Vehicle speed prediction method and device, electronic equipment and storage medium

Abstract

The application relates to the technical field of artificial intelligence, and provides a vehicle speed prediction method and device, electronic equipment and a storage medium. The method comprises the following steps: performing interval division on time series data of historical vehicle speed, performing graph construction on vehicle speed data in each divided interval to obtain a feature sequence comprising an adjacency matrix, splicing the adjacency matrix of the vehicle speed data in each interval to obtain an adjacency matrix of the historical vehicle speed data, and splicing the feature sequence of the vehicle speed data in each interval to obtain a feature sequence of the historical vehicle speed data. After time attenuation processing is performed on the adjacency matrix of the historical vehicle speed data, the trained graph convolution network is used to aggregate the time-attenuated adjacency matrix and the feature sequence of the historical vehicle speed data. Finally, the aggregation result is input into a trained prediction unit to obtain a vehicle speed prediction result. The method can deeply mine the vehicle speed space feature information in each time period in the historical vehicle speed data, increase the weight of recent vehicle speed data, and thus improve the accuracy and real-time performance of vehicle speed prediction.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a vehicle speed prediction method, device, electronic device, and storage medium. Background Technology

[0002] Developing new energy vehicles and increasing their market share has become an important means to alleviate oil resource shortages, solve environmental pollution problems, and achieve the restructuring and upgrading of the automotive industry. Under current technological conditions, range-extended electric vehicles (REEVs) are considered a promising energy-saving and emission-reduction solution, possessing the dual advantages of fuel efficiency and range extension. While achieving low emissions, low cost, and high efficiency, they can effectively solve the range anxiety problem of pure electric vehicles. Due to their unique energy-saving and emission-reduction advantages, REEVs have received widespread attention and research. When the total power of a REEV meets specific power requirements, energy management strategies can allocate power between the motor and engine to achieve energy saving and emission reduction. Since the output of different power sources in a REEV is directly related to driving conditions, accurately obtaining driving condition information is extremely important, which necessitates speed prediction.

[0003] Commonly used prediction methods in related technologies include those based on exponential function prediction models and those based on recurrent neural network (RNN) prediction models. The former is simple to implement, computationally inexpensive, and fast, but its prediction accuracy is low, and it is prone to large errors when the prediction time domain is too long. The latter only focuses on the temporal dependencies between historical vehicle speed data, ignoring the spatial correlations between vehicle speed data and the influence of correlation decay due to time distance, thus limiting the accuracy of the vehicle speed prediction model. Summary of the Invention

[0004] In view of this, embodiments of this application provide a vehicle speed prediction method, apparatus, electronic device, and storage medium to solve the problem of low accuracy in vehicle speed prediction in the prior art.

[0005] A first aspect of this application provides a vehicle speed prediction method, including:

[0006] Obtain the vehicle's historical speed data for this trip; the historical speed data is time-series data.

[0007] The historical vehicle speed data is divided into N interval vehicle speed data with the same time length. The time length of the interval vehicle speed data is the difference between the time corresponding to the last data and the time corresponding to the first data in the interval vehicle speed data, and N is a positive integer greater than 1.

[0008] Construct graph structures for the vehicle speed data of each interval, wherein the graph structure of each interval includes an adjacency matrix and a feature sequence;

[0009] The adjacency matrices of N interval speed data are concatenated to obtain the adjacency matrix of historical speed data. The feature sequences of N interval speed data are concatenated to obtain the feature sequence of historical speed data.

[0010] The adjacency matrix of historical vehicle speed data is subjected to time decay processing to obtain the updated adjacency matrix of historical vehicle speed data.

[0011] The pre-trained first graph convolutional network is used to aggregate the adjacency matrix of the updated historical vehicle speed data and the feature sequence of the historical vehicle speed data to obtain the feature sequence of the updated historical vehicle speed data.

[0012] The updated feature sequence of historical vehicle speed data is input into the trained prediction unit to obtain the vehicle speed prediction result.

[0013] A second aspect of this application provides a vehicle speed prediction device, comprising:

[0014] The acquisition module is configured to acquire the vehicle's historical speed data for the current trip, and the historical speed data is time series data.

[0015] The segmentation module is configured to divide historical vehicle speed data into N interval vehicle speed data with the same time length. The time length of the interval vehicle speed data is the difference between the time corresponding to the last data in the interval vehicle speed data and the time corresponding to the first data. N is a positive integer greater than 1.

[0016] The graph construction module is configured to construct the graph structure of the vehicle speed data for each interval, wherein the graph structure of the vehicle speed data for each interval includes an adjacency matrix and a feature sequence.

[0017] The splicing module is configured to splice the adjacency matrices of N interval vehicle speed data to obtain the adjacency matrix of historical vehicle speed data, and to splice the feature sequences of N interval vehicle speed data to obtain the feature sequences of historical vehicle speed data.

[0018] The attenuation module is configured to perform time attenuation processing on the adjacency matrix of historical vehicle speed data to obtain the updated adjacency matrix of historical vehicle speed data.

[0019] The aggregation module is configured to use the trained first graph convolutional network to aggregate the adjacency matrix of the updated historical vehicle speed data and the feature sequence of the historical vehicle speed data to obtain the feature sequence of the updated historical vehicle speed data.

[0020] The prediction module is configured to input the feature sequence of the updated historical vehicle speed data into the trained prediction unit to obtain the vehicle speed prediction result.

[0021] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0022] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0023] The beneficial effects of this application embodiment compared with the prior art are as follows: This application embodiment divides the time series data of historical vehicle speed into intervals, constructs a graph for each interval of vehicle speed data to obtain a feature sequence including an adjacency matrix, then concatenates the adjacency matrices of each interval of vehicle speed data to obtain the adjacency matrix of historical vehicle speed data, and concatenates the feature sequences of each interval of vehicle speed data to obtain the feature sequence of historical vehicle speed data. After performing time decay processing on the adjacency matrix of historical vehicle speed data, a trained graph convolutional network is used to aggregate the time decayed adjacency matrix and the feature sequence of historical vehicle speed data. Finally, the aggregation result is input into a trained prediction unit to obtain the vehicle speed prediction result. This allows for in-depth mining of the spatial feature information of vehicle speed in each time period in the historical vehicle speed data, as well as increasing the weight of recent vehicle speed data, thereby improving the accuracy and real-time performance of vehicle speed prediction and enhancing the user experience. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating a vehicle speed prediction method provided in an embodiment of this application.

