Vehicle speed prediction method and device and electronic equipment

Abstract

The invention provides a vehicle speed prediction method and device and electronic equipment, and the method comprises the steps: constructing a historical vehicle speed diagram according to historical vehicle speed information, and constructing a driving condition characteristic parameter diagram according to vehicle driving condition characteristic parameters; according to a pre-constructed adaptive redundant feature discarding module, removing corresponding first redundant features and second redundant features in the historical vehicle speed map and the driving condition feature parameter map to obtain target historical vehicle speed features and target driving condition feature parameter features; determining a loss value corresponding to each node according to the target historical vehicle speed characteristic and the target driving condition characteristic parameter characteristic; determining a target attention coefficient according to the loss value; and outputting the predicted vehicle speed through a preset neural network structure according to the target attention coefficient. According to the invention, the prediction accuracy of the whole node is improved, and the real-time accurate prediction of the future driving condition of the extended-range automobile under the small sample condition is realized.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of automobile application technology, and in particular to a vehicle speed prediction method, device and electronic equipment. Background Art

[0002] The predictive energy management strategy uses advanced prediction technology to predict energy demand and optimal energy distribution plans in the future based on driving conditions, vehicle status and other factors. Driving conditions are a type of time series information that changes over time. Unknown conditions will lead to large differences in vehicle fuel economy. Therefore, the future driving conditions of the vehicle can be predicted from a time perspective to plan the vehicle's power and power distribution in advance and improve overall fuel economy.

[0003] In order to predict the future driving conditions of the vehicle, short-term vehicle speed prediction can be used. In related technologies, short-term vehicle speed prediction mainly includes model-based and data-driven prediction methods. The model-based method completes the vehicle speed information prediction through the established model with certain theoretical assumptions, but is limited by the modeling complexity, limited accuracy and model generalization ability. The data-driven method mainly relies on information such as the vehicle's historical speed to predict the future short-term vehicle speed, and does not rely on detailed physical models and complex modeling processes. However, the existing data-driven method only uses a simple network model to predict the vehicle speed, which cannot capture the implicit deep relationship between nonlinear vehicle speed data, so that the prediction accuracy cannot be further improved. Summary of the invention

[0004] The embodiments of the present application provide a vehicle speed prediction method, device, system, electronic device and computer-readable storage medium to solve the problems in the related art in terms of automatic identification and removal of invalid black edges and realization of intelligent color adjustment, which are insufficient in real time and accuracy.

[0005] In a first aspect, an embodiment of the present application provides a vehicle speed prediction method, the method comprising:

[0006] Constructing a historical vehicle speed map based on historical vehicle speed information, and constructing a driving condition characteristic parameter map based on vehicle driving condition characteristic parameters;

[0007] According to a pre-built adaptive redundant feature discarding module, the first redundant feature and the second redundant feature corresponding to the historical vehicle speed map and the driving condition feature parameter map are removed respectively to obtain a target historical vehicle speed feature and a target driving condition feature parameter feature;

[0008] Determine the loss value corresponding to each node according to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics;

[0009] Determine the target attention coefficient corresponding to each of the nodes according to the loss value;

[0010] The target predicted speed of the target vehicle is output through the preset neural network structure according to the real-time speed information of the target vehicle and the target attention coefficient corresponding to each of the nodes.

[0011] Optionally, determining the loss value corresponding to each node according to the target historical vehicle speed feature and the target driving condition feature parameter feature includes:

[0012] All nodes corresponding to the target historical vehicle speed feature and the target driving condition feature parameter feature are output through a preset neural network structure to predict the vehicle speed corresponding to each node;

[0013] The loss value is obtained according to the predicted vehicle speed and the actual vehicle speed corresponding to each of the nodes obtained in advance.

[0014] Optionally, before the step of constructing a driving condition characteristic parameter map according to the vehicle driving condition characteristic parameters, the method includes:

[0015] Generating the vehicle driving condition characteristic parameters according to the pre-acquired historical vehicle speed information;

[0016] Among them, the vehicle driving condition characteristic parameters are generated through the historical vehicle speed information, and the vehicle driving condition characteristic parameters include at least one of the average vehicle speed, the average driving speed, the vehicle speed variance, the acceleration variance, the inertia energy variance, the acceleration time ratio and the deceleration time ratio.

[0017] Optionally, constructing a historical vehicle speed map according to historical vehicle speed information includes:

[0018] The historical vehicle speed information is graph-constructed based on cosine similarity to obtain the historical vehicle speed graph, wherein the historical vehicle speed graph includes the historical vehicle speed information and a first adjacency matrix.

[0019] Optionally, constructing a driving condition characteristic parameter map according to the vehicle driving condition characteristic parameters includes:

[0020] Obtaining a Euclidean distance matrix according to the Euclidean distances between the vehicle driving condition characteristic parameters;

[0021] A driving condition characteristic parameter graph is constructed according to the vehicle driving condition characteristic parameters and the Euclidean distance matrix, wherein the Euclidean distance matrix is ​​a second adjacency matrix corresponding to the driving condition characteristic parameter graph.

[0022] Optionally, the pre-constructed adaptive redundant feature discarding module removes the first redundant feature and the second redundant feature respectively corresponding to the historical vehicle speed map and the driving condition feature parameter map to obtain the target historical vehicle speed feature and the target driving condition feature parameter feature, including:

[0023] Inputting the historical vehicle speed graph and the driving condition characteristic parameter graph into dual-channel graph convolution to perform parallel feature extraction, thereby obtaining historical vehicle speed aggregation features and driving condition characteristic parameter aggregation features;

[0024] According to the historical vehicle speed aggregation feature and the driving condition feature parameter aggregation feature, each node feature in the historical vehicle speed graph and the driving condition feature parameter graph is respectively subjected to a dimensionality reduction operation, and a first attention coefficient and a second attention coefficient are obtained through a preset function;

[0025] Multiplying the first attention coefficient and the first adjacency matrix to obtain a first target adjacency matrix, and multiplying the second attention coefficient and the second adjacency matrix to obtain a second target adjacency matrix;

[0026] Screening the first target adjacency matrix and the second target adjacency matrix according to a first target threshold value to obtain a first screened adjacency matrix and a second screened adjacency matrix;

[0027] Performing a graph convolution operation on the first screening adjacency matrix and the second screening adjacency matrix to obtain a first feature and a second feature, and extracting aggregate features of the historical vehicle speed aggregate feature and the driving condition feature parameter aggregate feature through the first adjacency matrix and the second adjacency matrix to obtain a first aggregate feature and a second aggregate feature;

[0028] Combining the first feature with the first aggregate feature to obtain a first combined feature, and combining the second feature with the second aggregate feature to obtain a second combined feature;

[0029] Input the first combined features and the second combined features into a pre-built adaptive redundant feature discarding module to obtain the Euclidean distance between each pair of nodes corresponding to the first combined features and the Euclidean distance between each pair of nodes corresponding to the second combined features;

[0030] Determine a first redundant feature, a second redundant feature, and a redundant node corresponding to the first redundant feature and a redundant node corresponding to the second redundant feature according to the Euclidean distance;

[0031] The redundant nodes corresponding to the first redundant feature and the redundant nodes corresponding to the second redundant feature are removed to obtain the target historical vehicle speed feature and the target driving condition feature parameter feature.

