediction module is configured to use the attention coefficient to perform weighted enhancement on the multi-scale spatial interaction fusion feature to obtain an enhanced feature, and perform real-time prediction of the battery state of charge of the vehicle based on the enhanced feature.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images