[0026] Figure 2 This is a flowchart illustrating the method for constructing a graph structure of interval vehicle speed data provided in an embodiment of this application.

[0027] Figure 3 This is a flowchart illustrating the method for obtaining a second adjacency matrix by sparse processing of a first adjacency matrix according to an embodiment of this application.

[0028] Figure 4This is a flowchart illustrating the method for extracting key information from a second adjacency matrix to obtain a third adjacency matrix, as provided in an embodiment of this application.

[0029] Figure 5 This is a schematic diagram of the graph structure of the vehicle speed data for each section provided in the embodiments of this application.

[0030] Figure 6 This is a flowchart illustrating a method for concatenating adjacency matrices of N interval vehicle speed data to obtain an adjacency matrix of historical vehicle speed data, as provided in an embodiment of this application.

[0031] Figure 7 This is a flowchart illustrating a method for performing time decay processing on the adjacency matrix of historical vehicle speed data to obtain an updated adjacency matrix of historical vehicle speed data, as provided in an embodiment of this application.

[0032] Figure 8 This is a schematic diagram of the adjacency matrix and time decay factor matrix of historical vehicle speed data provided in the embodiments of this application.

[0033] Figure 9 This is a schematic diagram of a vehicle speed prediction device provided in an embodiment of this application.

[0034] Figure 10 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0035] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0036] A vehicle speed prediction method and apparatus according to embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0037] As mentioned above, the commonly used prediction methods in related technologies include those based on exponential function prediction models and those based on RNN prediction models. The former is simple to implement, computationally inexpensive, and fast, but its prediction accuracy is not high, and it is prone to excessive errors when the prediction time domain is too long. The latter only focuses on the temporal dependence between historical vehicle speed data, ignoring the spatial correlation between vehicle speed data and the influence of correlation decay due to time distance, thus limiting the accuracy of the vehicle speed prediction model.

[0038] In view of this, this application provides a vehicle speed prediction method. It divides historical vehicle speed time-series data into intervals, constructs a graph for each interval's vehicle speed data to obtain a feature sequence including an adjacency matrix, then concatenates the adjacency matrices of each interval's vehicle speed data to obtain the adjacency matrix of the historical vehicle speed data, and concatenates the feature sequences of each interval's vehicle speed data to obtain the feature sequence of the historical vehicle speed data. After time decay processing of the adjacency matrix of the historical vehicle speed data, a trained graph convolutional network is used to aggregate the time-decayed adjacency matrix and the feature sequence of the historical vehicle speed data. Finally, the aggregation result is input into a trained prediction unit to obtain the vehicle speed prediction result. This method can deeply mine the spatial feature information of vehicle speed in each time period of historical vehicle speed data and increase the weight of recent vehicle speed data during vehicle speed prediction, thereby improving the accuracy and real-time performance of vehicle speed prediction and enhancing user experience.

[0039] Figure 1 This is a flowchart illustrating a vehicle speed prediction method provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0040] In step S101, the vehicle's historical speed data for this trip is obtained.

[0041] Among them, the historical vehicle speed data is time series data.

[0042] In step S102, the historical vehicle speed data is divided into N interval vehicle speed data with the same time length.

[0043] The time length of the interval speed data is the difference between the time corresponding to the last data point and the time corresponding to the first data point in the interval speed data, where N is a positive integer greater than 1.

[0044] In step S103, a graph structure of the vehicle speed data for each section is constructed.

[0045] The graph structure of the vehicle speed data for each interval includes an adjacency matrix and a feature sequence.

[0046] In step S104, the adjacency matrices of N interval speed data are concatenated to obtain the adjacency matrix of historical speed data, and the feature sequences of N interval speed data are concatenated to obtain the feature sequences of historical speed data.

[0047] In step S105, the adjacency matrix of the historical vehicle speed data is subjected to time decay processing to obtain the updated adjacency matrix of the historical vehicle speed data.

[0048] In step S106, the trained first graph convolutional network is used to aggregate the adjacency matrix of the updated historical vehicle speed data and the feature sequence of the historical vehicle speed data to obtain the feature sequence of the updated historical vehicle speed data.

[0049] In step S107, the feature sequence of the updated historical vehicle speed data is input into the trained prediction unit to obtain the vehicle speed prediction result.

[0050] In this embodiment, the vehicle speed prediction method can be executed by a server or by a terminal with certain computing capabilities. For ease of description and explanation, the following description will use the example of the vehicle speed prediction method being executed by a server.

[0051] In this embodiment of the application, after receiving a vehicle speed prediction command, the server can obtain the historical vehicle speed data for the current trip, which is time-series data. In one example, if the server receives the vehicle speed prediction command at time t during the vehicle's current trip, it can obtain the historical vehicle speed data for the time interval from time t1 to time t2. Here, time t2 can be equal to or less than time t, and time t1 can be equal to 0 or greater than 0 and less than time t2.

[0052] In this embodiment, the acquired historical vehicle speed data can be divided into N interval vehicle speed data with the same time length. The time length of each interval vehicle speed data is the difference between the time corresponding to the last data point and the time corresponding to the first data point, where N is a positive integer greater than 1. In one example, if the acquired historical vehicle speed data is 30 seconds of speed data X, dividing X into segments with a time length T equal to 10 seconds results in three interval vehicle speed data segments, each with a time length of 10 seconds. In each interval vehicle speed data segment, the time difference between the time corresponding to the last data point and the time corresponding to the first data point is 10 seconds. Further, the three interval vehicle speed data segments can be represented by X1, X2, and X3, respectively.