[0032] Optionally, the determining of the first redundant feature and the second redundant feature according to the Euclidean distance, and the redundant nodes corresponding to the first redundant feature and the redundant nodes corresponding to the second redundant feature include:

[0033] When the Euclidean distance between each pair of nodes is greater than the second target threshold, each pair of nodes is used as a redundant node, and the node features corresponding to each pair of nodes are used as redundant features.

[0034] Optionally, determining a target attention coefficient corresponding to each of the nodes according to the loss value includes:

[0035] Adjusting the loss value range to a preset range through a preset function, wherein the preset range is from an initial value to a target value;

[0036] When the loss value corresponding to the node is within a preset range and is less than a first preset value, the attention coefficient corresponding to the node is the target value minus the loss value;

[0037] When the loss value corresponding to the node is within a preset range, greater than the first preset value and less than the second preset value, the attention coefficient corresponding to the node is the loss value;

[0038] When the loss value corresponding to the node is within a preset range and is greater than the second preset value, the attention coefficient corresponding to the node is the initial value;

[0039] Wherein, the second preset value is greater than the first preset value.

[0040] In a second aspect, an embodiment of the present application provides a vehicle speed prediction device, the device comprising:

[0041] A construction module, used to construct a historical vehicle speed map according to historical vehicle speed information, and to construct a driving condition characteristic parameter map according to vehicle driving condition characteristic parameters, wherein the vehicle driving condition characteristic parameters are generated by the historical vehicle speed information;

[0042] A feature module, used for removing the first redundant feature and the second redundant feature respectively corresponding to the historical vehicle speed map and the driving condition feature parameter map according to a pre-built adaptive redundant feature discarding module, to obtain a target historical vehicle speed feature and a target driving condition feature parameter feature;

[0043] A loss value acquisition module, used to determine the loss value corresponding to each node according to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics;

[0044] A target attention coefficient determination module, used to determine the target attention coefficient corresponding to each of the nodes according to the loss value;

[0045] The output module is used to output the target predicted speed of the target vehicle through the preset neural network structure according to the real-time speed information of the target vehicle and the target attention coefficient corresponding to each node.

[0046] Optionally, the loss value acquisition module includes:

[0047] A first loss value acquisition submodule is used to output the predicted vehicle speed corresponding to each node through a preset neural network structure for all nodes corresponding to the target historical vehicle speed feature and the target driving condition feature parameter feature;

[0048] The second loss value acquisition submodule is used to obtain the loss value according to the predicted vehicle speed and the actual vehicle speed corresponding to each of the nodes acquired in advance.

[0049] Optionally, the device further comprises:

[0050] A generation module is used to generate the vehicle driving condition characteristic parameters according to the pre-acquired historical vehicle speed information; wherein the vehicle driving condition characteristic parameters are generated by the historical vehicle speed information, and the vehicle driving condition characteristic parameters include at least one of the average vehicle speed, the average driving speed, the vehicle speed variance, the acceleration variance, the inertia energy variance, the acceleration time ratio and the deceleration time ratio.

[0051] Optionally, the building blocks include:

[0052] The first construction submodule is used to construct a graph of the historical vehicle speed information based on cosine similarity to obtain the historical vehicle speed graph, wherein the historical vehicle speed graph includes the historical vehicle speed information and a first adjacency matrix.

[0053] Optionally, the building blocks include:

[0054] A second construction submodule is used to obtain a Euclidean distance matrix according to the Euclidean distances between the vehicle driving condition characteristic parameters;

[0055] The third construction submodule is used to construct a driving condition characteristic parameter map according to the vehicle driving condition characteristic parameters and the Euclidean distance matrix, wherein the Euclidean distance matrix is ​​a second adjacency matrix corresponding to the driving condition characteristic parameter map.

[0056] Optionally, the feature module includes:

[0057] A first feature submodule is used to input the historical vehicle speed graph and the driving condition characteristic parameter graph into a dual-channel graph convolution to perform parallel feature extraction, so as to obtain historical vehicle speed aggregation features and driving condition characteristic parameter aggregation features;

[0058] The second feature submodule is used to perform a dimensionality reduction operation on each node feature in the historical vehicle speed graph and the driving condition feature parameter graph according to the historical vehicle speed aggregation feature and the driving condition feature parameter aggregation feature, and obtain a first attention coefficient and a second attention coefficient through a preset function;

[0059] A third feature submodule, configured to multiply the first attention coefficient and the first adjacency matrix to obtain a first target adjacency matrix, and to multiply the second attention coefficient and the second adjacency matrix to obtain a second target adjacency matrix;

[0060] A fourth feature submodule, configured to screen the first target adjacency matrix and the second target adjacency matrix according to a first target threshold value to obtain a first screened adjacency matrix and a second screened adjacency matrix;

[0061] a fifth feature submodule, configured to perform a graph convolution operation on the first screening adjacency matrix and the second screening adjacency matrix to obtain a first feature and a second feature, and extract aggregate features of the historical vehicle speed aggregate feature and the driving condition feature parameter aggregate feature through the first adjacency matrix and the second adjacency matrix to obtain a first aggregate feature and a second aggregate feature;

[0062] a sixth feature submodule, configured to combine the first feature with the first aggregate feature to obtain a first combined feature, and to combine the second feature with the second aggregate feature to obtain a second combined feature;

[0063] a seventh feature submodule, configured to input the first combined feature and the second combined feature into a pre-built adaptive redundant feature discarding module, and obtain the Euclidean distance between each pair of nodes corresponding to the first combined feature, and the Euclidean distance between each pair of nodes corresponding to the second combined feature;

[0064] an eighth feature submodule, configured to determine, according to the Euclidean distance, a first redundant feature, a second redundant feature, and a redundant node corresponding to the first redundant feature and a redundant node corresponding to the second redundant feature;

[0065] The ninth feature submodule is used to remove the redundant nodes corresponding to the first redundant feature and the redundant nodes corresponding to the second redundant feature to obtain the target historical vehicle speed feature and the target driving condition feature parameter feature.

[0066] Optionally, the eighth characteristic submodule includes:

[0067] The feature unit is used to treat each pair of nodes as redundant nodes and the node features corresponding to each pair of nodes as redundant features when the Euclidean distance between the nodes is greater than a second target threshold.

