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

By using multi-window adaptive fusion and multi-vehicle speed information enhancement modules in vehicle speed prediction, combined with graph convolution network and LSTM, the problem of difficult-to-express interaction between vehicles is solved, and the accuracy and robustness of vehicle speed prediction are improved.

CN119078853APending Publication Date: 2024-12-06CHONGQING JINKANG NEW ENERGY VEHICLE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411353778.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-26
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

When using complex road scenarios, existing vehicle speed prediction methods are difficult to accurately predict vehicle driving trajectories, especially because the interaction factors between vehicles are difficult to accurately express, resulting in poor results in prediction models.

Method used

By obtaining the historical vehicle speed data of the vehicle, dividing and extracting feature data according to different time windows, combining fusion features for vehicle speed prediction, using graph convolution network and long and short-term memory network (LSTM) to extract hidden features, and considering the interaction between vehicles. Feature enhancement is performed using the Euclidean distance matrix and preset ratio data.

Benefits of technology

It achieves efficient extraction of vehicle speed characteristics, improves the accuracy and robustness of vehicle speed prediction, and is suitable for vehicle speed prediction in complex road scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119078853A_ABST
    Figure CN119078853A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a vehicle speed prediction method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining first historical vehicle speed data of a first vehicle; dividing the first historical vehicle speed data into first vehicle speed sections according to a plurality of preset time windows, and extracting first vehicle speed feature data corresponding to each first vehicle speed section; generating first fusion feature data of the first vehicle according to the multiple pieces of first vehicle speed feature data; and performing speed prediction on the first vehicle based on the first fusion feature data. According to the embodiment of the invention, the hidden features of different time windows are fused, and the vehicle speed features are efficiently extracted, so that vehicle speed prediction is facilitated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a vehicle speed prediction method and device, electronic equipment, and a storage medium. Background Art

[0002] As the core technology of extended-range electric vehicle control, vehicle energy management strategy plays a vital role in improving fuel economy and reducing emissions. It is mainly divided into rule-based and optimization-based energy management strategies.

[0003] The main idea of ​​the rule-based energy management strategy is to determine the working state of the power drive system by defining a series of rules for the operation of the vehicle power system. It has the advantages of small online calculation amount, low technical difficulty, short development cycle, etc., and is widely used in industry; however, it cannot guarantee the optimality of the control effect, and the algorithm has poor portability. As one of the energy management strategies based on optimization, model predictive control transforms the global optimization problem in the complete working condition into a local optimization problem in the future control time domain. By predicting the working condition information in a rolling manner, the control quantity is optimized and solved, and the control quantity is applied to the underlying controller for feedback correction and enters the next control cycle. It takes into account both real-time and optimization, and has good robustness. Since the working condition information in the future control time domain will directly serve as the basis for the subsequent optimization control algorithm, the accuracy of the predicted working condition information will greatly affect the performance of the predictive energy management algorithm. Therefore, the rapid and accurate prediction of the working condition information in the future control time domain is the core and difficulty of the research on the model predictive control energy management strategy of extended-range electric vehicles. The working condition prediction in the model predictive control energy management strategy is mainly through the prediction of vehicle speed.

[0004] In general, vehicle speed prediction methods can be divided into statistical model-based and machine learning-based data-driven methods. Among them, prediction methods based on statistical models, such as the autoregressive moving average model, have simple models, mature parameter selection methods, and small computational complexity, but require that the time series data is stable or stable after differentiation, and can only describe the linear relationship between data in essence. However, vehicle speed is the result of the combined effect of the driver's intention and the driving environment, and is highly nonlinear. Therefore, prediction methods based solely on statistical models cannot achieve good prediction accuracy. Summary of the invention

[0005] In view of the above problems, a vehicle speed prediction method and device, electronic device, and storage medium are proposed to overcome the above problems or at least partially solve the above problems, including:

[0006] A vehicle speed prediction method, the method comprising:

[0007] Acquire first historical vehicle speed data of a first vehicle;

[0008] Dividing the first historical vehicle speed data into first vehicle speed segments according to a plurality of preset time windows, and extracting first vehicle speed characteristic data corresponding to each first vehicle speed segment;

[0009] generating first fused feature data of the first vehicle according to a plurality of first vehicle speed feature data;

[0010] A vehicle speed prediction is performed on the first vehicle based on the first fused feature data.

