Energy management optimization method, device and storage medium for extended-range vehicles

Abstract

The present application provides an energy management optimization method, device and storage medium for an extended-range vehicle. The method includes: extracting features from a target parameter signal to obtain feature information of different scales, and constructing a graph structure for features of each scale; processing the graph structure using a shared parameter graph convolution module to extract fused features, and aggregating the fused features to obtain aggregated features; extracting features from the graph structure using a non-shared parameter graph convolution module, and performing a pooling operation on the node features, and weighting them to obtain weighted features corresponding to different scales; fusing the aggregated features with the weighted features corresponding to different scales, and inputting the fused features into a prediction model for prediction to obtain the battery state of charge in the prediction time domain; generating energy management instructions based on the battery state of charge, and using the energy management instructions to optimize the energy management of the extended-range vehicle. The present application improves the prediction accuracy and optimization effect of the energy management optimization algorithm.

Description

Technical Field

[0001] The present application relates to the field of new energy vehicle technology, and in particular to an energy management optimization method, device and storage medium for an extended-range vehicle. Background Art

[0002] With increasingly stringent environmental protection regulations, the new energy vehicle industry has achieved rapid development. Among the many new energy vehicles, extended-range electric vehicles have become an important branch of this field with their advantages of reducing environmental pollution and alleviating electric vehicle range anxiety. The energy management strategy of extended-range electric vehicles, especially the power distribution between power sources and the vehicle's operating performance under variable conditions, plays a decisive role in the vehicle's economy and environmental protection. In this context, accurate estimation of the battery's state of charge (SOC) is crucial to optimizing energy distribution and flow, which can ensure that the battery provides power at the best time and reduce unnecessary charging and discharging activities.

[0003] Currently, the estimation of SOC mainly relies on model-based methods and data-driven methods. Model-based methods limit their scope of application due to their high requirements for parameter accuracy, while data-driven methods, although designed quickly, often require a compromise between accuracy and real-time performance. Although deep learning technology has been introduced to improve prediction accuracy, more accurate SOC predictions often mean more complex models and computational burdens, resulting in longer response times. In addition, although the introduction of graph convolutional neural networks (GCNs) has improved processing efficiency, their application under complex working conditions of real vehicles is still challenging. Existing technologies have failed to fully address the problem of balancing the optimization effect and real-time performance of algorithms in energy management strategies. Summary of the invention

[0004] In view of this, the embodiments of the present application provide an energy management optimization method, device and storage medium for an extended-range vehicle to solve the problems of low prediction accuracy and poor optimization effect of the energy management optimization algorithm in the prior art.

[0005] In a first aspect of an embodiment of the present application, a method for optimizing energy management of an extended-range vehicle is provided, comprising: extracting features from a target parameter signal of the vehicle to obtain feature information of different scales, constructing a graph structure for features of each scale based on the feature information of different scales, wherein the graph structure includes feature information of the corresponding scale and an adjacency matrix; processing the graph structure corresponding to each scale using a preset shared parameter graph convolution module to extract fused features corresponding to each scale, aggregating the fused features to obtain aggregated features representing shared information; extracting features from the graph structures of corresponding scales using preset non-shared parameter graph convolution modules corresponding to different scales, and performing a pooling operation on node features in the graph structure during the feature extraction process using a pooling attention guidance module, and weighting them to obtain weighted features corresponding to different scales; fusing the aggregated features with the weighted features corresponding to different scales, and inputting the fused features into a prediction model for prediction to obtain a battery state of charge in a prediction time domain; generating energy management instructions based on the battery state of charge, and optimizing energy management of the extended-range vehicle using the energy management instructions.

[0006] According to a second aspect of an embodiment of the present application, an energy management optimization device for an extended-range vehicle is provided, comprising: a construction module, configured to perform feature extraction on a target parameter signal of a vehicle to obtain feature information of different scales, and construct a graph structure for features of each scale according to the feature information of different scales, wherein the graph structure includes feature information of the corresponding scale and an adjacency matrix; a processing module, configured to process the graph structure corresponding to each scale respectively using a preset shared parameter graph convolution module to extract fused features corresponding to each scale, and aggregate the fused features to obtain aggregate features representing shared information; an extraction module, configured to perform feature extraction on the graph structure of the corresponding scale respectively using a preset non-shared parameter graph convolution module corresponding to different scales, and perform a pooling operation on the node features in the graph structure during the feature extraction process using a pooled attention guidance module, and weight them to obtain weighted features corresponding to different scales; a prediction module, configured to fuse the aggregated features with the weighted features corresponding to different scales, and input the fused features into a prediction model for prediction to obtain a battery state of charge in a prediction time domain; an optimization module, configured to generate energy management instructions according to the battery state of charge, and use the energy management instructions to optimize the energy management of the extended-range vehicle.

[0007] According to a third aspect of an embodiment of the present application, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above method are implemented.

[0008] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects:

[0009] By extracting features from the target parameter signals of the vehicle, feature information of different scales is obtained. Based on the feature information of different scales, a graph structure is constructed for the features of each scale, and the graph structure includes the feature information of the corresponding scale and the adjacency matrix; the graph structure corresponding to each scale is processed by a preset shared parameter graph convolution module to extract the fusion features corresponding to each scale, and the fusion features are aggregated to obtain the aggregated features that characterize the shared information; the graph structures of the corresponding scales are extracted by using the preset non-shared parameter graph convolution modules corresponding to different scales, and the node features in the graph structure are pooled during the feature extraction process using the pooled attention guidance module, and weighted to obtain the weighted features corresponding to different scales; the aggregated features are fused with the weighted features corresponding to different scales, and the fused features are input into the prediction model for prediction to obtain the battery state of charge in the prediction time domain; energy management instructions are generated according to the battery state of charge, and the energy management instructions are used to optimize the energy management of the extended-range vehicle. This application improves the prediction accuracy and optimization effect of the energy management optimization algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] 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.

[0011] Figure 1 It is a schematic diagram of the overall implementation process of the real-time optimization method for energy management of an extended-range vehicle based on SOC prediction provided in an embodiment of the present application;

[0012] Figure 2 It is a flow chart of the energy management optimization method for the extended-range vehicle provided in the embodiment of the present application;

[0013] Figure 3 is a schematic diagram of a multi-scale graph structure construction process provided in an embodiment of the present application;

[0014] Figure 4 It is a schematic diagram of the structure of a shared parameter graph convolution module provided in an embodiment of the present application;

[0015] Figure 5 It is a schematic diagram of the structure of a non-shared parameter graph convolution module provided in an embodiment of the present application;

[0016] Figure 6 is a schematic diagram of the structure of the pooled attention guidance module provided in an embodiment of the present application;

[0017] Figure 7It is a schematic diagram of calculating the minimum fuel consumption path using the energy management optimization algorithm provided in an embodiment of the present application;

[0018] Figure 8 It is a structural schematic diagram of an energy management optimization device for a range-extended vehicle provided in an embodiment of the present application;

[0019] Fig. 9 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] 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.