[0053] In this embodiment, a graph structure for the vehicle speed data of each interval can be constructed separately. Each interval's graph structure includes an adjacency matrix and a feature sequence. In one example, the constructed graph structure can be represented by G. i =(A i H i ) represents, where i equals 1, 2, or 3, and H i This represents the vehicle speed data X for the i-th interval. i The feature sequence obtained by feature extraction.

[0054] In this embodiment of the application, the directly constructed graph structure cannot reflect the time dependence between the features in the vehicle speed data. Therefore, time decay processing can be performed on the graph structure of the vehicle speed data to obtain a more accurate feature sequence of historical vehicle speed data, and then the feature sequence can be used to predict a more accurate predicted vehicle speed.

[0055] In this embodiment, the adjacency matrices of N interval speed data are first concatenated to obtain the adjacency matrix of historical speed data. Then, the feature sequences of the N interval speed data are concatenated to obtain the feature sequence of historical speed data. Next, time decay processing is applied to the adjacency matrix of the historical speed data to obtain the updated adjacency matrix. Finally, the trained first-graph convolutional network is used to aggregate the updated adjacency matrix and the feature sequence of the historical speed data to obtain the updated feature sequence of historical speed data.

[0056] The updated feature sequence of historical vehicle speed data can better reflect the time dependencies between various features in the vehicle speed data. By inputting the updated feature sequence of historical vehicle speed data into a trained prediction unit, the vehicle speed prediction result can be obtained. The prediction unit can be, for example, a fully connected layer in a trained neural network model, or a trained classifier.

[0057] According to the technical solution provided in the embodiments of this application, the time series data of historical vehicle speed is divided into intervals. The vehicle speed data of each interval is then used to construct a graph to obtain a feature sequence including an adjacency matrix. The adjacency matrices of the vehicle speed data of each interval are then concatenated to obtain the adjacency matrix of the historical vehicle speed data. The feature sequences of the vehicle speed data of each interval are then concatenated to obtain the feature sequence of the historical vehicle speed data. After time decay processing is applied to the adjacency matrix of the historical vehicle speed data, a trained graph convolutional network is used to aggregate the time decayed adjacency matrix and the feature sequence of the historical vehicle speed data. Finally, the aggregation result is input into a trained prediction unit to obtain the vehicle speed prediction result. This method can deeply mine the spatial feature information of vehicle speed in each time period in the historical vehicle speed data and increase the weight of recent vehicle speed data when making vehicle speed prediction, thereby improving the accuracy and real-time performance of vehicle speed prediction and enhancing the user experience.

[0058] Figure 2 This is a flowchart illustrating the method for constructing a graph structure of interval vehicle speed data provided in an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0059] In step S201, the trained feature extraction network is used to extract features from the interval vehicle speed data to obtain the first feature sequence of the interval vehicle speed data.

[0060] In step S202, the connection relationship between each feature in the first feature sequence and other features is determined, and the first adjacency matrix of the interval vehicle speed data is constructed using the connection relationship between each feature.

[0061] The first adjacency matrix consists of p rows and p columns. The matrix element in the i1th row and j1st column represents the connection relationship between the i1th feature and the j1st feature in the first feature sequence. Here, p is a positive integer greater than 1, and i1 and j1 are both positive integers greater than or equal to 1 and less than p.

[0062] In step S203, the first adjacency matrix is ​​sparsed to obtain the second adjacency matrix.

[0063] In step S204, the trained second graph convolutional network is used to aggregate the second adjacency matrix and the first feature sequence to obtain the second feature sequence of the interval vehicle speed data.

[0064] In step S205, key information is extracted from the second adjacency matrix to obtain the third adjacency matrix.

[0065] The third adjacency matrix consists of q rows and q columns, where q is a positive integer greater than 1 and less than p.

[0066] In step S206, the target features corresponding to each row in the third adjacency matrix are determined, and only the target features are retained in the second feature sequence to obtain the third feature sequence.

[0067] In step S207, the graph structure formed by the third adjacency matrix and the third feature sequence is determined to be the graph structure of the interval vehicle speed data.

[0068] In this embodiment of the application, when constructing the graph structure of the interval vehicle speed data, a trained feature extraction network can first be used to extract features from the interval vehicle speed data to obtain a first feature sequence of the interval vehicle speed data. Next, the connection relationships between each feature in the first feature sequence can be determined, and the first adjacency matrix of the interval vehicle speed data can be constructed using the determined connection relationships between each feature.

[0069] In one example, the Euclidean distance between features can be used to characterize the connectivity between features. Continuing with the example of the interval speed data including X1, X2, and X3, calculating the Euclidean distance between each speed data point in X1, X2, and X3 and other speed data points yields their respective first adjacency matrices E1, E2, and E3. For each first adjacency matrix, where element E... i1，j1 E1 represents the Euclidean distance between the i1th feature and the j1st feature. It is understood that E1, E2, and E3 are all symmetric matrices.

[0070] In this embodiment, since the first adjacency matrix includes all connections between each feature and all other features, processing all these connections in subsequent processing would result in high computational complexity and long processing time, significantly reducing the real-time performance of vehicle speed prediction. Therefore, this embodiment can perform sparse processing on the determined first adjacency matrix to obtain a second adjacency matrix. The sparsely processed second adjacency matrix retains only the connections between some features, reducing computational complexity and improving prediction real-time performance. In one example, the sparsely processed second adjacency matrix can be represented as A. i '.

[0071] In this embodiment, a trained second graph convolutional network can be used to aggregate the second adjacency matrix and the first feature sequence to obtain the second feature sequence of the interval vehicle speed data. The aggregation process can involve performing a symmetric normalization operation on the second adjacency matrix; then, using the activation function of the trained second graph convolutional network, performing activation calculations on the product of the symmetrically normalized second adjacency matrix, the first feature sequence, and the preset weight matrix of the interval vehicle speed data to obtain the second feature sequence of the interval vehicle speed data. That is, Among them, H i ' is the second characteristic sequence, W i The preset weight matrix for the interval vehicle speed data, Let σ be the symmetric normalized adjacency matrix of the second adjacency matrix, and σ be the ReLU activation function.