[0068] Optionally, the target attention coefficient determination module includes:

[0069] A first target attention coefficient determination submodule, used to adjust the loss value range to a preset range through a preset function, wherein the preset range is from an initial value to a target value;

[0070] A second target attention coefficient determination submodule, for when the loss value corresponding to the node is within a preset range and is less than a first preset value, the attention coefficient corresponding to the node is the target value minus the loss value;

[0071] A third target attention coefficient determination submodule, configured to determine the attention coefficient corresponding to the node as the loss value when the loss value corresponding to the node is within a preset range, greater than the first preset value, and less than a second preset value;

[0072] The fourth target attention coefficient determination submodule is used to determine that when the loss value corresponding to the node is within a preset range and is greater than the second preset value, the attention coefficient corresponding to the node is the initial value; wherein the second preset value is greater than the first preset value.

[0073] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a processor; and a memory for storing instructions executable by the processor, wherein the processor is configured to execute the instructions to implement any of the vehicle speed prediction methods described above.

[0074] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium. When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute any of the vehicle speed prediction methods described above.

[0075] In a fifth aspect, an embodiment of the present application further provides a vehicle, comprising the vehicle speed prediction device as described above.

[0076] In an embodiment of the present application, a historical vehicle speed graph is constructed based on historical vehicle speed information, and a driving condition characteristic parameter graph is constructed based on vehicle driving condition characteristic parameters; the first redundant features and the second redundant features corresponding to the historical vehicle speed graph and the driving condition characteristic parameter graph, respectively, are removed according to a pre-constructed adaptive redundant feature discarding module to obtain target historical vehicle speed features and target driving condition characteristic parameter features; the loss value corresponding to each node is determined according to the target historical vehicle speed features and the target driving condition characteristic parameter features; the target attention coefficient corresponding to each of the nodes is determined according to the loss value; and the target predicted speed of the target vehicle is output through the preset neural network structure according to the real-time vehicle speed information of the target vehicle and the target attention coefficient corresponding to each of the nodes. That is, the present application constructs a graph of characteristic parameters and historical vehicle speeds that characterize the driving conditions of a vehicle. Secondly, feature extraction is performed on the two generated graph structures in parallel, and redundant edges and redundant features are discarded through a pre-built adaptive redundant feature discarding module during the feature extraction process. This effectively avoids the problem of each node being connected to each other during the graph construction process, which leads to the introduction of a large number of redundant edges that affect the prediction performance of the graph convolutional network. It effectively extracts highly correlated features, removes redundant features, reduces the amount of parameters, and improves the calculation speed. Then, based on the predicted vehicle speed, the degree of correlation between each node and the prediction task is obtained, and reverse node feature enhancement is achieved according to the degree of correlation, and the target attention coefficient is updated, thereby improving the prediction accuracy of the overall node and realizing real-time and accurate prediction of the future driving conditions of extended-range vehicles under small sample conditions.

[0077] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0079] Figure 1 is a flowchart of the steps of a vehicle speed prediction method provided by an embodiment of the present application;

[0080] Figure 2 is a flowchart of another vehicle speed prediction method provided by an embodiment of the present application;

[0081] Figure 3 is a flowchart of the steps of another vehicle speed prediction method provided by the present application;

[0082] Figure 4 It is a logic block diagram of a vehicle speed prediction device provided in an embodiment of the present application;

[0083] Figure 5 is a structural diagram of an electronic device provided in an embodiment of the present application;

[0084] Figure 6 It is a schematic diagram of the architecture of a driving condition prediction model for an extended-range vehicle based on a reverse enhanced GCN provided in an embodiment of the present application;

[0085] Figure 7 It is a schematic diagram of a dual-channel graph structure construction process provided by an embodiment of the present application;

[0086] Figure 8 It is a schematic diagram of a dual-channel redundant edge discarding operation provided in an embodiment of the present application;

[0087] Fig. 9 It is a schematic diagram of the structure of an adaptive redundant feature discarding module provided in an embodiment of the present application;

[0088] Fig.10 It is a schematic diagram of a reverse enhancement module structure based on prediction results provided in an embodiment of the present application. DETAILED DESCRIPTION

[0089] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0090] Figure 1 is a flow chart of the steps of a vehicle speed prediction method provided by an embodiment of the present application, such as Figure 1 As shown, the method may include:

[0091] Step 101, constructing a historical vehicle speed map according to historical vehicle speed information, and constructing a driving condition characteristic parameter map according to vehicle driving condition characteristic parameters;

[0092] It should be noted that the overall plan refers to Figure 6 It can be seen that in the embodiment of the present application, the characteristic parameters characterizing the driving condition of the vehicle and the historical vehicle speed are respectively constructed.

[0093] Therefore, the first step is to collect a large number of samples of the relationship between vehicle speed and time, which provides a large amount of reliable training data sets for the model.

[0094] The historical speed is taken as a training sample according to the time window T = 10s, and several training samples are obtained. The history in the training sample is recorded as Vs = [v1, v2, ..., vn].

[0095] The characteristic parameters characterizing the vehicle driving condition in each window are obtained, including the average vehicle speed vm, the average driving speed vmd, the vehicle speed variance vvar, the acceleration variance avar, the inertia energy variance vavar, the acceleration time ratio Pacc, and the deceleration time ratio Pdec, denoted as Vj = [vm, vmd, vvar, avar, vavar, Pacc, Pdec].

[0096] Furthermore, constructing a historical vehicle speed map according to historical vehicle speed information includes:

[0097] The historical vehicle speed information is graph-constructed based on cosine similarity to obtain the historical vehicle speed graph, wherein the historical vehicle speed graph includes the historical vehicle speed information and a first adjacency matrix.

[0098] It should be noted that in the embodiment of the present application, the characteristic parameters representing the vehicle driving condition and the historical vehicle speed are respectively constructed, and the dual-channel graph structure construction process is as follows: Figure 7 As shown. First, input the historical speed Vs = [v1, v2, ..., vn], construct a graph of the historical speed information based on cosine similarity, and use cosine similarity as a metric to calculate the similarity between historical speed data. The historical speed data is regarded as a vector, and their similarity is measured by calculating the angle between them, thus obtaining an n×n matrix. The values ​​of the matrix are the values ​​of the similarity, reflecting the degree of similarity between the historical speed data points. The final historical speed graph can be expressed as the following formula 1:

[0099] G1=(X1,A1) (Formula 1)

[0100] Among them, in the above formula 1, X1 represents the historical vehicle speed data, and A1 is the adjacency matrix.

[0101] Furthermore, constructing a driving condition characteristic parameter map according to the vehicle driving condition characteristic parameters includes:

[0102] Generating the vehicle driving condition characteristic parameters according to the pre-acquired historical vehicle speed information;

[0103] Among them, the vehicle driving condition characteristic parameters are generated through the historical vehicle speed information, and the vehicle driving condition characteristic parameters include at least one of the average vehicle speed, the average driving speed, the vehicle speed variance, the acceleration variance, the inertia energy variance, the acceleration time ratio and the deceleration time ratio.