[0011] Optionally, the method further comprises:

[0012] Acquire second historical vehicle speed data of a second vehicle, where the second vehicle is a surrounding vehicle of the first vehicle;

[0013] The predicting the speed of the first vehicle based on the first fused feature data includes:

[0014] determining a target time window from the plurality of preset time windows based on the first vehicle speed characteristic data;

[0015] Dividing the second historical vehicle speed data into second vehicle speed segments according to the target time window, and extracting second vehicle speed characteristic data corresponding to each second vehicle speed segment;

[0016] generating second fused characteristic data of the second vehicle according to the second vehicle speed characteristic data;

[0017] Determine target fused feature data based on the first fused feature data and the second fused feature data;

[0018] The speed of the first vehicle is predicted based on the target fusion feature data.

[0019] Optionally, generating the first fusion feature data of the first vehicle according to the plurality of first vehicle speed feature data includes:

[0020] Within a preset first time period, obtaining target first vehicle speed characteristic data corresponding to the plurality of preset time windows;

[0021] forming a graph structure based on the target first vehicle speed characteristic data;

[0022] Determine an adjacency matrix and an eigenvector matrix corresponding to the graph structure;

[0023] First fused feature data of the first vehicle is generated according to the adjacency matrix and the feature vector matrix.

[0024] Optionally, generating first fused feature data of the first vehicle according to the adjacency matrix and the feature vector matrix includes:

[0025] Generating an attention coefficient of the graph structure according to the adjacency matrix and the eigenvector matrix;

[0026] First fused feature data of the first vehicle is generated based on the product of the attention coefficient and the feature vector matrix.

[0027] Optionally, determining target fused feature data based on the first fused feature data and the second fused feature data includes:

[0028] obtaining preset ratio data for controlling surrounding vehicle speed characteristics;

[0029] A target product of the second fused feature data and the preset ratio data is determined, and the target product is added to the first fused feature data to obtain target fused feature data.

[0030] Optionally, the attention coefficient includes an attention vector for each first vehicle speed segment, and determining a target time window from the plurality of preset time windows based on the first vehicle speed feature data includes:

[0031] Adding the attention vectors of the first vehicle speed segment in each preset time window to obtain the attention vector sum;

[0032] A target time window is determined based on the attention vector and from the plurality of preset time windows.

[0033] Optionally, generating second fusion feature data of the second vehicle according to the second vehicle speed feature data includes:

[0034] Acquire the position data of the second vehicle within a preset second time period;

[0035] Determine a Euclidean distance matrix between any two second vehicles based on the position data;

[0036] Determining second vehicle speed characteristic data within a preset second time period;

[0037] Second fused feature data of the second vehicle is generated according to the Euclidean distance matrix and the second vehicle speed feature data.

[0038] A vehicle speed prediction device, comprising:

[0039] A first historical vehicle speed data acquisition module, used to acquire first historical vehicle speed data of a first vehicle;

[0040] A first time window division module, used to divide the first historical vehicle speed data into first vehicle speed segments according to a plurality of preset time windows, and extract first vehicle speed characteristic data corresponding to each first vehicle speed segment;

[0041] A first fusion feature data generating module, used for generating first fusion feature data of the first vehicle according to a plurality of first vehicle speed feature data;

[0042] A vehicle speed prediction module is used to predict the vehicle speed of the first vehicle based on the first fused feature data.

[0043] An electronic device comprises a processor, a memory and a computer program stored in the memory and capable of running on the processor, wherein the computer program implements the vehicle speed prediction method as described above when executed by the processor.

[0044] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the vehicle speed prediction method as described above is implemented.

[0045] The embodiments of the present invention have the following advantages:

[0046] The embodiment of the present invention obtains first historical vehicle speed data of a first vehicle; divides the first historical vehicle speed data into first vehicle speed segments according to a plurality of preset time windows, and extracts first vehicle speed feature data corresponding to each first vehicle speed segment; generates first fused feature data of the first vehicle according to the plurality of first vehicle speed feature data; and predicts the vehicle speed of the first vehicle based on the first fused feature data, thereby integrating hidden features of vehicle speeds in different time windows and efficiently extracting vehicle speed features itself to facilitate vehicle speed prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solution of the present invention, the accompanying drawings required for use in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0048] Figure 1a is a flowchart of a vehicle speed prediction method provided by an embodiment of the present invention;

[0049] Figure 1b It is a schematic diagram of dividing vehicle speed data provided by an embodiment of the present invention;

[0050] Figure 1c It is a schematic diagram of dividing vehicle speed data provided by an embodiment of the present invention;

[0051] Figure 2 is a flowchart of another vehicle speed prediction method provided by an embodiment of the present invention;

[0052] Figure 3is a flowchart of another vehicle speed prediction method provided by an embodiment of the present invention;