[0072] In this embodiment, to further reduce computational complexity and improve prediction real-time performance, key information can be extracted from the second adjacency matrix to obtain a third adjacency matrix. The third adjacency matrix has fewer rows and columns than the second adjacency matrix. It is understood that for any adjacency matrix, each row represents a feature, and each column represents the connection relationship between a feature and other features. Based on this, the target feature corresponding to each row in the third adjacency matrix can be determined, and only this target feature can be retained in the second feature sequence, while the remaining features are deleted to obtain the third feature sequence.

[0073] Finally, it can be determined that the graph structure formed by the third adjacency matrix and the third feature sequence is the graph structure of the interval vehicle speed data.

[0074] Figure 3 This is a flowchart illustrating the method for obtaining a second adjacency matrix by performing sparse processing on a first adjacency matrix, as provided in an embodiment of this application. Figure 3 As shown, the method includes the following steps:

[0075] In step S301, the first k matrix elements with the smallest values ​​in each row of the first adjacency matrix are determined as the first target elements.

[0076] Where k is a positive integer greater than 1 and less than p.

[0077] In step S302, the values ​​of all matrix elements in the first adjacency matrix except for the first target element are set to 0 to obtain the second adjacency matrix.

[0078] In this embodiment, a minimum k(min-k) sorting mechanism can be used to sparse the first adjacency matrix. In one example, if k is set to 5, the five smallest values ​​in each row of the first adjacency matrix can be retained, and their remainders can be set to 0, thus transforming the dense first adjacency matrix into a sparse second adjacency matrix. Sparse matrices are efficient in both storage and computation, significantly improving the real-time performance of vehicle speed prediction.

[0079] Figure 4 This is a flowchart illustrating the method for extracting key information from a second adjacency matrix to obtain a third adjacency matrix, as provided in an embodiment of this application. Figure 4 As shown, the method includes the following steps:

[0080] In step S401, the values ​​of all elements in each row of the second adjacency matrix are summed to obtain the sum of the row elements of the second adjacency matrix.

[0081] In step S402, the first q row elements with the smallest values ​​in the row elements and their corresponding rows are determined as the target row.

[0082] In step S403, q target features corresponding to the target row are determined.

[0083] In step S404, the connection relationship between each of the q target features and other target features is determined, and the third adjacency matrix is ​​constructed using the connection relationship between the target features.

[0084] In the third adjacency matrix, the matrix element in the i2th row and j2nd column represents the connection relationship between the i2th feature and the j2nd feature in the target feature, where i2 and j2 are both positive integers greater than or equal to 1 and less than q.

[0085] As mentioned earlier, each row in the adjacency matrix corresponds to a feature, and each column in that row represents the connection relationship, or connection weight, between that feature and other features. For an adjacency matrix constructed using Euclidean distance, the smaller the sum of the values ​​in each column of each row, the stronger the relationship between that feature and other features. Therefore, the elements in each row of the second adjacency matrix can be filtered to extract key information.

[0086] In this embodiment, the sum of all elements in each row of the second adjacency matrix is ​​obtained. Then, q target features corresponding to the target row are determined. Finally, the connection relationships between each of the q target features and other target features are determined, and the connection relationships between the target features are used to construct the third adjacency matrix.

[0087] Taking the processing of the above interval speed data X1 as an example, if the value of q is 3, we can first sum all columns corresponding to each row in the second adjacency matrix of X1, then select the three rows with the smallest sum and determine the three features corresponding to these three rows, namely Y1, Y2, and Y3. Only these features Y1, Y2, and Y3 are retained in the second feature sequence, resulting in the third feature sequence H1” as (Y1, Y2, Y3). Next, the Euclidean distance between features Y1, Y2, and Y3 is calculated, resulting in the third adjacency matrix A1” of X1. Among them, Y 11 This represents the connection relationship (i.e., Euclidean distance) between the first feature Y1 and the first feature Y2. 12 Y represents the Euclidean distance between the first feature Y1 and the second feature Y2. 13 This represents the Euclidean distance between the first feature Y1 and the third feature Y3, and so on.

[0088] Similarly, if the three features corresponding to the key information of X2 are determined to be Y1', Y2', and Y3', then the third adjacency matrix A2" of X2 can be obtained as follows: The third feature sequence H2” of X2 is (Y1', Y2', Y3'). If the three features corresponding to the key information of X3 are determined to be Y1”, Y2”, and Y3”, then the third adjacency matrix A3” of X3 can be obtained as follows: The third feature sequence H3” of X3 is (Y1”, Y2”, Y3”). Further, the graph structure of the completed X1 is G1=(A1”, H1”), the graph structure of the completed X2 is G2=(A2”, H2”), and the graph structure of the completed X3 is G3=(A3”, H3”).

[0089] Figure 5 This is a schematic diagram of the graph structure of the vehicle speed data for each section provided in the embodiments of this application. For example... Figure 5 As shown, the vehicle speed data is divided into three segments: the first segment contains speed data from 0 to 10 seconds, the second segment from 10 to 10 seconds, and the third segment from 20 to 30 seconds. Euclidean distance is calculated for each segment, and sparsity processing and key information extraction are performed to obtain the constructed graph structure G1 = (A1”, H1”), G2 = (A2”, H2”), and G3 = (A3”, H3”).

[0090] Figure 6 This is a flowchart illustrating a method provided in this application for concatenating adjacency matrices of N interval vehicle speed data to obtain an adjacency matrix of historical vehicle speed data. For example... Figure 6 As shown, the method includes the following steps:

[0091] In step S601, the adjacency matrices of the first to Nth interval vehicle speed data are concatenated sequentially in the row direction to obtain the first q rows of the adjacency matrix of the historical vehicle speed data.

[0092] In step S602, the first q rows of the adjacency matrix of the historical vehicle speed data are copied N-1 times, and the N-1 copies of the data are concatenated sequentially in the column direction of the historical vehicle speed data to obtain the adjacency matrix of the historical vehicle speed data.