[0104] It should be noted that in the embodiment of the present application, a graph is constructed for the characteristic parameters Vj=[vm, vmd, vvar, avar, vavar, Pacc, Pdec] characterizing the driving condition of the vehicle, and the relationship measurement between the characteristic parameters is obtained by calculating the Euclidean distance between the characteristic parameters characterizing the driving condition of the vehicle. Euclidean distance is a common distance measurement method that can measure the similarity or difference between characteristic parameters. The Euclidean distance matrix E can be obtained by calculating the Euclidean distance between the characteristic parameters of the driving condition of the vehicle in each training sample.

[0105]

[0106] Among them, E is a symmetric matrix, and its element Ei j represents the Euclidean distance between nodes i and j. E is used as the adjacency matrix A2 of the characteristic parameters of the vehicle driving condition, so as to obtain the driving condition characteristic parameter diagram, which can be expressed as:

[0107] G2=(X2,A2) (Formula 3)

[0108] Wherein, X2 in Formula 3 is a characteristic parameter of the driving condition.

[0109] Step 102, removing the first redundant features and the second redundant features corresponding to the historical vehicle speed map and the driving condition characteristic parameter map respectively according to a pre-built adaptive redundant feature discarding module, to obtain a target historical vehicle speed feature and a target driving condition characteristic parameter feature;

[0110] It should be noted that in the embodiment of the present application, the two generated graph structures are input into the dual-channel graph convolution for parallel feature extraction, and a redundant edge discarding operation is designed to effectively avoid the problem of introducing a large number of redundant edges due to the connection between each node in the graph construction process, which affects the prediction performance of the graph convolution network; and an adaptive redundant feature discarding module is designed in the channel to effectively extract highly correlated features, remove redundant features, reduce the number of parameters and improve the calculation speed.

[0111] Further, refer to Figure 2 Step 102 includes removing the first redundant feature and the second redundant feature corresponding to the historical vehicle speed map and the driving condition characteristic parameter map respectively according to the pre-built adaptive redundant feature discarding module, and obtaining the target historical vehicle speed feature and the target driving condition characteristic parameter feature includes:

[0112] Step 1021, inputting the historical vehicle speed graph and the driving condition characteristic parameter graph into a dual-channel graph convolution to perform parallel feature extraction, so as to obtain historical vehicle speed aggregation features and driving condition characteristic parameter aggregation features;

[0113] It should be noted that in the embodiment of the present application, the two generated graph structures are input into the dual-channel graph convolution for parallel feature extraction, and an adaptive redundant feature discarding module is designed in the channel to effectively extract highly correlated features, remove redundant features, reduce the number of parameters and improve the calculation speed.

[0114] The historical vehicle speed graph G1 = (X1, A1) and the driving condition characteristic parameter graph G2 = (X2, A2) are convolved in two channels and features are extracted in parallel. The calculation process is shown in the following formulas 4 and 5:

[0115]

[0116] Among them, in the above formulas 4 and 5, and To normalize A1 and A2 symmetrically. 1 represents the historical vehicle speed, X2 is the characteristic parameter of the driving condition, and W 1 and W 2 Represents their respective weight matrices, H 1 and H 2 They represent the aggregated features generated by G1 and G2 after graph convolution aggregation.

[0117] Step 1022, performing a dimensionality reduction operation on each node feature in the historical vehicle speed graph and the driving condition characteristic parameter graph according to the historical vehicle speed aggregation feature and the driving condition characteristic parameter aggregation feature, and obtaining a first attention coefficient and a second attention coefficient through a preset function;

[0118] It should be noted that, in the embodiment of the present application, the dual-channel redundant edge discarding operation is as follows: Figure 8 As shown in the figure. Since the cosine similarity and Euclidean distance are used in the construction of the above graph structure, this method will cause each node to be connected to each other, and the number of edges will increase significantly, which will increase the cost of calculation and introduce a large number of redundant edges, affecting the prediction performance of the graph convolutional network.

[0119] Therefore, through the above-generated historical vehicle speed aggregation feature H1 and driving condition feature parameter aggregation feature H2, one-dimensional convolution is performed on each node feature, and the node feature is reduced in dimension. Then, through the preset functions, namely the tanh and sigmoid functions, a set of attention coefficients, namely the first attention coefficient Q1 and the second attention coefficient Q2, are obtained.

[0120] Specifically, it is calculated by Formula 6, which is as follows:

[0121] Q i =Sigmoid(tanh(1dConv(H i ))) (Formula 6)

[0122] Where i=1,2, Hi represents the historical vehicle speed aggregation feature H1, the driving condition feature parameter aggregation feature H2, 1dConv represents the feature extraction and dimensionality reduction operation of each node feature, and Qi represents the corresponding generated first attention coefficient Q1 and second attention coefficient Q2.

[0123] Step 1023, multiplying the first attention coefficient and the first adjacency matrix to obtain a first target adjacency matrix, and multiplying the second attention coefficient and the second adjacency matrix to obtain a second target adjacency matrix;

[0124] It should be noted that the attention coefficient is multiplied with the corresponding first adjacency matrix A1 and second adjacency matrix A2 respectively to guide the generation of updated adjacency matrices, namely the first target adjacency matrix A1Q and the second target adjacency matrix A2Q.

[0125] A iQ =Q i A i (Formula 7)

[0126] Among them, A iQ Represents the updated adjacency matrix, namely the first target adjacency matrix A1Q and the second target adjacency matrix A2Q.

[0127] Step 1024, screening the first target adjacency matrix and the second target adjacency matrix according to a first target threshold value to obtain a first screened adjacency matrix and a second screened adjacency matrix;

[0128] The generated adjacency matrices A1Q and A2Q are subjected to threshold screening, values ​​with similarities lower than 0.4 are returned to 0, and values ​​greater than or equal to 0.4 are retained, thereby obtaining a first screened adjacency matrix and a second screened adjacency matrix.

[0129] In this way, the adjacency matrix is ​​adaptively adjusted according to the learned features, and edges with low similarity are cleared, thereby avoiding redundant edges from introducing a large amount of interference information during the aggregation process.

[0130] Step 1025, performing a graph convolution operation on the first screening adjacency matrix and the second screening adjacency matrix to obtain a first feature and a second feature, and extracting aggregate features of the historical vehicle speed aggregate feature and the driving condition feature parameter aggregate feature through the first adjacency matrix and the second adjacency matrix to obtain a first aggregate feature and a second aggregate feature;

[0131] The newly generated adjacency matrix is ​​subjected to graph convolution again. Specifically, refer to the following formula 8:

[0132]

[0133] In the above formula 8, H iQ Represents the first feature H1Q and the second feature H2Q of the aggregated features generated by graph convolution based on the newly generated adjacency matrix.