[0021] In order to cope with strict emission regulations, new energy vehicles have developed rapidly. Among them, extended-range electric vehicles not only reduce pollution and emissions, but also alleviate the range anxiety of electric vehicles, becoming the backbone of new energy vehicles. The energy management strategy not only determines the power distribution between the power sources of extended-range electric vehicles, but also affects the various operating performances of the vehicle under different working conditions. Therefore, it is crucial to design a reasonable energy management strategy. By accurately estimating the state of charge (SOC) of the battery, the energy management system can optimize the energy flow and distribution strategy based on the difference between the current SOC and the predicted SOC. This ensures that the battery provides enough energy when needed and avoids charging or discharging when it is not necessary. For the prediction of SOC, parameters such as current, voltage and temperature are often used for modeling and prediction.

[0022] Existing SOC estimation methods mainly include model-based methods and data-driven methods. Model-based methods refer to the direct use of models to estimate SOC. Direct estimation of related models has high requirements on the composition parameters. Compared with the above methods, data-driven methods have shorter design time. Although deep learning-based methods have made great progress in SOC estimation, SOC prediction requires a trade-off between high accuracy and real-time performance. More accurate SOC predictions usually require more complex models and algorithms, which may result in higher computing costs and delays.

[0023] Graph Convolutional Neural Network (GCN) is a deep learning model based on graph structured data. It can perform parallel calculations on different nodes of the graph data, which can significantly improve the computational efficiency of the model and thus reduce the delay of prediction. This is very important for tasks that require rapid decision-making and prediction in real-time scenarios. GCN can use parameters such as current, voltage, and temperature as node features, and capture the mutual influence and dependency between them by learning the relationship between nodes. This can more comprehensively model the various influencing factors in the battery system and improve the accuracy of SOC prediction.

[0024] However, the existing GCN method is difficult to handle such changing working conditions under the complex working conditions of the actual driving process of the vehicle. In addition, it is also an important challenge for the optimization algorithm in the energy management strategy to balance the optimization effect and real-time performance. Therefore, it is an urgent problem to accurately predict SOC and combine the optimization algorithm to design a real-time energy management strategy for extended-range vehicles with the goal of minimizing fuel consumption.

[0025] In view of the problems existing in the prior art, the present application provides a real-time optimization method for energy management of extended-range electric vehicles based on SOC prediction, aiming to solve the problems of low accuracy of state of charge (SOC) prediction of existing extended-range electric vehicles under complex working conditions and difficulty in real-time optimization of energy management optimization algorithms. Before describing the embodiments of the present application in detail, the overall implementation process of the technical solution of the present application is first summarized in conjunction with the accompanying drawings and specific embodiments. Figure 1 1 is a schematic diagram of the overall implementation process of the real-time optimization method for energy management of an extended-range vehicle based on SOC prediction provided in the embodiment of the present application. Figure 1 As shown, the real-time optimization method for energy management of an extended-range vehicle based on SOC prediction may specifically include the following contents:

[0026] This application proposes a real-time optimization method for energy management of extended-range vehicles based on SOC prediction to solve the problems that the traditional method is affected by complex working conditions during the actual driving of extended-range vehicles, resulting in low SOC prediction accuracy, and the energy management optimization algorithm is difficult to balance the optimization effect and real-time performance. The overall flow chart is as follows: Figure 1 As shown in the figure, the main methods include: firstly, a multi-scale fusion graph convolution module is designed, and feature information of different scales is obtained for voltage, current and temperature through multi-scale one-dimensional convolution, and graph structures are constructed for features of different scales.

[0027] Then, we designed a shared parameter graph convolution and non-shared parameter graph convolution for different scales to better capture the multi-scale features of the data and adapt to the changing conditions of different scales. In addition, we designed a pooling attention guidance module in the non-shared parameter graph convolution module for the features of different scales to guide the graph convolution network to pay more attention to important nodes during the feature extraction process.

[0028] After completing the above operations, the shared parameter and non-shared parameter graph convolution module features are added and fused, and the fused features are input into the long short-term memory network (LSTM) to accurately predict the SOC in the future time domain. Finally, based on the Dijkstra algorithm, the minimum fuel consumption path in the predicted time domain is obtained according to the SOC obtained from the above prediction, and finally the real-time optimization of the energy management of the extended-range vehicle is realized to improve the fuel economy.

[0029] The contents of the technical solution of the present application are described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] Figure 2 It is a flow chart of the energy management optimization method for extended-range vehicles provided in an embodiment of the present application. Figure 2 The energy management optimization method of the extended-range vehicle can be executed by the on-board energy management system. Figure 2 As shown, the energy management optimization method of the extended-range vehicle may specifically include:

[0031] S201, extracting features from the target parameter signal of the vehicle to obtain feature information of different scales, and constructing a graph structure for the features of each scale according to the feature information of different scales, wherein the graph structure includes the feature information of the corresponding scale and an adjacency matrix;

[0032] S202, using a preset shared parameter graph convolution module to process the graph structure corresponding to each scale respectively, so as to extract the fusion features corresponding to each scale, aggregate the fusion features, and obtain the aggregate features representing the shared information;

[0033] S203, using the preset non-shared parameter graph convolution modules corresponding to different scales, respectively extracting features from the graph structures of the corresponding scales, and using the pooling attention guidance module in the feature extraction process to perform a pooling operation on the node features in the graph structure, and weighting them to obtain weighted features corresponding to different scales;

[0034] S204, fusing the aggregated features with the weighted features corresponding to different scales output by the non-shared parameter graph convolution module, and inputting the fused features into the prediction model for prediction to obtain the battery state of charge in the prediction time domain;

[0035] S205, generating an energy management instruction according to the battery state of charge, and optimizing energy management of the extended-range vehicle using the energy management instruction.

[0036] In some embodiments, feature extraction is performed on the target parameter signal of the vehicle to obtain feature information of different scales, including:

[0037] A multi-scale one-dimensional convolution network is used to perform a multi-scale one-dimensional convolution operation on the target parameter signal to obtain first-scale feature information and second-scale feature information; wherein the multi-scale one-dimensional convolution network includes one-dimensional convolution layers with different convolution kernel sizes and depths, and the target parameter signal includes a voltage signal, a current signal and a temperature signal.