[0093] In this embodiment of the application, when concatenating the adjacency matrices of N interval vehicle speed data, the adjacency matrices of the first to Nth interval vehicle speed data can be concatenated sequentially in the row direction to obtain the first q rows of the adjacency matrix of the historical vehicle speed data. Then, the first q rows of the adjacency matrix of the historical vehicle speed data are copied N-1 times, and the N-1 copies of the data are concatenated sequentially in the column direction of the historical vehicle speed data to obtain the adjacency matrix of the historical vehicle speed data.

[0094] Taking the concatenation of the third adjacency matrix of the interval vehicle speed data X1, X2, and X3 as an example, the resulting adjacency matrix of the historical data can be: Furthermore, by concatenating the third feature sequences of X1, X2, and X3, the feature sequence of the historical vehicle speed data can be (Y1, Y2, Y3, Y1', Y2', Y3', Y1", Y2", Y3").

[0095] Figure 7 This is a flowchart illustrating a method for performing time decay processing on the adjacency matrix of historical vehicle speed data to obtain an updated adjacency matrix of historical vehicle speed data, as provided in an embodiment of this application. Figure 7 As shown, the method includes the following steps:

[0096] In step S701, an empty matrix consisting of N*q rows and N*q columns is constructed.

[0097] Here, * represents the multiplication operator.

[0098] In step S702, N*N submatrices are determined in the empty matrix.

[0099] Each submatrix consists of the elements of row a*q+1 to row (a+1)*q and column a*q+1 to column (a+1)*q of the empty matrix, where a is an integer greater than or equal to 0 and less than N.

[0100] In step S703, it is determined that the time interval of each matrix element in the submatrix of row i3 and column j3 is the product of the difference between i3 and j3 and the time length of the interval vehicle speed data, and the value of each matrix element is determined to be the natural exponential value of the product of the attenuation rate and the time interval.

[0101] Where i3 and j3 are positive integers greater than 0 and less than or equal to N.

[0102] In step S704, the values ​​of each matrix element are assigned to an empty matrix to obtain the time decay factor matrix.

[0103] In step S705, the adjacency matrix of the historical vehicle speed data is multiplied by the time decay factor matrix to obtain the updated adjacency matrix of the historical vehicle speed data.

[0104] In this embodiment, a time decay factor matrix can be constructed first, and then the adjacency matrix of the historical vehicle speed data can be multiplied by the time decay factor matrix to obtain the updated adjacency matrix of the historical vehicle speed data after the adjacency matrix has undergone time decay processing. The time decay factor matrix can be constructed in the following manner:

[0105] First, construct an empty matrix with N*q rows and N*q columns. Then, determine N*N submatrices within the empty matrix, where each submatrix consists of elements from row (a*q+1) to row (a+1)*q and column (a+1)*q of the empty matrix. Next, determine that the time interval of each element in the submatrix at row i3 and column j3 is the product of the difference between i3 and j3 and the time length of the interval vehicle speed data, and determine that the value of each element is the natural exponent of the product of the decay rate and the time interval. Finally, assign the values ​​of each element to the empty matrix to obtain the time decay factor matrix.

[0106] To the above Figure 6 Taking the 9*9 adjacency matrix of historical data generated in the illustrated embodiment as an example, when constructing the time decay factor matrix corresponding to this adjacency matrix, a 9*9 empty matrix can be constructed first. This 9*9 empty matrix includes 9 sub-matrices, each of which is a 3*3 matrix. These 9 sub-matrices are denoted as A. 11 A 12 A 33 The empty matrix can be obtained as follows: Next, we can calculate the value of each element in the empty matrix. The attenuation factor of each element can be calculated using the following formula: Attenuation factor = exp(-λ*Δt), where λ is the attenuation rate and Δt is the time interval.

[0107] In the first submatrix A 11In this context, the element value can be the attenuation factor between the first segment speed data X1 and the second segment speed data X2. Since the time interval between X1 and X2 is 0, therefore A 11 The element with a value of 1 in the second submatrix A is 1. 12 In this context, the element value can be the attenuation factor between the first segment's speed data X1 and the second segment's speed data X2. Since the time interval between X1 and X2 is 10, therefore A 12 The element value is exp(-λ*10); in the third submatrix A 13 In this context, the element value can be the attenuation factor between the first segment speed data X1 and the third segment speed data X3. Since the time interval between X1 and X3 is 20, therefore A 13 The element value is exp(-λ*20).

[0108] In the fourth submatrix A 21 In this context, the element value can be the attenuation factor between the second segment speed data X2 and the first segment speed data X1. Since the time interval between X2 and X1 is 10, therefore A 21 The element value in the fifth submatrix A is exp(-λ*10); 22 In this context, the element value can be the attenuation factor between the second segment speed data X2 and the second segment speed data X2. Since the time interval between X2 and X2 is 0, therefore A 22 The element value is 1; in the sixth submatrix A 23 In this context, the element value can be the attenuation factor between the second segment speed data X2 and the third segment speed data X3. Since the time interval between X2 and X3 is 10, therefore A 23 The element value is exp(-λ*10).

[0109] In the seventh submatrix A 31 In this context, the element value can be the attenuation factor between the third segment speed data X3 and the first segment speed data X1. Since the time interval between X3 and X1 is 20, therefore A 31 The element value in the eighth submatrix A is exp(-λ*20); 32 In this context, the element value can be the attenuation factor between the third segment speed data X3 and the second segment speed data X2. Since the time interval between X3 and X2 is 10, therefore A 32 The element value is exp(-λ*10); in the ninth submatrix A 33 In this context, the element value can be the attenuation factor between the third segment speed data X3 and the third segment speed data X3. Since the time interval between X3 and X3 is 0, therefore A 33 The element value is 1.