[0134] However, in this process of removing redundant edges, the selection of thresholds is extremely important. For complex working conditions, when the node similarity is low, valid edges are easily removed, which is detrimental to the subsequent model performance. Therefore, the original adjacency matrix graph convolution channel is retained, and while extracting features from the new adjacency matrix generated by threshold screening, the original adjacency matrix is ​​also subjected to graph convolution feature aggregation. For details, refer to the following formula

[0135] Formula 9:

[0136]

[0137] In the above formula 9, i = 1, 2, H iO The corresponding H1O and H2O respectively represent the features obtained by aggregating H1 and H2 generated after the first graph convolution aggregation through the original adjacency matrix, that is, the first aggregation feature and the second aggregation feature.

[0138] Step 1026, combining the first feature and the first aggregate feature to obtain a first combined feature, and combining the second feature and the second aggregate feature to obtain a second combined feature;

[0139] Step 1027, inputting the first combined features and the second combined features into a pre-built adaptive redundant feature discarding module, obtaining the Euclidean distance between each pair of nodes corresponding to the first combined features, and the Euclidean distance between each pair of nodes corresponding to the second combined features;

[0140] Step 1028, determining a first redundant feature, a second redundant feature, and a redundant node corresponding to the first redundant feature and a redundant node corresponding to the second redundant feature according to the Euclidean distance;

[0141] Further, the determining of the first redundant feature and the second redundant feature according to the Euclidean distance, and the redundant nodes corresponding to the first redundant feature and the redundant nodes corresponding to the second redundant feature include:

[0142] When the Euclidean distance between each pair of nodes is greater than the second target threshold, each pair of nodes is used as a redundant node, and the node features corresponding to each pair of nodes are used as redundant features.

[0143] Step 1029, removing the redundant nodes corresponding to the first redundant feature and the redundant nodes corresponding to the second redundant feature, and obtaining the target historical vehicle speed feature and the target driving condition feature parameter feature.

[0144] The above steps 1026-1029 design an adaptive redundant feature discarding module in each channel, and the structure is as follows: Fig. 9 As shown, highly correlated features are effectively extracted, redundant features are removed, the number of parameters is reduced, and the calculation speed is improved.

[0145] Specifically, H iQ and H iO As the input of this module, H1Q and H1O are combined to obtain (H1Q, H1O), and H2Q and H2O are combined to obtain (H2Q, H2O). iQ Only the edges are removed, and there is no loss in the number of nodes. Therefore, the number of nodes of the two features in the two combinations is the same. The similarity of each node is calculated. The above Euclidean distance formula is directly used to calculate the two nodes in the same position in the two features, so as to obtain the Euclidean distance of each pair of nodes. The Euclidean distance is normalized to obtain a set of numbers in [0,1]. The closer the Euclidean distance is to 0, the more similar the two feature values ​​are. The node features with a normalized Euclidean distance less than 0.4 are retained. Nodes greater than 0.4 mean that a large amount of interference information is generated in the process of extracting features in the two ways. Finally, the two corresponding redundant features H1D and H2D are obtained in the two channels of the characteristic parameters of the historical vehicle speed and the driving condition. In this way, the dependence on the selection of the threshold in the process of removing redundant edges is effectively avoided, and highly correlated features are effectively extracted. While improving the robustness of the model, the number of nodes is reduced and the calculation speed of the model is improved.

[0146] Step 103, determining the loss value corresponding to each node according to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics;

[0147] Further, step 103, i.e., determining the loss value corresponding to each node according to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics, includes: outputting the predicted vehicle speed corresponding to each node through a preset neural network structure for all nodes corresponding to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics; and obtaining the loss value according to the predicted vehicle speed and the actual vehicle speed corresponding to each of the nodes obtained in advance.

[0148] Step 104, determining a target attention coefficient corresponding to each of the nodes according to the loss value;

[0149] Step 105: output the target predicted speed of the target vehicle through the preset neural network structure according to the real-time speed information of the target vehicle and the target attention coefficient corresponding to each of the nodes.

[0150] It should be noted that in the above steps 103-105, a reverse enhancement module based on the prediction results is constructed, and the structure is as follows: Fig.10 As shown in the figure, the correlation between each node and the prediction task is obtained according to the prediction results, and the reverse node feature enhancement is implemented according to the correlation degree, thereby improving the prediction accuracy of the overall node. The specific process is as follows:

[0151] Then all nodes in H1D and H2D are passed through a layer of multi-layer perceptron (MLP) to output the predicted speed ys. The true label corresponding to the data is yrea l. Then the absolute error L is selected as the loss function, and the loss is calculated by L(ys,yrea l):

[0152]

[0153] In the above formula 10, ys is the predicted vehicle speed, yrea l is the actual vehicle speed corresponding to each of the nodes obtained in advance, i is the number of nodes, and L is the loss value.

[0154] It should be noted that the smaller L is, the more accurate the vehicle speed predicted by the node is, and the higher the relevance of the node to the vehicle speed prediction task is. Conversely, the smaller the relevance of the node to the prediction task is.

[0155] In an embodiment of the present application, a historical vehicle speed map is constructed based on historical vehicle speed information, and a driving condition characteristic parameter map is constructed based on vehicle driving condition characteristic parameters, wherein the vehicle driving condition characteristic parameters are generated by the historical vehicle speed information; the first redundant features and the second redundant features corresponding to the historical vehicle speed map and the driving condition characteristic parameter map, respectively, are removed according to a pre-constructed adaptive redundant feature discarding module to obtain target historical vehicle speed features and target driving condition characteristic parameter features; all nodes corresponding to the target historical vehicle speed features and the target driving condition characteristic parameter features are output through a preset neural network structure, and the predicted vehicle speed corresponding to each node is output; a loss value is obtained based on the predicted vehicle speed and the actual vehicle speed corresponding to each of the nodes obtained in advance; a target attention coefficient corresponding to each of the nodes is determined based on the loss value; and the target predicted vehicle speed of the target vehicle is output through the preset neural network structure based on the real-time vehicle speed information of the target vehicle and the target attention coefficient corresponding to each of the nodes. That is, the present application constructs a graph of characteristic parameters and historical vehicle speeds that characterize the driving conditions of a vehicle. Secondly, feature extraction is performed on the two generated graph structures in parallel, and redundant edges and redundant features are discarded through a pre-built adaptive redundant feature discarding module during the feature extraction process. This effectively avoids the problem of each node being connected to each other during the graph construction process, which leads to the introduction of a large number of redundant edges that affect the prediction performance of the graph convolutional network. It effectively extracts highly correlated features, removes redundant features, reduces the amount of parameters, and improves the calculation speed. Then, based on the predicted vehicle speed, the degree of correlation between each node and the prediction task is obtained, and reverse node feature enhancement is achieved according to the degree of correlation, and the target attention coefficient is updated, thereby improving the prediction accuracy of the overall node and realizing real-time and accurate prediction of the future driving conditions of extended-range vehicles under small sample conditions.