[0038] Specifically, this embodiment provides a multi-scale feature extraction method, which uses a multi-scale one-dimensional convolutional network to process the voltage, current and temperature signals of the extended-range electric vehicle, thereby obtaining feature information at different scales to optimize the energy management strategy.

[0039] In this embodiment, the target parameter signals are first set, including voltage, current and temperature signals. These signals are obtained from the battery management system of the extended-range electric vehicle, and they reflect the performance status of the battery under real-time working conditions.

[0040] Next, a multi-scale one-dimensional convolutional network is designed, which includes two convolutional layers of different scales. The first convolutional layer (1dCNN_1) uses a smaller convolution kernel (10×3) to extract high-resolution detail features, namely small-scale features X1 (first-scale features). These detail features can reflect parameter changes in the short term and are essential for capturing rapid fluctuations and dynamic changes in a small range. In the following embodiments, the first scale is also referred to as the small scale, the first-scale feature is the small-scale feature, and the first-scale graph structure is the small-scale graph structure.

[0041] The second convolution layer (1dCNN_2) uses a larger convolution kernel (30×3) to capture a wider range of features, namely, large-scale features X2 (second-scale features). These large-scale features help understand the overall trend and pattern of parameter signals over a larger time range, which is very useful for long-term energy management strategy formulation. In the following embodiments, the second scale is also referred to as large scale, the second scale feature is the large-scale feature, and the second scale graph structure is the large-scale graph structure.

[0042] Each convolution layer is adapted to its specific scale through different configurations of its convolution kernel and depth to ensure that relevant information can be retained to the maximum extent during feature extraction. After proper training, this multi-scale one-dimensional convolutional network can automatically extract useful feature information from the input voltage, current and temperature signals and output them in the form of X1 and X2.

[0043] According to the method of the present embodiment, the advantage of the method is that it can simultaneously process and integrate feature information of different scales to provide a more comprehensive representation of the battery performance status. Thus, the energy management system can use this information to make more accurate decisions on how to allocate power and when to charge or discharge, so as to maximize fuel efficiency and minimize emissions.

[0044] In some embodiments, based on feature information at different scales, a graph structure is constructed for features at each scale, including:

[0045] The first scale feature information is used as a node to construct a first scale graph structure corresponding to the first scale feature, and the second scale feature information is used as a node to construct a second scale graph structure corresponding to the second scale feature; the first scale feature information and the second scale feature information are divided according to the number of nodes, and the distances between the nodes in the graph structures of different scales are calculated respectively to form adjacency matrices corresponding to different scales.

[0046] Specifically, this embodiment also introduces in detail a multi-scale graph structure construction method for energy management of extended-range electric vehicles. The method mainly includes constructing a small-scale (i.e., first-scale) and a large-scale (i.e., second-scale) graph structure, and generating a corresponding adjacency matrix, so as to more effectively predict the battery state of charge (SOC) and manage energy. Figure 3 is a schematic diagram of the multi-scale graph structure construction process provided by the embodiment of the present application, such as Figure 3 As shown, the multi-scale graph structure construction process of the embodiment of the present application may include the following contents:

[0047] This embodiment constructs graph structures for features of different scales respectively, and the graph structure is represented by G=(V, E, A), where V represents the set of nodes in the graph, E is the set of edges, and A represents the adjacency matrix of the graph G. The number of nodes in the feature graph structures of different scales is the same, all N, and the features of different scales are automatically divided into N according to the number of nodes. For each node under the small-scale feature, more detailed features will be included, and for each node under the large-scale feature, the number of features included is small, but it is rich in more global features, which can better capture the overall or global feature information.

[0048] Through such automatic division, nodes at different scales can fully utilize the characteristics of their scales and maintain the important constraint of consistent number of nodes, thereby enhancing the representation ability and adaptability of the model. Then, by calculating the Euclidean distance of the feature vectors of each node at the two scales, the small-scale Euclidean distance matrix E1 and the large-scale Euclidean distance matrix E2 can be obtained. Obviously, E1 and E2 are symmetric matrices, and their elements E ij represents the Euclidean distance between nodes i and j.

[0049]

[0050] Among them, E1 and E2 are used as the small-scale and large-scale adjacency matrices A1 and A2 respectively.

[0051] For example, in a specific example, constructing small-scale and large-scale graph structures and generating corresponding adjacency matrices may include the following steps:

[0052] 1) Acquisition of feature information: First, the voltage, current and temperature signals are collected from the vehicle’s battery management system, and the small-scale features X1 and large-scale features X2 are extracted through the multi-scale one-dimensional convolutional network described previously.

[0053] 2) Construction of graph structure: Use small-scale features X1 to construct a small-scale graph structure, in which each node represents a set of high-resolution detail features; use large-scale features X2 to construct a large-scale graph structure, in which each node represents a set of features covering a wider range.

[0054] 3) Node partitioning: Determine the number of nodes N in the graph structure, and divide the small-scale and large-scale features into N parts respectively, ensuring that each graph structure has the same number of nodes. This step ensures that features of each scale can be represented equally in the model.

[0055] 4) Generation of adjacency matrix: Calculate the Euclidean distance between nodes in the small-scale graph structure to generate the small-scale Euclidean distance matrix E1; calculate the Euclidean distance between nodes in the large-scale graph structure to generate the large-scale Euclidean distance matrix E2; ensure that E1 and E2 are symmetric matrices, whose elements Eij represent the Euclidean distance between node i and node j.

[0056] 5) Optimization of graph structure: Based on the calculated Euclidean distance matrix, the graph structure is optimized so that the graph convolutional network can process node features more effectively and improve the accuracy and adaptability of model learning.

[0057] Through the above method of this embodiment, the energy management system can more accurately understand and process different aspects of battery characteristics, from details to global perspectives. This multi-scale graph structure construction method helps capture the complex behavior of batteries under different operating conditions and provides strong support for formulating efficient energy management strategies.

[0058] In some embodiments, the graph structures corresponding to each scale are processed respectively using a preset shared parameter graph convolution module to extract fusion features corresponding to each scale, including:

[0059] Initialize the feature representation of the first scale graph structure and the second scale graph structure, and input the node features into the shared parameter graph convolution module;

[0060] Performing symmetric normalization processing on the adjacency matrix of the first scale graph structure and the second scale graph structure to obtain a normalized adjacency matrix;

[0061] Perform graph convolution operations on the first-scale graph structure and the second-scale graph structure, and use the same weight matrix to update the node features of each layer to extract fusion features of each scale;

[0062] The output of each layer of graph convolution is activated using a nonlinear activation function, and the multi-layer graph convolution operation is repeated until the fusion features that meet the depth requirements are extracted.