[0053] Figure 4 It is a schematic diagram of the structure of a vehicle speed prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] In recent years, deep learning methods have received great attention due to their powerful nonlinear fitting capabilities. Long short-term memory (LSTM) neural network has achieved significant improvement in prediction accuracy for vehicle speed prediction compared to traditional machine learning algorithms such as support vector machine (SVM) and BP neural network. However, existing methods all use fixed windows for feature extraction, ignoring the impact of different window sizes. In addition, in real road scenes, the factors that affect vehicle driving trajectories are very complex. Driving is not only constrained by road structure and traffic rules, but also by random factors such as the driver's subjective consciousness. Among them, the interaction between vehicles is an important factor affecting vehicle trajectory, but it is usually difficult to accurately express, resulting in poor prediction model results.

[0056] In an embodiment of the present invention, the historical speed data of the vehicle can be divided according to different time windows, and then features can be extracted for feature integration, and then the fused feature data can be combined to predict the vehicle speed. Thus, by fusing the hidden features of the vehicle speed in different time windows, the vehicle speed features can be efficiently extracted, thereby achieving accurate prediction of the vehicle speed.

[0057] Reference Figure 1a , shows a flow chart of a vehicle speed prediction method provided by an embodiment of the present invention, which may specifically include the following steps:

[0058] Step S101, obtaining first historical vehicle speed data of a first vehicle;

[0059] The first vehicle is a vehicle for which a speed is to be preset, and the first historical speed data can be obtained through an onboard sensor in the first vehicle, which can be a speed sensor, such as a wheel speed sensor, a GPS, etc. The first historical speed data is a speed data set of the first vehicle in a certain period of time in the past, including speed data at each collection time in a certain period of time in the past. In practical applications, the first historical speed data of the first vehicle in the most recent period of time can be obtained.

[0060] Step S102, dividing the first historical vehicle speed data into first vehicle speed segments according to a plurality of preset time windows, and extracting first vehicle speed characteristic data corresponding to each first vehicle speed segment;

[0061] After obtaining the first historical vehicle speed data, the first historical vehicle speed data may be split into a plurality of vehicle speed segments. Specifically, different time windows may be set, and then the splitting may be performed according to the set time windows.

[0062] After the time window splitting is completed, feature extraction can be performed for each first vehicle speed segment to obtain the first vehicle speed feature data of each first vehicle speed segment. For each time window, the corresponding first vehicle end feature data can be obtained. In an embodiment of the present invention, the first vehicle speed end can be feature extracted using an LSTM (Long Short-Term Memory) model to obtain hidden state features.

[0063] It should be noted that the time window in the embodiment of the present invention can be set according to the actual scenario, and there are no excessive restrictions on this in the embodiment of the present invention.

[0064] Reference Figure 1b , showing a schematic diagram of vehicle speed data division provided by an embodiment of the present invention.

[0065] The historical speed information of the first vehicle can be divided into three windows of different sizes, namely, T1=5s, T2=10s and T3=20s. That is, when divided according to the window of T1=5s, every 5s of speed information is divided into a first speed segment. Similarly, when divided according to the window of T2=10s, every 10s of speed information is divided into a first speed segment. When divided according to the window of T3=20s, every 20s of speed information is divided into a first speed segment.

[0066] Step S103, generating first fusion feature data of the first vehicle according to the plurality of first vehicle speed feature data;

[0067] After obtaining a plurality of first vehicle speed characteristic data, the vehicle speed characteristic data of different preset time windows can be fused to obtain the first fused characteristic data of the first vehicle. By fusing the characteristics of different time windows, the influence of different time windows on the vehicle speed prediction can be overcome, thereby improving the accuracy of the vehicle speed prediction.

[0068] In one embodiment of the present invention, first fused feature data of a first vehicle is generated based on multiple first vehicle speed feature data, including: obtaining target first vehicle speed feature data corresponding to multiple preset time windows within a preset first time period; constructing a graph structure based on the target first vehicle speed feature data; determining an adjacency matrix and an eigenvector matrix corresponding to the graph structure; and generating the first fused feature data of the first vehicle according to the adjacency matrix and the eigenvector matrix.

[0069] In practical applications, the preset first time period can be determined according to multiple preset time windows, and the preset first time period includes at least an integer number of preset time windows. In one example, the preset first time period can be set according to a common multiple of the multiple preset time windows. For example, when the time windows are 5s, 10s, and 20s, the first preset time period can be 20s, and 20s can include 4 5s time windows, 2 10s time windows, and 1 20s time window.