[0156] Figure 3 is a flow chart of the steps of a vehicle speed prediction method provided by an embodiment of the present application, such as Figure 3 As shown, the method may include:

[0157] Step 301, generating the vehicle driving condition characteristic parameters according to the pre-acquired historical vehicle speed information; wherein, the vehicle driving condition characteristic parameters are generated by the historical vehicle speed information, and the vehicle driving condition characteristic parameters include at least one of the average vehicle speed, the driving average vehicle speed, the vehicle speed variance, the acceleration variance, the inertia energy variance, the acceleration time ratio and the deceleration time ratio.

[0158] It should be noted that, in the embodiment of the present application, the average vehicle speed vm in the time window is expressed as the following formula 11:

[0159]

[0160] Where vi is the vehicle speed at the i-th time point in the time window, n is the time window size, and i is any time point in the time window.

[0161] The average driving speed vmd after removing the points where the vehicle speed is zero is shown in the following formula 12:

[0162]

[0163] Where vi is the speed of the vehicle at the i-th time point in the time window. k is the length of time when the vehicle speed is non-zero in the time window. i is any time point in the time window.

[0164] The vehicle speed variance vvar is shown in the following formula 13:

[0165]

[0166] Where vi is the vehicle speed at the i-th time point in the time window, vm is the average vehicle speed in the time window, and n is the time window size.

[0167] The acceleration variance avar is shown in the following formula 14:

[0168]

[0169] Where ai is the acceleration at the i-th moment in the time window, am is the average acceleration in the time window, and n is the size of the time window.

[0170] The inertial energy variance vavar is shown in the following formula 15:

[0171]

[0172] Where vai is the value of the velocity multiplied by the acceleration at the i-th moment in the time window, vam is the average value of the velocity multiplied by the acceleration at the i-th moment in the time window, and n is the size of the time window.

[0173] The acceleration time ratio Pacc is the percentage of the entire acceleration time in the time window.

[0174] Step 302, constructing a historical vehicle speed map according to historical vehicle speed information, and constructing a driving condition characteristic parameter map according to vehicle driving condition characteristic parameters;

[0175] Step 303, removing the first redundant features and the second redundant features corresponding to the historical vehicle speed map and the driving condition characteristic parameter map respectively according to a pre-built adaptive redundant feature discarding module, and obtaining a target historical vehicle speed feature and a target driving condition characteristic parameter feature;

[0176] Step 304, determining the loss value corresponding to each node according to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics;

[0177] Step 305, determining a target attention coefficient corresponding to each of the nodes according to the loss value;

[0178] Further, step 305 includes:

[0179] Adjusting the loss value range to a preset range through a preset function, wherein the preset range is from an initial value to a target value;

[0180] When the loss value corresponding to the node is within a preset range and is less than a first preset value, the attention coefficient corresponding to the node is the target value minus the loss value;

[0181] When the loss value corresponding to the node is within a preset range, greater than the first preset value and less than the second preset value, the attention coefficient corresponding to the node is the loss value;

[0182] When the loss value corresponding to the node is within a preset range and is greater than the second preset value, the attention coefficient corresponding to the node is the initial value;

[0183] Wherein, the second preset value is greater than the first preset value.

[0184] It should be noted that in the embodiments of the present application, reference is made to Fig.10 As shown, first, the L range is adjusted to the preset range through the Sigma function. For example, the preset range is [0,1], then the corresponding target value is 1, the initial value is 0, the first preset value can be 0.4, the second preset value can be 0.6, and L is the loss value.

[0185] Specifically, when L is less than 0.4, it means that the node is highly correlated with the vehicle speed prediction task, and the corresponding attention coefficient is 1-L, which means that the attention coefficient will be close to 1, indicating that the node has a high importance in the vehicle speed prediction task.

[0186] When L is greater than 0.6, it means that the node is weakly correlated with the vehicle speed prediction task, and the corresponding attention coefficient is 1-L. The attention coefficient will be close to 0, but it means that the importance of the node in the vehicle speed prediction task is low.

[0187] When L is between 0.4 and 0.6, L is used as the attention coefficient of the corresponding node. In this way, we can pay more attention to the nodes related to the vehicle speed prediction task and reduce the attention to the nodes that are less related to the task. Real-time and accurate prediction of the future driving conditions of extended-range vehicles can be achieved under small sample conditions.

[0188] Step 306: output the target predicted speed of the target vehicle through the preset neural network structure according to the real-time speed information of the target vehicle and the target attention coefficient corresponding to each of the nodes.

[0189] The present application constructs a graph of characteristic parameters and historical vehicle speeds that characterize the driving conditions of a vehicle. Secondly, feature extraction is performed on the two generated graph structures in parallel, and redundant edges and redundant features are discarded through a pre-built adaptive redundant feature discarding module during the feature extraction process. This effectively avoids the problem of each node being connected to each other during the graph construction process, which leads to the introduction of a large number of redundant edges that affect the prediction performance of the graph convolutional network. Highly correlated features are effectively extracted, redundant features are removed, the amount of parameters is reduced, and the calculation speed is increased. The degree of correlation between each node and the prediction task is obtained based on the predicted vehicle speed, and reverse node feature enhancement is achieved according to the degree of correlation. The target attention coefficient is updated, thereby improving the prediction accuracy of the overall node and achieving real-time and accurate prediction of the future driving conditions of extended-range vehicles under small sample conditions.

[0190] Corresponding to the method provided in the above-mentioned vehicle speed prediction method embodiment of the present application, see Figure 4 The present application also provides a logic block diagram of a vehicle speed prediction device. In this embodiment, the device 400 includes:

[0191] A construction module 401 is used to construct a historical vehicle speed map according to historical vehicle speed information, and to construct a driving condition characteristic parameter map according to vehicle driving condition characteristic parameters, wherein the vehicle driving condition characteristic parameters are generated by the historical vehicle speed information;

[0192] A feature module 402 is used to remove the first redundant feature and the second redundant feature respectively corresponding to the historical vehicle speed map and the driving condition feature parameter map according to a pre-built adaptive redundant feature discarding module, so as to obtain a target historical vehicle speed feature and a target driving condition feature parameter feature;

[0193] A loss value acquisition module 403 is used to determine the loss value corresponding to each node according to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics;

[0194] A target attention coefficient determination module 404 is used to determine a target attention coefficient corresponding to each of the nodes according to the loss value;

[0195] The output module 405 is used to output the target predicted speed of the target vehicle through the preset neural network structure according to the real-time speed information of the target vehicle and the target attention coefficient corresponding to each of the nodes.