[0063] Specifically, the embodiment of the present application designs a shared parameter graph convolution module and a non-shared parameter graph convolution module designed for different scales to better capture the multi-scale characteristics of the data and adapt to changing working conditions at different scales.

[0064] This embodiment also introduces in detail a method for processing a multi-scale graph structure using a shared parameter graph convolution module, aiming to extract the fusion features of voltage, current and temperature signals in the battery management system of an extended-range electric vehicle. Figure 4 is a schematic diagram of the structure of the shared parameter graph convolution module provided in the embodiment of the present application, such as Figure 4 As shown, the process of processing the multi-scale graph structure using the shared parameter graph convolution module in the embodiment of the present application may include the following contents:

[0065] First, the original voltage, current and temperature signals collected in the vehicle's battery management system are first processed by a multi-scale one-dimensional convolutional network to generate small-scale features X1 and large-scale features X2. Then, these features are assigned to the nodes of the small-scale graph structure and the large-scale graph structure respectively.

[0066] Furthermore, according to the obtained small-scale and large-scale features, the corresponding graph structure is constructed, and an adjacency matrix is ​​generated for each graph structure. Each node in the graph represents a set of features, and the edge represents the connection relationship between nodes.

[0067] Furthermore, symmetric normalization is performed on the adjacency matrices A1 and A2 of the small-scale and large-scale graph structures to maintain numerical stability in the graph convolution operation.

[0068] Furthermore, a shared parameter graph convolution module is used to process the graph structure of each scale. This module uses a weight matrix to update the node features of each layer to extract deep fusion features. The shared weight matrix here enables graph structures of different scales to learn common information across scales while maintaining their own characteristics.

[0069] Furthermore, the nonlinear characteristics of the output of the graph convolution layer are enhanced by the ReLU activation function. This process is iterated multiple times until a deep fusion feature that can represent small-scale and large-scale information is obtained.

[0070] Further, the above graph convolution and activation process is repeated until the required depth is reached. Subsequently, the small-scale fusion features and large-scale fusion features are comprehensively considered to construct a feature set that comprehensively represents the battery status.

[0071] For example, in a specific example, Figure 4 As shown, in the shared parameter graph convolution module, the fused features are extracted from the small-scale graph (A1, X1) The calculation process is as follows:

[0072]

[0073] Among them, W c (l) is the weight matrix of the lth layer in the shared parameter graph convolution module, is the output of layer l, is the l-1th layer graph convolution fusion feature, To perform symmetric normalization on the adjacency matrix A1, ReLU is used as the activation function.

[0074] Then, extract the fused features from the large-scale image (A2, X2) In order to extract shared information, the same weight matrix W is used in each layer of graph convolution. c (l) , the calculation process is as follows:

[0075]

[0076] in, is the output of layer l, is the output of the l-1th layer, To perform symmetric normalization operation on the adjacency matrix A2.

[0077] Then, the two scale output features obtained from the shared parameter graph convolution are added and aggregated and the average value is taken. The calculation process is as follows:

[0078]

[0079] Through the method of the present embodiment described above, the multi-scale feature extraction and fusion method described in this embodiment is conducive to understanding and predicting the battery behavior of the extended-range electric vehicle under complex working conditions, thereby providing more accurate energy management decision support. Through this method, the energy management system can effectively adjust power distribution, optimize the charging and discharging cycle, improve fuel economy, and reduce environmental impact.

[0080] In some embodiments, using preset non-shared parameter graph convolution modules corresponding to different scales, feature extraction is performed on graph structures of corresponding scales respectively, including:

[0081] Initialize node feature representations of the first scale graph structure and the second scale graph structure;

[0082] Performing symmetric normalization operations on the first scale graph structure and the second scale graph structure respectively to generate a normalized adjacency matrix;

[0083] Applying the corresponding non-shared parameter graph convolution module to the first scale graph structure, and performing graph convolution operation using the weight matrix of the first scale, so as to update the node feature representation of each layer;

[0084] The corresponding non-shared parameter graph convolution module is applied to the second-scale graph structure, and the graph convolution operation is performed using the second-scale weight matrix to update the node feature representation of each layer.

[0085] Specifically, this embodiment also describes in detail a method for extracting features from graph structures of different scales in an extended-range electric vehicle energy management system. The method uses a non-shared parameter graph convolution module to independently process node features in small-scale and large-scale graph structures to extract and optimize key information in the battery management system. Figure 5 is a schematic diagram of the structure of the non-shared parameter graph convolution module provided in the embodiment of the present application, such as Figure 5 As shown, the process of independently processing node features in small-scale and large-scale graph structures using a non-shared parameter graph convolution module in the embodiment of the present application may include the following:

[0086] First, voltage, current and temperature data are collected from the vehicle’s battery management system, and corresponding small-scale and large-scale features are generated through preprocessing and feature extraction steps. These features initialize the node feature representation of the small-scale graph structure and the large-scale graph structure respectively.

[0087] Furthermore, in order to ensure the numerical stability and efficiency of the graph convolution operation, the adjacency matrices of the small-scale graph structure (A1) and the large-scale graph structure (A2) are symmetrically normalized. This step generates a normalized adjacency matrix, thereby providing a suitable input format for the graph convolution operation.

[0088] Furthermore, a non-shared parameter graph convolution module designed for small-scale features is applied, and a small-scale weight matrix is ​​used to perform graph convolution operations. Through this operation, the node features of each layer of the small-scale graph are updated to extract deeper local features.

[0089] Furthermore, a graph convolution module with non-shared parameters customized for large-scale features is used with a large-scale weight matrix to perform graph convolution operations. This step updates the node features of the large-scale graph to extract global features at the macro level.

[0090] Furthermore, after each graph convolution operation, the output of each layer is activated by the ReLU activation function to introduce nonlinearity and enhance the expressiveness of the model. Multiple layers of graph convolution operations are repeated until the required deep feature extraction requirements are achieved.

[0091] Furthermore, the resulting small-scale and large-scale deep features capture the detailed local changes and overall trends of battery behavior, respectively. These fused features can be used for further energy management decision analysis, such as predicting the optimal charge and discharge time of the battery, adjusting power output, and optimizing energy allocation strategies.