[0070] After determining the preset first time period, multiple target first vehicle speed characteristic data within the preset first time period can be divided into a group of vehicle speed characteristic data under the preset time window under each preset time window. Under the same preset first time period, a group of vehicle speed characteristic data in each preset time window constitutes a graph structure.

[0071] For example, 20 seconds may include four 5-second time windows, two 10-second time windows, and one 20-second time window. Then, within the 20 seconds, the target first vehicle speed characteristic data corresponding to the four 5-second time windows are regarded as a group, the target first vehicle speed characteristic data corresponding to the two 10-second time windows are regarded as a group, and a group of vehicle speed characteristic data in one 20-second time window are regarded as a group. The three groups of data are combined to form the graph structure within the 20 seconds, and so on, to obtain the graph structure composed of all the first historical vehicle speed data.

[0072] After the graph structure is generated, the adjacency matrix and the eigenvector matrix corresponding to the graph structure may be determined; and the first fused feature data of the first vehicle may be generated according to the adjacency matrix and the eigenvector matrix.

[0073] In one embodiment of the present invention, first fused feature data of the first vehicle is generated according to an adjacency matrix and an eigenvector matrix, including: generating an attention coefficient of a graph structure according to the adjacency matrix and the eigenvector matrix; and generating the first fused feature data of the first vehicle based on the product of the attention coefficient and the eigenvector matrix.

[0074] Reference Figure 1c , shows a schematic diagram of vehicle speed data division provided by an embodiment of the present invention. After the first vehicle speed data is divided, the divided first vehicle speed segments can be respectively input into the LSTM, and the hidden state of the LSTM in each time window is output.

[0075] For example, when the window size is T1=5s, it can be expressed as H1, where H1=[h1,h2,…,hn] corresponds to the hidden state features obtained after n 5s time windows are processed by LSTM.

[0076] When the window size is T2=10s, it can be expressed as H2, where H2=[p1,p2,…,pn / 2] corresponds to the hidden state features obtained after n / 2 10s time windows are processed by LSTM.

[0077] When the window size is T3=20s, it can be expressed as H3, where H3=[d1,d2,…,dn / 4] corresponds to the hidden state features obtained after n / 4 20s time windows are processed by LSTM.

[0078] In the embodiment of the present invention, hidden state features after LSTM extraction under three different time windows can be obtained. A smaller window can capture short-term dependencies, while a larger window can capture long-term dependencies.

[0079] The four hidden layer features of the window size T1 = 5s are grouped together, the two hidden layer features of the window size T2 = 10s are grouped together, and each hidden feature of the window size T3 = 20s is grouped together. In this way, n / 4 groups of feature vectors of the same size can be obtained under each window. Each group is regarded as a node, and the feature vector in this group is the node feature. The three groups corresponding to the three windows in the same time period can form a graph structure composed of three nodes. The nodes are connected to each other to obtain an adjacency matrix. In this way, n / 4 graph structures can be obtained, which can be specifically expressed as Gi = (A i , F i ), where i=1,2,…,n / 4, A i represents the adjacency matrix in the i-th time period, F i The feature vector matrix representing the three nodes in the i-th time period (i.e., the preset first time period) is then obtained by extracting the spatial features of different windows through the graph convolutional network and then through Sigmoid;

[0080] Right now:

[0081] where i=1,2,…,n / 4,a i Represents the attention coefficient generated by the i-th graph structure, with a size of 3×1. Each node corresponds to an attention vector. is the symmetric normalized adjacency matrix, σ is the Sigmoid function, W i Represents the respective weight matrices.

[0082] Then the attention coefficient a i With the eigenvector F iMultiply and adaptively fuse the features of the three windows to obtain the first fused vehicle speed feature H F The calculation process is as follows:

[0083] H F =a i ·F i

[0084] Step S104: predicting the speed of the first vehicle based on the first fused feature data.

[0085] After the first fused feature data is obtained, the vehicle speed may be predicted based on the first fused feature data.

[0086] In an embodiment of the present invention, first historical vehicle speed data of a first vehicle is acquired; the first historical vehicle speed data is divided into first vehicle speed segments according to a plurality of preset time windows, and first vehicle speed feature data corresponding to each first vehicle speed segment is extracted; first fused feature data of the first vehicle is generated according to the plurality of first vehicle speed feature data; and the vehicle speed of the first vehicle is predicted based on the first fused feature data, thereby realizing the fusion of hidden features of different time windows and efficiently extracting the vehicle speed features itself for the convenience of speed prediction.