[0110] Figure 8 This is a schematic diagram of the adjacency matrix and time decay factor matrix of historical vehicle speed data provided in an embodiment of this application. Figure 8 As shown, the adjacency matrix of historical vehicle speed data can be divided into 9 regions, each region corresponding to a submatrix A in the time decay factor matrix. 11 A 12 A 13 A 21 A 22 A 23 A 31 A 32 A 33 In the time decay factor matrix, submatrix A 11 The element with a value of 1 in submatrix A 12 The element value in the matrix is ​​exp(-λ*10), and the submatrix A is... 13 The element value in the matrix is ​​exp(-λ*20), and the submatrix A is... 21 The element value in the matrix is ​​exp(-λ*10), and the submatrix A is... 22 The element with a value of 1 in submatrix A 23 The element value in the matrix is ​​exp(-λ*10), and the submatrix A is... 31 The element value in the matrix is ​​exp(-λ*20), and the submatrix A is... 32 The element value in the matrix is ​​exp(-λ*10), and the submatrix A is... 33 The value of the element in is 1.

[0111] In this embodiment, the trained first graph convolutional network is used to aggregate the adjacency matrix and feature sequence of the updated historical vehicle speed data to obtain the feature sequence of the updated historical vehicle speed data. This can be achieved by performing a symmetric normalization operation on the adjacency matrix of the updated historical vehicle speed data, and then using the activation function of the trained first graph convolutional network to perform activation calculation on the product of the symmetric normalized adjacency matrix, the feature sequence of the historical vehicle speed data, and the preset weight matrix to obtain the feature sequence of the updated historical vehicle speed data.

[0112] That is, if the feature sequence of the spliced ​​historical vehicle speed data is H all Let A denote the adjacency matrix of the updated historical vehicle speed data. all Then it can be done through the formula The updated historical vehicle speed data feature sequence is calculated, where W all A preset weight matrix for historical vehicle speed data.

[0113] The technical solution of this application first selects vehicle speed data under different driving conditions, divides the speed data into time periods, and extracts the spatial information of vehicle speed within each time period through a graph convolutional network. For each time period, a graph is constructed to represent the spatial relationship of the speed data, and the spatial feature information of vehicle speed within each time period is deeply mined. Then, a time-factor decay graph convolutional module is designed, introducing a time decay factor into the graph convolution of each time period, so that the model can pay more attention to the influence of recent speed data and correspondingly reduce the weight of past data, thereby improving the accuracy and real-time performance of speed prediction.

[0114] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0115] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0116] Figure 9 This is a schematic diagram of a vehicle speed prediction device provided in an embodiment of this application. Figure 9 As shown, the device includes:

[0117] The acquisition module 901 is configured to acquire the vehicle's historical speed data for the current trip, and the historical speed data is time series data.

[0118] The segmentation module 902 is configured to divide the historical vehicle speed data into N interval vehicle speed data with the same time length. The time length of the interval vehicle speed data is the difference between the time corresponding to the last data and the time corresponding to the first data in the interval vehicle speed data, and N is a positive integer greater than 1.

[0119] Graph construction module 903 is configured to construct graph structures for each interval of vehicle speed data, wherein the graph structure for each interval of vehicle speed data includes an adjacency matrix and feature sequences.

[0120] The splicing module 904 is configured to splice the adjacency matrices of N interval vehicle speed data to obtain the adjacency matrix of historical vehicle speed data, and to splice the feature sequences of N interval vehicle speed data to obtain the feature sequences of historical vehicle speed data.

[0121] The attenuation module 905 is configured to perform time attenuation processing on the adjacency matrix of historical vehicle speed data to obtain the updated adjacency matrix of historical vehicle speed data.

[0122] The aggregation module 906 is configured to use the trained first graph convolutional network to aggregate the adjacency matrix of the updated historical vehicle speed data and the feature sequence of the historical vehicle speed data to obtain the feature sequence of the updated historical vehicle speed data.

[0123] The prediction module 907 is configured to input the feature sequence of the updated historical vehicle speed data into the trained prediction unit to obtain the vehicle speed prediction result.

[0124] According to the technical solution provided in the embodiments of this application, the time series data of historical vehicle speed is divided into intervals. The vehicle speed data of each interval is then used to construct a graph to obtain a feature sequence including an adjacency matrix. The adjacency matrices of the vehicle speed data of each interval are then concatenated to obtain the adjacency matrix of the historical vehicle speed data. The feature sequences of the vehicle speed data of each interval are then concatenated to obtain the feature sequence of the historical vehicle speed data. After time decay processing is applied to the adjacency matrix of the historical vehicle speed data, a trained graph convolutional network is used to aggregate the time decayed adjacency matrix and the feature sequence of the historical vehicle speed data. Finally, the aggregation result is input into a trained prediction unit to obtain the vehicle speed prediction result. This method can deeply mine the spatial feature information of vehicle speed in each time period in the historical vehicle speed data and increase the weight of recent vehicle speed data when making vehicle speed prediction, thereby improving the accuracy and real-time performance of vehicle speed prediction and enhancing the user experience.

[0125] In this embodiment, constructing a graph structure for interval vehicle speed data includes: using a trained feature extraction network to extract features from the interval vehicle speed data to obtain a first feature sequence of the interval vehicle speed data; determining the connection relationships between each feature in the first feature sequence and other features; and using the connection relationships between the features to construct a first adjacency matrix of the interval vehicle speed data. The first adjacency matrix includes p rows and p columns, and the matrix element in the i1th row and j1th column represents the connection relationship between the i1th feature and the j1th feature in the first feature sequence, where p is a positive integer greater than 1, and i1 and j1 are both positive integers greater than or equal to 1 and less than p. The process involves: 1) Sparse processing the first adjacency matrix to obtain the second adjacency matrix; 2) Aggregating the second adjacency matrix and the first feature sequence using a trained second graph convolutional network to obtain the second feature sequence of the interval speed data; 3) Extracting key information from the second adjacency matrix to obtain the third adjacency matrix, which consists of q rows and q columns, where q is a positive integer greater than 1 and less than p; 4) Determining the target features corresponding to each row in the third adjacency matrix, retaining only the target features in the second feature sequence to obtain the third feature sequence; 5) Determining that the graph structure formed by the third adjacency matrix and the third feature sequence is the graph structure of the interval speed data.

[0126] In this embodiment of the application, the first adjacency matrix is ​​sparsely processed to obtain the second adjacency matrix, including: determining the first k matrix elements with the smallest values ​​in each row of the first adjacency matrix as the first target elements, where k is a positive integer greater than 1 and less than p; setting the values ​​of other matrix elements in the first adjacency matrix, except for the first target elements, to 0, to obtain the second adjacency matrix.