[0092] For example, in a specific example, non-shared parameter graph convolution modules are designed for different scales, and the structure of the non-shared parameter graph convolution module is as follows: Figure 5 As shown. First, perform graph convolution operation on the small scale (A1, X1), and the calculation process is as follows:

[0093]

[0094] in, is the l-th layer output of the small-scale graph in the non-shared parameter graph convolution, is the output of the l-1th layer, W1 (i) is the weight matrix of the lth layer of the small-scale graph in the non-shared parameter graph convolution.

[0095] Then, the graph convolution operation is performed on the large scale (A2, X2), and the calculation process is as follows:

[0096]

[0097] in, is the l-th layer output of the large-scale graph in the non-shared parameter graph convolution, is the output of the l-1th layer, W2 (l) is the weight matrix of the lth layer in the non-shared parameter graph convolution of the large-scale graph.

[0098] Through this non-shared parameter graph convolution processing method, this embodiment not only improves the sensitivity and response speed to changes in battery performance, but also optimizes the overall performance of the energy management system by precisely controlling and adjusting the weight matrix of each scale, thereby providing a more reliable and efficient energy utilization solution for extended-range electric vehicles.

[0099] In some embodiments, a pooling attention guidance module is used in the feature extraction process to perform a pooling operation on node features in the graph structure, including:

[0100] Performing an average pooling operation on each node of the first scale graph structure to extract detailed features of each node and form an average pooling feature set;

[0101] Performing a maximum pooling operation on each node of the second-scale graph structure to capture the global salient features of each node and form a maximum pooling feature set;

[0102] For each node in the average pooled feature set, the first attention coefficient is calculated using a fully connected layer and an activation function, and the first attention coefficient is weighted with the corresponding first-scale node feature;

[0103] For each node in the maximum pooling feature set, the second attention coefficient is calculated using a fully connected layer and an activation function, and the second attention coefficient is weighted with the corresponding second-scale node feature.

[0104] Specifically, this embodiment also describes in detail a pooling attention guidance module used in an energy management system of an extended-range electric vehicle. The module performs specific pooling operations on node features in graph structures of different scales and enhances the effect of feature extraction through an attention mechanism. Figure 6 is a schematic diagram of the structure of the pooled attention guidance module provided in the embodiment of the present application, such as Figure 6 As shown, the process of performing a pooling operation on node features in graph structures of different scales using a pooled attention guidance module in the embodiment of the present application may include the following contents:

[0105] First, an average pooling operation is performed on the node features extracted from the small-scale graph structure. This step involves calculating the average value of each node feature to capture more detailed information and form an average pooled feature set. At the same time, a maximum pooling operation is performed on the node features in the large-scale graph structure to highlight the most significant features of each node and form a maximum pooled feature set.

[0106] Furthermore, a fully connected layer and Sigmoid activation function are applied to the average pooled feature set to calculate the first attention coefficient a1 for each node. These coefficients reflect the importance of each node in the overall feature. Similarly, a similar mechanism is applied to the maximum pooled feature set to obtain the second attention coefficient a2, which emphasizes the importance of global salient features.

[0107] Furthermore, the calculated attention coefficient a1 is used to weight each node feature of the small-scale graph to form a weighted feature; and the attention coefficient a2 is used to weight each node feature of the large-scale graph to form a weighted feature.

[0108] For example, in a specific example, Max Pooling selects the maximum value in the input feature as the output value. It can well retain prominent features and capture prominent information. Average Pooling calculates the average value of the values ​​in the input window as the output value. It can smooth features and retain more detail information. Since large-scale features contain more prominent global information and small-scale features contain more detail information, a pooling attention guidance module is designed after the non-shared parameter graph convolution module. Figure 6 As shown, in order to guide the graph convolutional network to pay more attention to important nodes in the feature extraction process, the specific operation process is as follows:

[0109] First, the obtained small-scale features The average pooling is used for each node feature to extract the smooth features of each node so that more detailed information can be retained. Then, the attention coefficient a1 of each node based on the average pooling feature is obtained through the full connection layer and the Sigmoid activation function, which can guide the model to enhance the attention to detailed information and improve the model performance and expression ability. The calculation process of the attention coefficient a1 is as follows:

[0110]

[0111]

[0112] Among them, F i is the feature of each node in the small-scale graph, where i = 1, 2, ..., N, N is the number of nodes, and l is the length of the feature. Then the attention coefficient a1 is combined with the small-scale feature Multiply to get the weighted feature H1′:

[0113]

[0114] Next, the obtained large-scale features The maximum pooling is used for each node feature to extract the most representative salient information of each node, and then the attention coefficient of each node based on the maximum pooling feature is obtained through the fully connected layer and the Sigmoid activation function, which can guide the model to enhance the attention to the global salient information and improve the model performance and expression ability. The calculation process of the attention coefficient a2 is as follows:

[0115]

[0116]

[0117] Among them, F iis the feature of each node in the large-scale graph, where i = 1, 2, ..., N, N is the number of nodes, and l is the length of the feature. Then the attention coefficient a1 is combined with the small-scale feature Multiply to get the weighted feature H2′:

[0118]

[0119] Secondly, the shared parameter graph convolution output feature H c The small-scale weighted features H1′ and the large-scale weighted features H2′ in the non-shared parameter graph convolution of the fusion pooling attention guidance module are added together to obtain a more complete output feature representation, which is represented by H all express:

[0120] H all =H c +H1′+H2′

[0121] Finally, the fused feature H all The input is fed into the long short-term memory recurrent neural network (LSTM) for modeling and prediction, which can effectively process time series data and capture its long-term dependencies, learn the patterns and trends of time series data, and perform SOC prediction to obtain the SOC (battery state of charge) in the future time domain. t+m Indicates that SOC t is the SOC value at the current moment.

[0122] Through the above method of this embodiment, the system can more effectively distinguish and emphasize key details and global information in vehicle energy management. Small-scale processing enhances sensitivity to local changes, while large-scale processing helps the system grasp the overall trend. This not only improves the prediction accuracy of the model, but also makes the energy management strategy more sophisticated and efficient.

[0123] In some embodiments, generating an energy management instruction according to the battery state of charge, and optimizing energy management of the extended-range vehicle using the energy management instruction, includes:

[0124] A directed graph is constructed based on the battery state of charge, and the nodes in the directed graph are traversed using a predetermined energy management optimization algorithm to determine the path with minimum fuel consumption within the predicted time domain. Energy management instructions are generated based on the path with minimum fuel consumption, and the energy management of the extended-range vehicle is optimized using the energy management instructions.

[0125] Specifically, this embodiment also provides a method for generating and applying energy management instructions for an extended-range electric vehicle based on the battery state of charge (SOC). The method mainly relies on constructing a directed graph and applying an energy management optimization algorithm to minimize fuel consumption, thereby improving the energy efficiency of the vehicle.