[0087] Reference Figure 2 , shows a flowchart of another vehicle speed prediction method provided by an embodiment of the present invention, which may specifically include the following steps:

[0088] Step S201, obtaining first historical vehicle speed data of a first vehicle, and obtaining second historical vehicle speed data of a second vehicle, where the second vehicle is a surrounding vehicle of the first vehicle;

[0089] The second vehicle may be any one or more of a left vehicle, a right vehicle or a front vehicle of the first vehicle.

[0090] Step S202, dividing the first historical vehicle speed data into first vehicle speed segments according to a plurality of preset time windows, and extracting first vehicle speed characteristic data corresponding to each first vehicle speed segment;

[0091] Step S203, generating first fusion feature data of the first vehicle according to the plurality of first vehicle speed feature data;

[0092] Step S204, determining a target time window from a plurality of preset time windows based on the first vehicle speed characteristic data;

[0093] In one embodiment of the present invention, the attention coefficient includes an attention vector for each first vehicle speed segment, and a target time window is determined from multiple preset time windows based on the first vehicle speed characteristic data, including: adding the attention vectors of the first vehicle speed segment in each preset time window to obtain the attention vector sum; and determining the target time window from multiple preset time windows based on the attention vector sum.

[0094] Among them, the attention vector and the largest time window can be used as the target time window.

[0095] Step S205, dividing the second historical vehicle speed data into second vehicle speed segments according to the target time window, and extracting second vehicle speed characteristic data corresponding to each second vehicle speed segment;

[0096] Step S206, generating second fusion characteristic data of the second vehicle according to the second vehicle speed characteristic data;

[0097] In one embodiment of the present invention, second fused feature data of a second vehicle is generated based on second vehicle speed feature data, including: obtaining position data of the second vehicle within a preset second time period; determining a Euclidean distance matrix of any two second vehicles based on the position data; determining the second vehicle speed feature data within a preset second time period; and generating second fused feature data of the second vehicle according to the Euclidean distance matrix and the second vehicle speed feature data.

[0098] Among them, the preset second time period can be the same as the preset first time period or different. The preset second time period can be set according to actual scene requirements, and no excessive restrictions are made on this in the embodiments of the present invention.

[0099] In an embodiment of the present invention, GCN (graph convolutional network) can be applied to aggregate the speeds of three vehicles in each time period.

[0100] In an embodiment of the present invention, a multi-vehicle speed information enhancement module can be designed. The multi-vehicle speed information enhancement module can add the 4 / n attention vectors under each of the above-mentioned time windows, and by comparing the size of the sum, select the time window size corresponding to the highest attention vector addition value as the optimal time window, and divide the historical speeds of the three surrounding vehicles into windows based on the optimal time window.

[0101] Assuming that the optimal time window is T2 = 10s, the historical speeds of the three surrounding cars V2, V3, and V4 are divided into n / 2 segments according to the window size of 10s. Then, according to the actual position relationship of the three cars in each segment, the three cars are regarded as three nodes. The Euclidean distance matrix of the three cars is obtained by Euclidean distance calculation as the adjacency matrix input of the graph convolution, denoted as A m , where m = 1, 2, n / 2, and the speed information of the three cars in this time period is used as the node feature, denoted by Fm , and then the speeds of the three vehicles in each time period (i.e. the preset second time period) are aggregated through GCN to obtain the spatial characteristics of the surrounding vehicles. The calculation process is as follows:

[0102]

[0103] Among them, H m represents the aggregated features generated by graph convolution aggregation of the speeds of three vehicles in each time period, W m Represents the weight matrix in the corresponding time period.

[0104] Step S207, determining target fused feature data based on the first fused feature data and the second fused feature data;

[0105] In one embodiment of the present invention, target fusion feature data is determined based on the first fusion feature data and the second fusion feature data, including: obtaining preset ratio data for controlling the surrounding vehicle speed characteristics; determining a target product of multiplying the second fusion feature data by the preset ratio data, and adding the target product to the first fusion feature data to obtain the target fusion feature data.

[0106] That is, target fusion feature data = first fusion feature data + second fusion feature data * preset ratio data.

[0107] In practical applications, LSTM can be used to m Extract and then integrate the vehicle speed feature H F Additive aggregation is used to enhance the vehicle speed feature and obtain the output feature H out , the calculation process is as follows:

[0108] H out =H F +βH m

[0109] Among them, β is the ratio of controlling the surrounding vehicle speed characteristics, and its value range is 0.1-0.3.

[0110] After getting the output feature H out After that, the future vehicle speed can be predicted through a multi-layer perceptron (MLP).

[0111] Step S208: predicting the speed of the first vehicle based on the target fusion feature data.