[0127] In this embodiment, key information is extracted from the second adjacency matrix to obtain the third adjacency matrix. This includes: summing the values ​​of all elements in each row of the second adjacency matrix to obtain the row element sum of the second adjacency matrix; determining the rows corresponding to the q smallest row element sums; determining the q target features corresponding to the target rows; determining the connection relationship between each of the q target features and other target features; and constructing the third adjacency matrix using the connection relationships between the target features. The matrix element in the i2th row and j2nd column of the third adjacency matrix represents the connection relationship between the i2th feature and the j2nd feature in the target features, where i2 and j2 are both positive integers greater than or equal to 1 and less than q.

[0128] In this embodiment of the application, the adjacency matrices of N interval vehicle speed data are concatenated to obtain the adjacency matrix of historical vehicle speed data. This includes: concatenating the adjacency matrices of the first to Nth interval vehicle speed data in the row direction to obtain the first q rows of the adjacency matrix of historical vehicle speed data; copying the first q rows of the adjacency matrix of historical vehicle speed data N-1 times, and concatenating the N-1 copies of data in the column direction of historical vehicle speed data to obtain the adjacency matrix of historical vehicle speed data.

[0129] In this embodiment, time decay processing is performed on the adjacency matrix of historical vehicle speed data to obtain an updated adjacency matrix of historical vehicle speed data. This includes: constructing an empty matrix with N*q rows and N*q columns, where * is the multiplication operator; determining N*N submatrices in the empty matrix, each submatric consisting of matrix elements from row (a*q+1) to row (a+1)*q and column (a*q+1) to column (a+1)*q) of the empty matrix, where a is an integer greater than or equal to 0 and less than N; determining that the time interval of each matrix element in the submatrix at row i3 and column j3 is the product of the difference between i3 and j3 and the time length of the interval vehicle speed data, and determining that the value of each matrix element is the natural exponent of the product of the decay rate and the time interval, where i3 and j3 are both positive integers greater than 0 and less than or equal to N; assigning the values ​​of each matrix element to the empty matrix to obtain a time decay factor matrix; and multiplying the adjacency matrix of historical vehicle speed data by the time decay factor matrix to obtain the updated adjacency matrix of historical vehicle speed data.

[0130] In this embodiment, a trained first graph convolutional network is used to aggregate the adjacency matrix of the updated historical vehicle speed data and the feature sequence of the historical vehicle speed data to obtain the feature sequence of the updated historical vehicle speed data. This includes: performing a symmetric normalization operation on the adjacency matrix of the updated historical vehicle speed data; and using the activation function of the trained first graph convolutional network to perform activation calculation on the product of the symmetric normalized adjacency matrix, the feature sequence of the historical vehicle speed data, and the preset weight matrix to obtain the feature sequence of the updated historical vehicle speed data.

[0131] It should be understood that the sequence number of each step in the above embodiments does not imply 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 this application.

[0132] Figure 10 This is a schematic diagram of the electronic device provided in an embodiment of this application. For example... Figure 10 As shown, the electronic device 10 of this embodiment includes: a processor 1001, a memory 1002, and a computer program 1003 stored in the memory 1002 and executable on the processor 1001. When the processor 1001 executes the computer program 1003, it implements the steps in the various method embodiments described above. Alternatively, when the processor 1001 executes the computer program 1003, it implements the functions of each module / unit in the various device embodiments described above.

[0133] Electronic device 10 may be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 10 may include, but is not limited to, a processor 1001 and a memory 1002. Those skilled in the art will understand that... Figure 10 This is merely an example of electronic device 10 and does not constitute a limitation on electronic device 10. It may include more or fewer components than shown, or different components.

[0134] The processor 1001 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0135] The memory 1002 can be an internal storage unit of the electronic device 10, such as a hard disk or RAM of the electronic device 10. The memory 1002 can also be an external storage device of the electronic device 10, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, FlashCard, etc., equipped on the electronic device 10. The memory 1002 can also include both internal and external storage units of the electronic device 10. The memory 1002 is used to store computer programs and other programs and data required by the electronic device.

[0136] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to 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 embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0137] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0138] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A vehicle speed prediction method, characterized in that, include: Obtain the vehicle's historical speed data for this current trip, wherein the historical speed data is time-series data; The historical vehicle speed data is divided into N interval vehicle speed data with the same time length. The time length of the interval vehicle speed data is the difference between the time corresponding to the last data and the time corresponding to the first data in the interval vehicle speed data, and N is a positive integer greater than 1. Construct graph structures for the vehicle speed data of each interval, wherein the graph structure of each interval includes an adjacency matrix and a feature sequence; The adjacency matrices of the N interval speed data are concatenated to obtain the adjacency matrix of the historical speed data. The feature sequences of the N interval speed data are concatenated to obtain the feature sequence of the historical speed data. The adjacency matrix of the historical vehicle speed data is subjected to time decay processing to obtain an updated adjacency matrix of the historical vehicle speed data. The time decay processing includes: constructing a time decay factor matrix, multiplying the adjacency matrix of the historical vehicle speed data with the time decay factor matrix to obtain an updated adjacency matrix of the historical vehicle speed data after the adjacency matrix has been subjected to time decay processing, wherein the matrix element value of the time decay factor matrix is ​​the natural exponent value of the product of the decay rate and the time interval. The pre-trained first graph convolutional network is used to aggregate the adjacency matrix of the updated historical vehicle speed data and the feature sequence of the historical vehicle speed data to obtain the feature sequence of the updated historical vehicle speed data. The feature sequence of the updated historical vehicle speed data is input into the trained prediction unit to obtain the vehicle speed prediction result.