[0126] First, the system collects real-time battery state of charge data from the battery management system of the extended-range electric vehicle, which reflects the current energy level of the battery. Based on the collected SOC data, the system predicts the state of charge in a certain period of time in the future to help plan energy use and allocation.

[0127] Furthermore, a directed graph is constructed using the predicted SOC data, where each node represents the SOC state at a specific time point, the edge represents the possibility of state transition, and the weight of the edge corresponds to the amount of fuel consumed to transition from one state to another. Such a directed graph enables the energy management system to visualize and quantify the effects of different energy management strategies.

[0128] Furthermore, an energy management optimization algorithm (such as Dijkstra's algorithm) is applied to traverse the nodes in the directed graph, with the goal of finding the minimum fuel consumption path from the current state to the predicted end state. This path represents an optimized energy management strategy that maximizes battery efficiency while minimizing fuel consumption.

[0129] Furthermore, based on the minimum fuel consumption path obtained through the optimization algorithm, the system generates specific energy management instructions. These instructions specify in detail how to adjust the battery charging and discharging state and the operating mode of the power system at different time points. These energy management instructions are then transmitted to the vehicle's control system to implement specific energy distribution and adjustment measures.

[0130] In some embodiments, a directed graph is constructed based on the battery state of charge, and nodes in the directed graph are traversed using a predetermined energy management optimization algorithm to determine a path with minimum fuel consumption within a predicted time domain, including:

[0131] Discretize the battery state of charge and time, and establish multiple discrete nodes, each of which represents the remaining battery power value at a specific moment;

[0132] A directed graph is constructed using nodes, where the edges between nodes represent the transition from one state to another, and weights are assigned to the edges, where the weights represent the fuel consumption of the state transition;

[0133] Define the untraversed node set and the traversed node set. Starting from the start node, put the traversed nodes into the traversed node set and put the adjacent nodes into the untraversed node set.

[0134] Select a node with the smallest cumulative weight with the starting node from the untraversed node set as the next traversal node, and update the shortest path weight from the next traversal node to the starting node;

[0135] Move the selected node from the untraversed node set to the traversed node set, and repeat the traversal operation until the end node is moved to the traversed node set;

[0136] When the terminal node moves into the traversed node set, the shortest path from the starting node to the terminal node is found by backtracking, and the shortest path is used as the path with the minimum fuel consumption in the prediction time domain.

[0137] Specifically, energy management strategy optimization based on energy management optimization algorithm is a process of formulating an efficient energy use plan for extended-range electric vehicles. This embodiment uses the battery state of charge (SOC) information predicted by the long short-term memory network (LSTM) to optimize fuel consumption and achieve real-time optimization of energy management of extended-range electric vehicles. Figure 7 is a schematic diagram of calculating the minimum fuel consumption path using the energy management optimization algorithm provided in an embodiment of the present application, such as Figure 7 As shown, the process of calculating the minimum fuel consumption path using the energy management optimization algorithm in the embodiment of the present application may include the following contents:

[0138] This embodiment uses an energy management optimization algorithm (such as the Dijkstra algorithm) to obtain the minimum fuel consumption path in the predicted time domain according to the SOC obtained by the above prediction. When traversing the nodes in the directed graph, the Dijkstra algorithm uses two sets to represent the nodes to be traversed and the nodes that have been traversed. The set of nodes to be traversed is represented by Y, and the set of nodes that have been traversed is represented by N.

[0139] First, put the starting point into Y, then put the adjacent nodes of the starting point, that is, the nodes that the starting point can reach, into Y, and put the starting point in Y into N, because it has been traversed. The next step is to select a node from Y for the next traversal. When selecting a node, the priority is determined based on the cost of the current node from the starting point, and the cost of the previous node from the starting point is calculated based on the weight of the edge of the directed graph. When the end point appears in N, a feasible path from the starting point to the end point has been found, and the feasible path can be output by backtracking.

[0140] Furthermore, this embodiment focuses on the fuel consumption of the vehicle, which is always greater than or equal to 0. A directed graph is established with this as the edge weight. Based on the above-mentioned predicted SOC, the values ​​of the SOC at the start and end are determined, that is, there is a starting point and an end point, and the Dijkstra algorithm can be effectively used to solve it. Because the input of the Dijkst ra algorithm is a directed graph, the first step is to discretize the operating conditions. The discrete time step length is selected as 1s, and the range of SOC changes is set to 0.8 to 0.3; the minimum change interval of SOC is selected as 0.001. By discretizing SOC and time, a directed graph with a certain operating condition is established.

[0141] Furthermore, each node in the directed graph contains two pieces of information: SOC and time. Figure 7 The starting point of the Dijkstra algorithm represents the initial state of the battery, and SOC is SOC t , time is 0; Figure 7 The middle endpoint indicates the battery's final state SOC t+m . The diffusion of nodes represents the change of SOC over time, and the edge between nodes represents the direction of SOC change. Since time is irreversible, the diffusion direction of nodes can only be from left to right. At the same time, different control strategies will lead to different fuel consumption and SOC changes, so each edge in the directed graph has its own "length". The length of the edge is the fuel consumption when the SOC changes. In a directed graph, there are many paths to choose from from the starting point to the end point, and each path represents different fuel consumption. In the above way, the Dijkstra algorithm can find a path with the least fuel consumption, and ultimately achieve real-time optimization of energy management of extended-range vehicles and improve fuel economy.

[0142] According to the technical solution provided in the above-mentioned embodiment of the present application, the purpose of the technical solution of the present application is to provide a real-time optimization method for energy management of an extended-range vehicle based on SOC prediction. The embodiment of the present application designs a multi-scale fused graph convolution module, which better handles the changing working conditions under complex working conditions during the actual driving of the vehicle through the graph convolution module with shared parameters and non-shared parameters, and fuses the pooled attention guidance module to output more representative features, thereby improving the accuracy of real-time SOC prediction, and then obtains the minimum fuel consumption path based on the predicted SOC in the time domain based on the Dijkstra algorithm, thereby improving the fuel economy of the extended-range vehicle.

[0143] 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.