[0112] In the embodiment of the present invention, a method for predicting the speed of an extended-range vehicle considering the influence of multiple vehicles is proposed to solve the problem that the existing methods ignore the influence of different window sizes and the interaction between vehicles is difficult to accurately express, resulting in poor results of the speed prediction model. In the speed prediction method, the historical speed of the vehicle itself and the historical speed of the surrounding vehicles and the coordinate information of the surrounding vehicles are first selected as the input of the subsequent model; then a multi-window adaptive fusion module is designed to divide the historical speed of the vehicle into three different size windows and input them into LSTM to extract the time information under different windows, and then the hidden features of different windows are adaptively fused through the graph convolution network to efficiently extract the speed features of the vehicle itself; secondly, a multi-vehicle speed information enhancement module is designed to divide the speed information of multiple vehicles into windows by obtaining the optimal window setting, and then the spatial features of multiple speeds in each time period are obtained according to the actual position relationship through GCN, and then the time-related features are extracted through LSTM, and then the speed features of the vehicle itself are enhanced, so as to provide a reliable speed prediction method for the energy management strategy of extended-range vehicles with better economy and real-time performance.

[0113] In the embodiment of the present invention, the vehicle's own historical speed is divided into windows of different sizes to extract hidden features, and the hidden features of different windows are fused to efficiently extract the vehicle's own speed features. In addition, by obtaining the optimal time window setting, the speed information of multiple vehicles around the vehicle is divided into windows, and spatial features are obtained according to the actual position relationship, thereby enhancing the vehicle's own speed features and improving the accuracy of speed prediction.

[0114] Reference Figure 3 , shows a vehicle speed prediction method in the implementation of the present invention, which mainly includes the following process:

[0115] First, the historical speed of the car itself V1 and the historical speeds of the three surrounding vehicles are selected, which are the historical speeds of the left vehicle, the front vehicle and the right vehicle, respectively, represented by V2, V3, and V4, and the coordinate information of the three surrounding vehicles is obtained at the same time. Then a multi-window adaptive fusion module is designed to divide the historical speed of the car into three windows of different sizes, namely T1 = 5s, T2 = 10s and T3 = 20s.

[0116] The divided speed segments are input into LSTM, and the hidden state of LSTM in each time window is output. When the window size is T1=5s, it can be expressed as H1, where H1=[h1,h2,…,hn] corresponds to the hidden state features obtained after n 5s time windows are processed by LSTM; when the window size is T2=10s, it can be expressed as H2, where H2=[p1,p2,…,pn / 2] corresponds to the hidden state features obtained after n / 2 10s time windows are processed by LSTM; when the window size is T3=20s, it can be expressed as H3, where H3=[d1,d2,…,dn / 4] corresponds to the hidden state features obtained after n / 4 20s time windows are processed by LSTM.

[0117] Since the length of the historical speed is constant, different window sizes will result in different numbers of hidden state features extracted. The four hidden layer features of the window size T1 = 5s are grouped together, the two hidden layer features of the window size T2 = 10s are grouped together, and each hidden feature of the window size T3 = 20s is grouped together. In this way, n / 4 groups of feature vectors of the same size can be obtained under each window;

[0118] Each group is regarded as a node, and the feature vector in this group is the node feature. The three groups corresponding to the three windows in the same time period can form a graph structure composed of three nodes. The nodes are connected to each other to obtain the adjacency matrix, thus obtaining n / 4 graph structures, using Gi=(A i , F i ), where i = 1, 2, ..., n / 4, A i represents the adjacency matrix in the i-th time period, F i Represents the feature vector matrix of the three nodes in the i-th time period. Then, after extracting the spatial features of different windows through the graph convolutional network, an attention coefficient between [0,1] is obtained through Sigmoid, and each node corresponds to an attention vector.

[0119] The attention coefficient a i With the eigenvector F i Multiply and adaptively fuse the features of the three windows to obtain the fused vehicle speed feature H F (ie the first fused feature data).

[0120] Based on the sum of the attention coefficients, the optimal window size is selected, and the historical speeds of the three surrounding vehicles are divided according to the optimal window size to obtain the second speed segment.

[0121] The position relationship data of surrounding vehicles are divided according to the optimal window size, and then the Euclidean distance is calculated based on the position relationship data, and an adjacency matrix is ​​generated according to the Euclidean distance.

[0122] The adjacency matrix and the divided second vehicle speed segment are combined and input into the graph convolution for convolution calculation to obtain the spatial features of the surrounding vehicles, i.e., the second fused feature data.

[0123] Then, the target fusion feature data is generated by combining the first fusion data and the second fusion feature data, and then input into the MLP to realize the prediction of the vehicle speed.