2. The method according to claim 1, characterized in that, Construct a graph structure for the interval vehicle speed data, including: The trained feature extraction network is used to extract features from the interval vehicle speed data to obtain the first feature sequence of the interval vehicle speed data. The connection relationships between each feature in the first feature sequence and other features are determined, and the connection relationships between each feature are used to construct the first adjacency matrix of the interval vehicle speed data. The first adjacency matrix includes p rows and p columns. The matrix element in the i1th row and j1th column represents the connection relationship between the i1th feature and the j1st feature in the first feature sequence, where p is a positive integer greater than 1, and i1 and j1 are both positive integers greater than or equal to 1 and less than p. The first adjacency matrix is ​​sparsed to obtain the second adjacency matrix; The trained second graph convolutional network is used to aggregate the second adjacency matrix and the first feature sequence to obtain the second feature sequence of the interval vehicle speed data; Key information is extracted from the second adjacency matrix to obtain a third adjacency matrix, which consists of q rows and q columns, where q is a positive integer greater than 1 and less than p; Determine the target features corresponding to each row in the third adjacency matrix, and retain only the target features in the second feature sequence to obtain the third feature sequence; The graph structure formed by the third adjacency matrix and the third feature sequence is determined to be the graph structure of the interval vehicle speed data.

3. The method according to claim 2, characterized in that, The step of sparse processing the first adjacency matrix to obtain the second adjacency matrix includes: In the first adjacency matrix, the k smallest matrix elements in each row are determined as the first target elements, where k is a positive integer greater than 1 and less than p; Set the values ​​of all matrix elements in the first adjacency matrix except for the first target element to 0 to obtain the second adjacency matrix.

4. The method according to claim 2, characterized in that, The step of extracting key information from the second adjacency matrix to obtain the third adjacency matrix includes: The sum of all the values ​​of each element in each row of the second adjacency matrix is ​​obtained to obtain the sum of the row elements of the second adjacency matrix; Determine the q smallest row elements and their corresponding rows as the target rows; Determine q target features corresponding to the target row; The connection relationship between each of the q target features and other target features is determined, and the connection relationship between each target feature is used to construct the third adjacency matrix. The matrix element in the i2th row and j2nd column of the third adjacency matrix represents the connection relationship between the i2th feature and the j2nd feature in the target features, where i2 and j2 are both positive integers greater than or equal to 1 and less than q.

5. The method according to claim 1, characterized in that, The step of concatenating the adjacency matrices of the N interval vehicle speed data to obtain the adjacency matrix of historical vehicle speed data includes: The adjacency matrices of the first to Nth interval vehicle speed data are concatenated sequentially in the row direction to obtain the first q rows of the adjacency matrix of the historical vehicle speed data; The first q rows of the adjacency matrix of the historical vehicle speed data are copied N-1 times, and the N-1 copies of the data are concatenated sequentially in the column direction of the historical vehicle speed data to obtain the adjacency matrix of the historical vehicle speed data.

6. The method according to claim 5, characterized in that, The step of performing time decay processing on the adjacency matrix of the historical vehicle speed data to obtain the updated adjacency matrix of the historical vehicle speed data includes: Construct an empty matrix with N*q rows and N*q columns, where * is the multiplication operator; In the empty matrix, determine N*N submatrices. Each submatrix consists of matrix elements from row a*q+1 to row (a+1)*q and column a*q+1 to column (a+1)*q in the empty matrix, where a is an integer greater than or equal to 0 and less than N. In the submatrix of row i3 and column j3, the time interval of each matrix element is determined to be the product of the difference between i3 and j3 and the time length of the interval vehicle speed data. The value of each matrix element is determined to be the natural exponent of the product of the attenuation rate and the time interval, where i3 and j3 are positive integers greater than 0 and less than or equal to N. The values ​​of each matrix element are assigned to the empty matrix to obtain the time decay factor matrix; Multiply the adjacency matrix of the historical vehicle speed data by the time decay factor matrix to obtain the updated adjacency matrix of the historical vehicle speed data.

7. The method according to claim 1, characterized in that, The pre-trained first graph convolutional network is used to aggregate the adjacency matrix of the updated historical vehicle speed data and the feature sequence of the historical vehicle speed data to obtain the feature sequence of the updated historical vehicle speed data, including: Perform symmetric normalization on the adjacency matrix of the updated historical vehicle speed data; Using the activation function of the trained first graph convolutional network, the activation calculation is performed on the product of the symmetric normalized adjacency matrix, the feature sequence of the historical vehicle speed data, and the preset weight matrix to obtain the updated feature sequence of the historical vehicle speed data.

8. A vehicle speed prediction device, characterized in that, include: The acquisition module is configured to acquire the vehicle's historical speed data for the current trip, wherein the historical speed data is time-series data; The segmentation module is configured to divide the historical vehicle speed data into N interval vehicle speed data with the same time length, wherein the time length of the interval vehicle speed data is the difference between the time corresponding to the last data and the time corresponding to the first data in the interval vehicle speed data, and N is a positive integer greater than 1. The graph construction module is configured to construct the graph structure of the vehicle speed data for each interval, wherein the graph structure of the vehicle speed data for each interval includes an adjacency matrix and a feature sequence. The splicing module is configured to splice the adjacency matrices of the N interval vehicle speed data to obtain the adjacency matrix of the historical vehicle speed data, and to splice the feature sequences of the N interval vehicle speed data to obtain the feature sequences of the historical vehicle speed data. The attenuation module is configured to perform time attenuation processing on the adjacency matrix of the historical vehicle speed data to obtain an updated adjacency matrix of the historical vehicle speed data. The time attenuation processing includes: constructing a time attenuation factor matrix, multiplying the adjacency matrix of the historical vehicle speed data with the time attenuation factor matrix to obtain an updated adjacency matrix of the historical vehicle speed data after time attenuation processing, wherein each element of the time attenuation factor matrix is ​​the natural exponent of the product of the attenuation rate and the time interval. The aggregation module is configured to use a trained first graph convolutional network to aggregate the adjacency matrix of the updated historical vehicle speed data and the feature sequence of the historical vehicle speed data to obtain the feature sequence of the updated historical vehicle speed data. The prediction module is configured to input the feature sequence of the updated historical vehicle speed data into the trained prediction unit to obtain the vehicle speed prediction result.

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, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images