[0144] Figure 8 Schematic diagram of the structure of the energy management optimization device for the extended-range vehicle provided in the embodiment of the present application. Figure 8 As shown, the energy management optimization device of the extended-range vehicle includes:

[0145] The construction module 801 is configured to extract features from the target parameter signal of the vehicle to obtain feature information of different scales, and construct a graph structure for the features of each scale according to the feature information of different scales, wherein the graph structure includes the feature information of the corresponding scale and an adjacency matrix;

[0146] The processing module 802 is configured to process the graph structure corresponding to each scale respectively using a preset shared parameter graph convolution module to extract fusion features corresponding to each scale, aggregate the fusion features, and obtain aggregate features representing shared information;

[0147] The extraction module 803 is configured to use the preset non-shared parameter graph convolution modules corresponding to different scales to extract features of the graph structures of corresponding scales respectively, and use the pooling attention guidance module in the feature extraction process to perform a pooling operation on the node features in the graph structure, and weight them to obtain weighted features corresponding to different scales;

[0148] The prediction module 804 is configured to fuse the aggregated features with the weighted features corresponding to different scales, and input the fused features into the prediction model for prediction to obtain the battery state of charge in the prediction time domain;

[0149] The optimization module 805 is configured to generate energy management instructions according to the battery state of charge, and use the energy management instructions to optimize the energy management of the extended-range vehicle.

[0150] In some embodiments, Figure 8 The construction module 801 uses a multi-scale one-dimensional convolution network to perform a multi-scale one-dimensional convolution operation on the target parameter signal to obtain first-scale feature information and second-scale feature information; wherein the multi-scale one-dimensional convolution network includes one-dimensional convolution layers with different convolution kernel sizes and depths, and the target parameter signal includes a voltage signal, a current signal, and a temperature signal.

[0151] In some embodiments, Figure 8 The construction module 801 uses the first scale feature information as a node to construct a first scale graph structure corresponding to the first scale feature, and uses the second scale feature information as a node to construct a second scale graph structure corresponding to the second scale feature; the first scale feature information and the second scale feature information are divided according to the number of nodes, and the distances between the nodes in the different scale graph structures are calculated respectively to form adjacency matrices corresponding to different scales.

[0152] In some embodiments, Figure 8 The processing module 802 initializes the feature representation of the first-scale graph structure and the second-scale graph structure, and inputs the node features into the shared parameter graph convolution module; performs symmetric normalization processing on the adjacency matrix of the first-scale graph structure and the second-scale graph structure to obtain a normalized adjacency matrix; performs graph convolution operation on the first-scale graph structure and the second-scale graph structure, and uses the same weight matrix to update the node features of each layer to extract fusion features of each scale; uses a nonlinear activation function to activate the output of each layer of graph convolution, and repeats the multi-layer graph convolution operation until the fusion features that meet the depth requirements are extracted.

[0153] In some embodiments, Figure 8 The extraction module 803 initializes the node feature representation of the first-scale graph structure and the second-scale graph structure; performs symmetric normalization operations on the first-scale graph structure and the second-scale graph structure respectively to generate normalized adjacency matrices; applies the corresponding non-shared parameter graph convolution module to the first-scale graph structure, and performs graph convolution operations using the weight matrix of the first scale to update the node feature representation of each layer; applies the corresponding non-shared parameter graph convolution module to the second-scale graph structure, and performs graph convolution operations using the weight matrix of the second scale to update the node feature representation of each layer.

[0154] In some embodiments, Figure 8 The extraction module 803 performs an average pooling operation on each node of the first-scale graph structure to extract the detail features of each node and form an average pooling feature set; performs a maximum pooling operation on each node of the second-scale graph structure to capture the global salient features of each node and form a maximum pooling feature set; for each node in the average pooling feature set, a first attention coefficient is calculated using a fully connected layer and an activation function, and the first attention coefficient is weighted with the corresponding first-scale node feature; for each node in the maximum pooling feature set, a second attention coefficient is calculated using a fully connected layer and an activation function, and the second attention coefficient is weighted with the corresponding second-scale node feature.

[0155] In some embodiments, Figure 8 The optimization module 805 constructs a directed graph based on the battery state of charge, and uses a predetermined energy management optimization algorithm to traverse the nodes in the directed graph to determine the minimum fuel consumption path within the predicted time domain, generates energy management instructions based on the minimum fuel consumption path, and uses the energy management instructions to optimize the energy management of the extended-range vehicle.

[0156] In some embodiments, Figure 8The optimization module 805 discretizes the battery state of charge and time, establishes multiple discrete nodes, and each node represents the remaining battery power value at a specific moment; uses the nodes to construct a directed graph, in which the edge between the nodes represents the transition from one state to another state, and assigns weights to the edges, and the weights represent the fuel consumption of the state transfer; defines an untraversed node set and a traversed node set, starting from the starting node, puts the traversed nodes into the traversed node set, and puts the adjacent nodes into the untraversed node set; selects a node with the smallest cumulative weight with the starting node in the untraversed node set as the next traversed node, and updates the shortest path weight from the next traversed node to the starting node; moves the selected node from the untraversed node set to the traversed node set, and repeats the traversal operation until the end node is moved into the traversed node set; when the end node is moved into the traversed node set, finds the shortest path from the starting node to the end node by backtracking, and uses the shortest path as the path with minimum fuel consumption in the prediction time domain.

[0157] 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.

[0158] Fig. 9 Schematic diagram of the structure of the electronic device 9 provided in the embodiment of the present application. Fig. 9 As shown, the electronic device 9 of this embodiment includes: a processor 901, a memory 902, and a computer program 903 stored in the memory 902 and executable on the processor 901. When the processor 901 executes the computer program 903, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor 901 executes the computer program 903, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0159] Exemplarily, the computer program 903 may be divided into one or more modules / units, which are stored in the memory 902 and executed by the processor 901 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 903 in the electronic device 9.

[0160] The electronic device 9 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 9 may include, but is not limited to, a processor 901 and a memory 902. Those skilled in the art will appreciate that Fig. 9It is only an example of the electronic device 9 and does not constitute a limitation of the electronic device 9. 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.

[0161] The processor 901 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.

[0162] The memory 902 may be an internal storage unit of the electronic device 9, for example, a hard disk or memory of the electronic device 9. The memory 902 may also be an external storage device of the electronic device 9, 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 9. Further, the memory 902 may also include both an internal storage unit of the electronic device 9 and an external storage device. The memory 902 is used to store computer programs and other programs and data required by the electronic device. The memory 902 may also be used to temporarily store data that has been output or is to be output.

[0163] 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.

[0164] 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.

[0165] 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.

[0166] 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.

[0167] 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.

[0168] 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.

[0169] 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.