[0196] Optionally, the loss value acquisition module includes:

[0197] A first loss value acquisition submodule is used to output the predicted vehicle speed corresponding to each node through a preset neural network structure for all nodes corresponding to the target historical vehicle speed feature and the target driving condition feature parameter feature;

[0198] The second loss value acquisition submodule is used to obtain the loss value according to the predicted vehicle speed and the actual vehicle speed corresponding to each of the nodes acquired in advance.

[0199] Optionally, the device further comprises:

[0200] A generation module is used to generate the vehicle driving condition characteristic parameters according to the pre-acquired historical vehicle speed information; wherein the vehicle driving condition characteristic parameters are generated by the historical vehicle speed information, and the vehicle driving condition characteristic parameters include at least one of the average vehicle speed, the average driving speed, the vehicle speed variance, the acceleration variance, the inertia energy variance, the acceleration time ratio and the deceleration time ratio.

[0201] Optionally, the building blocks include:

[0202] The first construction submodule is used to construct a graph of the historical vehicle speed information based on cosine similarity to obtain the historical vehicle speed graph, wherein the historical vehicle speed graph includes the historical vehicle speed information and a first adjacency matrix.

[0203] Optionally, the building blocks include:

[0204] A second construction submodule is used to obtain a Euclidean distance matrix according to the Euclidean distances between the vehicle driving condition characteristic parameters;

[0205] The third construction submodule is used to construct a driving condition characteristic parameter map according to the vehicle driving condition characteristic parameters and the Euclidean distance matrix, wherein the Euclidean distance matrix is ​​a second adjacency matrix corresponding to the driving condition characteristic parameter map.

[0206] Optionally, the feature module includes:

[0207] A first feature submodule is used to input the historical vehicle speed graph and the driving condition characteristic parameter graph into a dual-channel graph convolution to perform parallel feature extraction, so as to obtain historical vehicle speed aggregation features and driving condition characteristic parameter aggregation features;

[0208] The second feature submodule is used to perform a dimensionality reduction operation on each node feature in the historical vehicle speed graph and the driving condition feature parameter graph according to the historical vehicle speed aggregation feature and the driving condition feature parameter aggregation feature, and obtain a first attention coefficient and a second attention coefficient through a preset function;

[0209] A third feature submodule, configured to multiply the first attention coefficient and the first adjacency matrix to obtain a first target adjacency matrix, and to multiply the second attention coefficient and the second adjacency matrix to obtain a second target adjacency matrix;

[0210] A fourth feature submodule, configured to screen the first target adjacency matrix and the second target adjacency matrix according to a first target threshold value to obtain a first screened adjacency matrix and a second screened adjacency matrix;

[0211] a fifth feature submodule, configured to perform a graph convolution operation on the first screening adjacency matrix and the second screening adjacency matrix to obtain a first feature and a second feature, and extract aggregate features of the historical vehicle speed aggregate feature and the driving condition feature parameter aggregate feature through the first adjacency matrix and the second adjacency matrix to obtain a first aggregate feature and a second aggregate feature;

[0212] a sixth feature submodule, configured to combine the first feature with the first aggregate feature to obtain a first combined feature, and to combine the second feature with the second aggregate feature to obtain a second combined feature;

[0213] a seventh feature submodule, configured to input the first combined feature and the second combined feature into a pre-built adaptive redundant feature discarding module, and obtain the Euclidean distance between each pair of nodes corresponding to the first combined feature, and the Euclidean distance between each pair of nodes corresponding to the second combined feature;

[0214] an eighth feature submodule, configured to determine, according to the Euclidean distance, a first redundant feature, a second redundant feature, and a redundant node corresponding to the first redundant feature and a redundant node corresponding to the second redundant feature;

[0215] The ninth feature submodule is used to remove the redundant nodes corresponding to the first redundant feature and the redundant nodes corresponding to the second redundant feature to obtain the target historical vehicle speed feature and the target driving condition feature parameter feature.

[0216] Optionally, the eighth characteristic submodule includes:

[0217] The feature unit is used to treat each pair of nodes as redundant nodes and the node features corresponding to each pair of nodes as redundant features when the Euclidean distance between the nodes is greater than a second target threshold.

[0218] Optionally, the target attention coefficient determination module includes:

[0219] A first target attention coefficient determination submodule, used to adjust the loss value range to a preset range through a preset function, wherein the preset range is from an initial value to a target value;

[0220] A second target attention coefficient determination submodule, for when the loss value corresponding to the node is within a preset range and is less than a first preset value, the attention coefficient corresponding to the node is the target value minus the loss value;

[0221] A third target attention coefficient determination submodule, configured to determine the attention coefficient corresponding to the node as the loss value when the loss value corresponding to the node is within a preset range, greater than the first preset value, and less than a second preset value;

[0222] The fourth target attention coefficient determination submodule is used to determine that when the loss value corresponding to the node is within a preset range and is greater than the second preset value, the attention coefficient corresponding to the node is the initial value; wherein the second preset value is greater than the first preset value.

[0223] In summary, an embodiment of the present application provides a vehicle speed prediction device. The present application constructs a graph for characteristic parameters and historical vehicle speeds that characterize the driving conditions of a vehicle. Secondly, the two generated graph structures are subjected to parallel feature extraction, and redundant edges and redundant features are discarded through a pre-constructed adaptive redundant feature discarding module during the feature extraction process. This effectively avoids the problem that each node in the graph construction process is connected to each other, resulting in the introduction of a large number of redundant edges that affect the prediction performance of the graph convolutional network. It effectively extracts highly correlated features, removes redundant features, reduces the amount of parameters, and improves the calculation speed. Furthermore, based on the predicted vehicle speed, the degree of correlation between each node and the prediction task is obtained, and reverse node feature enhancement is achieved according to the degree of correlation. The target attention coefficient is updated, thereby improving the prediction accuracy of the overall node and realizing real-time and accurate prediction of the future driving conditions of extended-range vehicles under small sample conditions.

[0224] Figure 5 This is a structural diagram of an electronic device M00 provided in an embodiment of the present application. In the figure, the electronic device M00 includes a processor M01 and a memory M02. The memory M02 stores programs or instructions that can be run on the processor M01. When the program or instructions are executed by the processor M01, the various steps of the above-mentioned vehicle speed prediction method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, they are not repeated here.

[0225] In an embodiment of the present application, the memory M02 can be used to store software programs and various data. The memory M02 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory M02 may include a volatile memory or a non-volatile memory, or the memory x09 may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM) and a direct memory bus random access memory (DRRAM). The memory M02 in the embodiment of the present application includes but is not limited to these and any other suitable types of memories.

[0226] The processor M01 may include one or more processing units; optionally, the processor M01 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor M01.