[0124] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0125] Reference Figure 4 , shows a schematic diagram of the structure of a vehicle speed prediction device provided by an embodiment of the present invention, which may specifically include the following modules:

[0126] A first historical vehicle speed data acquisition module 401 is used to acquire first historical vehicle speed data of a first vehicle;

[0127] A first time window division module 402, configured to divide the first historical vehicle speed data into first vehicle speed segments according to a plurality of preset time windows, and extract first vehicle speed characteristic data corresponding to each first vehicle speed segment;

[0128] A first fusion feature data generating module 403, configured to generate first fusion feature data of the first vehicle according to a plurality of first vehicle speed feature data;

[0129] The vehicle speed prediction module 404 is used to predict the vehicle speed of the first vehicle based on the first fused feature data.

[0130] In one embodiment of the present invention, the device further comprises:

[0131] A second historical vehicle speed data acquisition module, used to acquire second historical vehicle speed data of a second vehicle, where the second vehicle is a surrounding vehicle of the first vehicle;

[0132] The vehicle speed prediction module 404 may include:

[0133] a target time window determination submodule, configured to determine a target time window from the plurality of preset time windows based on the first vehicle speed characteristic data;

[0134] A second vehicle speed characteristic data extraction submodule, used for dividing the second historical vehicle speed data into second vehicle speed segments according to the target time window, and extracting second vehicle speed characteristic data corresponding to each second vehicle speed segment;

[0135] A second fused feature data generating submodule, configured to generate second fused feature data of the second vehicle according to the second vehicle speed feature data;

[0136] a target fusion feature data determination submodule, configured to determine target fusion feature data based on the first fusion feature data and the second fusion feature data;

[0137] A vehicle speed prediction submodule is used to predict the vehicle speed of the first vehicle based on the target fusion feature data.

[0138] In one embodiment of the present invention, the first fusion feature data generating module 403 may include:

[0139] A target first vehicle speed characteristic data determination submodule is used to obtain the target first vehicle speed characteristic data corresponding to the plurality of preset time windows within a preset first time period;

[0140] A graph structure acquisition submodule, configured to construct a graph structure based on the target first vehicle speed characteristic data;

[0141] A matrix data determination submodule, used to determine the adjacency matrix and eigenvector matrix corresponding to the graph structure;

[0142] The first fused feature data generating submodule is used to generate first fused feature data of the first vehicle according to the adjacency matrix and the feature vector matrix.

[0143] In one embodiment of the present invention, the first fusion feature data generating submodule may include:

[0144] An attention coefficient generating unit, used for generating an attention coefficient of the graph structure according to the adjacency matrix and the eigenvector matrix;

[0145] The first fused feature data generating unit is used to generate first fused feature data of the first vehicle based on the product of the attention coefficient and the feature vector matrix.

[0146] In one embodiment of the present invention, the target fusion feature data determination submodule may include:

[0147] a preset ratio data acquisition unit, for acquiring preset ratio data for controlling surrounding vehicle speed characteristics;

[0148] The target fusion feature data determination unit is used to determine a target product of the second fusion feature data and the preset ratio data, and add the target product to the first fusion feature data to obtain the target fusion feature data.

[0149] In one embodiment of the present invention, the attention coefficient includes an attention vector for each first vehicle speed segment, and the target time window determination submodule may include:

[0150] an attention vector sum determination unit, configured to add the attention vectors of the first vehicle speed segment in each preset time window to obtain an attention vector sum;

[0151] A target time window determination unit is used to determine a target time window based on the attention vector and from the multiple preset time windows.

[0152] In one embodiment of the present invention, the second fusion feature data generating submodule may include:

[0153] A position data acquisition unit, used to acquire the position data of the second vehicle within a preset second time period;

[0154] a Euclidean distance matrix determining unit, configured to determine the Euclidean distance matrix of any two second vehicles based on the position data;

[0155] A second vehicle speed characteristic data determining unit, used to determine second vehicle speed characteristic data within a preset second time period;

[0156] The second fused feature data generating unit is used to generate second fused feature data of the second vehicle according to the Euclidean distance matrix and the second vehicle speed feature data.

[0157] In an embodiment of the present invention, first historical vehicle speed data of a first vehicle is acquired; the first historical vehicle speed data is divided into first vehicle speed segments according to a plurality of preset time windows, and first vehicle speed feature data corresponding to each first vehicle speed segment is extracted; first fused feature data of the first vehicle is generated according to the plurality of first vehicle speed feature data; and the vehicle speed of the first vehicle is predicted based on the first fused feature data, thereby realizing the fusion of hidden features of different time windows and efficiently extracting the vehicle speed features itself for the convenience of speed prediction.