[0170] 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. An energy management optimization method for an extended-range vehicle, characterized in that: include: Extracting features of the target parameter signal of the vehicle to obtain feature information of different scales, and constructing a graph structure for the features of each scale according to the feature information of different scales, wherein the graph structure includes the feature information of the corresponding scale and an adjacency matrix; The graph structures corresponding to each scale are processed respectively using a preset shared parameter graph convolution module to extract fusion features corresponding to each scale, and the fusion features are aggregated to obtain aggregate features representing shared information; Using the preset non-shared parameter graph convolution modules corresponding to different scales, feature extraction is performed on the graph structures of corresponding scales respectively, and in the feature extraction process, the pooling attention guidance module is used to perform pooling operation on the node features in the graph structure, and weighted to obtain weighted features corresponding to different scales; The aggregated features are fused with weighted features corresponding to different scales, and the fused features are input into a prediction model for prediction to obtain a battery state of charge in a prediction time domain; An energy management instruction is generated according to the battery state of charge, and the energy management of the extended-range vehicle is optimized using the energy management instruction.

2. The method according to claim 1, characterized in that The feature extraction of the target parameter signal of the vehicle is performed to obtain feature information of different scales, including: A multi-scale one-dimensional convolution operation is performed on the target parameter signal using a multi-scale one-dimensional convolution network to obtain first-scale feature information and second-scale feature information; wherein the multi-scale one-dimensional convolution network includes one-dimensional convolution layers with different convolution kernel sizes and depths, and the target parameter signal includes a voltage signal, a current signal, and a temperature signal.

3. The method according to claim 2, characterized in that The step of constructing a graph structure for features of each scale according to the feature information of different scales includes: The first scale feature information is used as a node to construct a first scale graph structure corresponding to the first scale feature, and the second scale feature information is used as a node to construct a second scale graph structure corresponding to the second scale feature; the first scale feature information and the second scale feature information are divided according to the number of nodes, and the distances between the nodes in the different scale graph structures are calculated respectively to form adjacency matrices corresponding to different scales.

4. The method according to claim 3, characterized in that The method of using a preset shared parameter graph convolution module to process the graph structure corresponding to each scale respectively to extract the fusion features corresponding to each scale includes: Initializing feature representations of the first scale graph structure and the second scale graph structure, and inputting node features into the shared parameter graph convolution module; Performing symmetric normalization processing on the adjacency matrices of the first scale graph structure and the second scale graph structure to obtain a normalized adjacency matrix; Performing a graph convolution operation on the first-scale graph structure and the second-scale graph structure, and updating the node features of each layer using the same weight matrix to extract fusion features of each scale; The output of each layer of graph convolution is activated using a nonlinear activation function, and the multi-layer graph convolution operation is repeated until the fusion features that meet the depth requirements are extracted.

5. The method according to claim 1, characterized in that The method uses the preset non-shared parameter graph convolution modules corresponding to different scales to extract features of the graph structures of the corresponding scales, including: Initialize node feature representations of the first scale graph structure and the second scale graph structure; Performing symmetric normalization operations on the first scale graph structure and the second scale graph structure respectively to generate a normalized adjacency matrix; Applying a corresponding non-shared parameter graph convolution module to the first scale graph structure, and performing a graph convolution operation using a weight matrix of the first scale, so as to update a node feature representation of each layer; A corresponding non-shared parameter graph convolution module is applied to the second-scale graph structure, and a graph convolution operation is performed using a weight matrix of the second scale to update the node feature representation of each layer.

6. The method according to claim 5, characterized in that The method of using a pooling attention guidance module in the feature extraction process to perform a pooling operation on the node features in the graph structure includes: Performing an average pooling operation on each node of the first scale graph structure to extract detail features of each node and form an average pooling feature set; Performing a maximum pooling operation on each node of the second scale graph structure to capture the global salient features of each node and form a maximum pooling feature set; For each node in the average pooled feature set, a first attention coefficient is calculated using a fully connected layer and an activation function, and the first attention coefficient is weighted with a corresponding first-scale node feature; For each node in the maximum pooling feature set, a second attention coefficient is calculated using a fully connected layer and an activation function, and the second attention coefficient is weighted with the corresponding second-scale node feature.

7. The method according to claim 1, characterized in that The generating of energy management instructions according to the battery state of charge, and using the energy management instructions to optimize energy management of the extended-range vehicle, comprises: A directed graph is constructed based on the battery state of charge, and nodes in the directed graph are traversed using a predetermined energy management optimization algorithm to determine a minimum fuel consumption path within the predicted time domain, energy management instructions are generated according to the minimum fuel consumption path, and energy management optimization is performed on the extended-range vehicle using the energy management instructions.

8. The method according to claim 7, characterized in that The step of constructing a directed graph based on the battery state of charge and traversing nodes in the directed graph using a predetermined energy management optimization algorithm to determine a path with minimum fuel consumption within the predicted time domain includes: Discretize the battery state of charge and time to establish multiple discrete nodes, each node representing the remaining battery power value at a specific moment; A directed graph is constructed using nodes, wherein an edge between the nodes represents a transition from one state to another state, and a weight is assigned to the edge between the nodes, wherein the weight represents the fuel consumption of the state transition; Define the untraversed node set and the traversed node set. Starting from the start node, put the traversed nodes into the traversed node set and put the adjacent nodes into the untraversed node set. Selecting a node with the smallest cumulative weight with the starting node from the untraversed node set as the next traversal node, and updating the shortest path weight from the next traversal node to the starting node; Moving the selected node from the untraversed node set into the traversed node set, and repeating the traversal operation until the end node is moved into the traversed node set; When the terminal node moves into the traversed node set, the shortest path from the starting node to the terminal node is found by backtracking, and the shortest path is used as the minimum fuel consumption path in the predicted time domain.

9. An energy management optimization device for an extended-range vehicle, characterized in that: include: A construction module is configured to extract features from the target parameter signal of the vehicle to obtain feature information of different scales, and to construct a graph structure for features of each scale according to the feature information of different scales, wherein the graph structure includes feature information of the corresponding scale and an adjacency matrix; A processing module is configured to process the graph structure corresponding to each scale respectively using a preset shared parameter graph convolution module to extract fusion features corresponding to each scale, aggregate the fusion features, and obtain aggregate features representing shared information; The extraction module is configured to use the preset non-shared parameter graph convolution modules corresponding to different scales to extract features of the graph structures of the corresponding scales respectively, and use the pooling attention guidance module in the feature extraction process to perform a pooling operation on the node features in the graph structure, and weight them to obtain weighted features corresponding to different scales; A prediction module is configured to fuse the aggregated features with weighted features corresponding to different scales, and input the fused features into a prediction model for prediction to obtain a battery state of charge in a prediction time domain; The optimization module is configured to generate an energy management instruction according to the battery state of charge, and use the energy management instruction to optimize the energy management of the extended-range vehicle.

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 8 are implemented.

Images