[0227] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the above-mentioned vehicle speed prediction method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0228] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0229] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned vehicle speed prediction method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0230] It should be understood that the chip involved in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0231] An embodiment of the present application provides a computer program product, which is stored in a storage medium. The program product is executed by at least one processor to implement the various processes of the above-mentioned vehicle speed prediction method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0232] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0233] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the relevant technology, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0234] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A vehicle speed prediction method, characterized in that: The method comprises: Constructing a historical vehicle speed map based on historical vehicle speed information, and constructing a driving condition characteristic parameter map based on vehicle driving condition characteristic parameters; According to a pre-built adaptive redundant feature discarding module, the first redundant feature and the second redundant feature corresponding to the historical vehicle speed map and the driving condition feature parameter map are removed respectively to obtain a target historical vehicle speed feature and a target driving condition feature parameter feature; Determine the loss value corresponding to each node according to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics; Determine the target attention coefficient corresponding to each of the nodes according to the loss value; The target predicted speed of the target vehicle is output through the preset neural network structure according to the real-time speed information of the target vehicle and the target attention coefficient corresponding to each of the nodes.

2. The method according to claim 1, characterized in that Determining the loss value corresponding to each node according to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics includes: All nodes corresponding to the target historical vehicle speed feature and the target driving condition feature parameter feature are output through a preset neural network structure to predict the vehicle speed corresponding to each node; The loss value is obtained according to the predicted vehicle speed and the actual vehicle speed corresponding to each of the nodes obtained in advance.

3. The method according to claim 1, characterized in that Before the step of constructing a driving condition characteristic parameter map according to the vehicle driving condition characteristic parameters, the method includes: Generating the vehicle driving condition characteristic parameters according to the pre-acquired historical vehicle speed information; Among them, the vehicle driving condition characteristic parameters are generated through the historical vehicle speed information, and the vehicle driving condition characteristic parameters include at least one of the average vehicle speed, the average driving speed, the vehicle speed variance, the acceleration variance, the inertia energy variance, the acceleration time ratio and the deceleration time ratio.

4. The method according to claim 1, characterized in that: The constructing of the historical vehicle speed map according to the historical vehicle speed information comprises: The historical vehicle speed information is graph-constructed based on cosine similarity to obtain the historical vehicle speed graph, wherein the historical vehicle speed graph includes the historical vehicle speed information and a first adjacency matrix.

5. The method according to claim 1, characterized in that The constructing of a driving condition characteristic parameter map according to the vehicle driving condition characteristic parameters comprises: Obtaining a Euclidean distance matrix according to the Euclidean distances between the vehicle driving condition characteristic parameters; A driving condition characteristic parameter graph is constructed according to the vehicle driving condition characteristic parameters and the Euclidean distance matrix, wherein the Euclidean distance matrix is ​​a second adjacency matrix corresponding to the driving condition characteristic parameter graph.

6. The method according to claim 1, characterized in that The pre-built adaptive redundant feature discarding module removes the first redundant feature and the second redundant feature corresponding to the historical vehicle speed map and the driving condition feature parameter map, respectively, to obtain the target historical vehicle speed feature and the target driving condition feature parameter feature, including: Inputting the historical vehicle speed graph and the driving condition characteristic parameter graph into dual-channel graph convolution to perform parallel feature extraction, thereby obtaining historical vehicle speed aggregation features and driving condition characteristic parameter aggregation features; According to the historical vehicle speed aggregation feature and the driving condition feature parameter aggregation feature, each node feature in the historical vehicle speed graph and the driving condition feature parameter graph is respectively subjected to a dimensionality reduction operation, and a first attention coefficient and a second attention coefficient are obtained through a preset function; Multiplying the first attention coefficient and the first adjacency matrix to obtain a first target adjacency matrix, and multiplying the second attention coefficient and the second adjacency matrix to obtain a second target adjacency matrix; Screening the first target adjacency matrix and the second target adjacency matrix according to a first target threshold value to obtain a first screened adjacency matrix and a second screened adjacency matrix; Performing a graph convolution operation on the first screening adjacency matrix and the second screening adjacency matrix to obtain a first feature and a second feature, and extracting aggregate features of the historical vehicle speed aggregate feature and the driving condition feature parameter aggregate feature through the first adjacency matrix and the second adjacency matrix to obtain a first aggregate feature and a second aggregate feature; Combining the first feature with the first aggregate feature to obtain a first combined feature, and combining the second feature with the second aggregate feature to obtain a second combined feature; Input the first combined features and the second combined features into a pre-built adaptive redundant feature discarding module to obtain the Euclidean distance between each pair of nodes corresponding to the first combined features and the Euclidean distance between each pair of nodes corresponding to the second combined features; Determine a first redundant feature, a second redundant feature, and a redundant node corresponding to the first redundant feature and a redundant node corresponding to the second redundant feature according to the Euclidean distance; The redundant nodes corresponding to the first redundant feature and the redundant nodes corresponding to the second redundant feature are removed to obtain the target historical vehicle speed feature and the target driving condition feature parameter feature.

7. The method according to claim 6, characterized in that The determining of the first redundant feature and the second redundant feature according to the Euclidean distance, and the redundant nodes corresponding to the first redundant feature and the redundant nodes corresponding to the second redundant feature include: When the Euclidean distance between each pair of nodes is greater than the second target threshold, each pair of nodes is used as a redundant node, and the node features corresponding to each pair of nodes are used as redundant features.

8. The method according to claim 1, characterized in that: Determining the target attention coefficient corresponding to each of the nodes according to the loss value includes: Adjusting the loss value range to a preset range through a preset function, wherein the preset range is from an initial value to a target value; When the loss value corresponding to the node is within a preset range and is less than a first preset value, the attention coefficient corresponding to the node is the target value minus the loss value; When the loss value corresponding to the node is within a preset range, greater than the first preset value and less than the second preset value, the attention coefficient corresponding to the node is the loss value; When the loss value corresponding to the node is within a preset range and is greater than the second preset value, the attention coefficient corresponding to the node is the initial value; Wherein, the second preset value is greater than the first preset value.

9. A vehicle speed prediction device, characterized in that: The device comprises: A construction module, used to construct a historical vehicle speed map according to historical vehicle speed information, and to construct a driving condition characteristic parameter map according to vehicle driving condition characteristic parameters, wherein the vehicle driving condition characteristic parameters are generated by the historical vehicle speed information; A feature module, used for removing the first redundant feature and the second redundant feature respectively corresponding to the historical vehicle speed map and the driving condition feature parameter map according to a pre-built adaptive redundant feature discarding module, to obtain a target historical vehicle speed feature and a target driving condition feature parameter feature; A loss value acquisition module, used to determine the loss value corresponding to each node according to the target historical vehicle speed characteristics and the target driving condition characteristic parameter characteristics; A target attention coefficient determination module, used to determine the target attention coefficient corresponding to each of the nodes according to the loss value; The output module is used to output the target predicted speed of the target vehicle through the preset neural network structure according to the real-time speed information of the target vehicle and the target attention coefficient corresponding to each node.

10. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the vehicle speed prediction method as described in any one of claims 1 to 8.

Images