[0158] An embodiment of the present invention further provides an electronic device, which may include a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, the above vehicle speed prediction method is implemented.

[0159] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above vehicle speed prediction method is implemented.

[0160] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0161] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0162] Those skilled in the art will appreciate that the embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0163] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate a device for implementing the functions specified in one process or multiple processes in the flowchart and / or one box or multiple boxes in the block diagram.

[0164] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce computer-implemented processing, so that the instructions executed on the computer or other programmable terminal device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0166] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0167] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0168] The above provides a detailed introduction to a vehicle speed prediction method and device, electronic device, and storage medium. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A vehicle speed prediction method, characterized in that: The method comprises: Acquire first historical vehicle speed data of a first vehicle; Dividing the first historical vehicle speed data into first vehicle speed segments according to a plurality of preset time windows, and extracting first vehicle speed characteristic data corresponding to each first vehicle speed segment; generating first fused feature data of the first vehicle according to a plurality of first vehicle speed feature data; A vehicle speed prediction is performed on the first vehicle based on the first fused feature data.

2. The method according to claim 1, characterized in that: The method further comprises: Acquire second historical vehicle speed data of a second vehicle, where the second vehicle is a surrounding vehicle of the first vehicle; The predicting the speed of the first vehicle based on the first fused feature data includes: determining a target time window from the plurality of preset time windows based on the first vehicle speed characteristic data; Dividing the second historical vehicle speed data into second vehicle speed segments according to the target time window, and extracting second vehicle speed characteristic data corresponding to each second vehicle speed segment; generating second fused characteristic data of the second vehicle according to the second vehicle speed characteristic data; Determine target fused feature data based on the first fused feature data and the second fused feature data; The speed of the first vehicle is predicted based on the target fusion feature data.

3. The method according to claim 1 or 2, characterized in that: The step of generating first fusion feature data of the first vehicle according to a plurality of first vehicle speed feature data comprises: Within a preset first time period, obtaining target first vehicle speed characteristic data corresponding to the plurality of preset time windows; forming a graph structure based on the target first vehicle speed characteristic data; Determine an adjacency matrix and an eigenvector matrix corresponding to the graph structure; First fused feature data of the first vehicle is generated according to the adjacency matrix and the feature vector matrix.

4. The method according to claim 3, characterized in that The step of generating first fused feature data of the first vehicle according to the adjacency matrix and the feature vector matrix includes: Generating an attention coefficient of the graph structure according to the adjacency matrix and the eigenvector matrix; First fused feature data of the first vehicle is generated based on the product of the attention coefficient and the feature vector matrix.

5. The method according to claim 2, characterized in that: The determining target fusion feature data based on the first fusion feature data and the second fusion feature data includes: obtaining preset ratio data for controlling surrounding vehicle speed characteristics; A target product of the second fused feature data and the preset ratio data is determined, and the target product is added to the first fused feature data to obtain target fused feature data.

6. The method according to claim 4, characterized in that The attention coefficient includes an attention vector for each first vehicle speed segment, and determining a target time window from the plurality of preset time windows based on the first vehicle speed feature data includes: Adding the attention vectors of the first vehicle speed segment in each preset time window to obtain the attention vector sum; A target time window is determined based on the attention vector and from the plurality of preset time windows.

7. The method according to claim 2, characterized in that The step of generating second fusion feature data of the second vehicle according to the second vehicle speed feature data comprises: Acquire the position data of the second vehicle within a preset second time period; Determine a Euclidean distance matrix between any two second vehicles based on the position data; Determining second vehicle speed characteristic data within a preset second time period; Second fused feature data of the second vehicle is generated according to the Euclidean distance matrix and the second vehicle speed feature data.

8. A vehicle speed prediction device, characterized in that: The device comprises: A first historical vehicle speed data acquisition module, used to acquire first historical vehicle speed data of a first vehicle; A first time window division module, used to divide the first historical vehicle speed data into first vehicle speed segments according to a plurality of preset time windows, and extract first vehicle speed characteristic data corresponding to each first vehicle speed segment; A first fusion feature data generating module, used for generating first fusion feature data of the first vehicle according to a plurality of first vehicle speed feature data; A vehicle speed prediction module is used to predict the vehicle speed of the first vehicle based on the first fused feature data.

9. An electronic device, characterized in that: The method comprises a processor, a memory and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the vehicle speed prediction method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the vehicle speed prediction method according to any one of claims 1 to 7 is implemented.