Vehicle behavior prediction method, device and electronic equipment

Abstract

The present invention provides a vehicle behavior prediction method, device and electronic device. The method comprises the following steps: building an LSTM-MLP model under a specified architecture, extracting input features required for training and testing through an open source data set NGSIM, the input features comprising training data and test data, the input features classifying each vehicle trajectory sequence into M categories through the open source data set NGSIM, and arranging the M-category vehicle trajectory sequences into a matrix form according to time to obtain preprocessed training data and test data, the training data and the test data both have corresponding data labels, training the LSTM-MLP model using the preprocessed training data to obtain a trained LSTM-MLP model, testing the trained LSTM-MLP model using the preprocessed test data to obtain a tested LSTM-MLP model, and the tested LSTM-MLP model is used to output a prediction result according to the input features to be identified of the vehicle. Its purpose is to solve the problem of low precision and low accuracy of the existing prediction method.

Description

Technical Field

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

[0002] The emergence of smart cars has improved the problems of traffic congestion, environmental pollution and waste of resources caused by traditional cars, and has attracted widespread attention at home and abroad. In recent decades, with the continuous development of artificial intelligence, smart car technology has also been increasingly advanced. However, in February 2016, Google's smart car collided with a bus on the side while changing lanes. This accident showed the shortcomings of smart cars in lane change prediction.

[0003] There are two main existing vehicle trajectory prediction methods:

[0004] (1) One is a trajectory prediction method based on RNN, but RNN has the problems of gradient vanishing and gradient exploding, so RNN is not suitable for the prediction of longer time series. Patent application number CN202010765063.1 discloses a lane change intention prediction method based on LSTM, which describes the prediction of vehicle lane change intention based on LSTM model alone, which can solve the gradient problem of RNN, and LSTM has a strong ability to learn and copy the historical trajectory of the vehicle, but the LSTM network cannot capture the dependency between multiple related historical trajectories. Therefore, the accuracy of trajectory prediction using LSTM alone is not high.

[0005] (2) Another method is based on convolutional neural network (CNN). Some studies have shown that CNN methods are better than some LSTM methods, but convolutional neural network (CNN) is a local feature extraction method, that is, close-range features. Moreover, this type of method has limited ability to process interactive information, and the accuracy of trajectory prediction is not high. Summary of the invention

[0006] In view of this, an object of the present invention is to provide a vehicle behavior prediction method, device and electronic device to solve the problem of low precision and accuracy of existing prediction methods.

[0007] The present invention solves the above technical problems by the following technical means:

[0008] In a first aspect, an embodiment of the present application provides a vehicle behavior prediction method, comprising:

[0009] Building an LSTM-MLP model under a specified architecture, wherein the LSTM-MLP model includes a long short-term memory network LSTM model and a multi-layer perceptron MLP model;

[0010] Extract input features required for training and testing through the open source data set NGSIM, where the input features include training data and test data;

[0011] The input feature divides each vehicle trajectory sequence into M categories through the open source data set NGSIM, and organizes the M categories of vehicle trajectory sequences into a matrix form according to time to obtain preprocessed training data and test data, wherein the training data and the test data both have corresponding data labels, and M is an integer greater than 1;

[0012] Using the preprocessed training data to train the LSTM-MLP model to obtain a trained LSTM-MLP model;

[0013] The trained LSTM-MLP model is tested using the preprocessed test data to obtain a tested LSTM-MLP model, wherein the tested LSTM-MLP model is used to output a prediction result according to input features to be identified of the vehicle, wherein the prediction result includes a driving mode of the vehicle, and the driving mode includes any one of going straight, changing lanes to the left, and changing lanes to the right.

[0014] In combination with the first aspect, in some optional implementations, building an LSTM-MLP model under a specified architecture includes:

[0015] Set up the LSTM neural network under the specified architecture, select the adam gradient optimization algorithm, the ReLU activation function, set the initial learning rate, initialize the weights, and set the gradient threshold and forgetting bias;

[0016] According to the characteristics of the LSTM neural network output, the number of neurons in the input layer of the MLP network is determined.

[0017] In combination with the first aspect, in some optional implementations, the LSTM model is used to output vehicle predicted trajectory information within a first specified time period.

[0018] In combination with the first aspect, in some optional implementations, the input features include the position of the vehicle, the vehicle speed, the vehicle acceleration, the vehicle heading angle and the position information of the lane in the historical time period, wherein the input vector input to the LSTM neural network is represented by X Train Indicates that the training output sequence corresponding to the input vector is denoted by Y train express.

[0019] In combination with the first aspect, in some optional embodiments, the X Train Each unit of includes 6 input features, and the 6 input features are:

[0020] Vehicle lateral position, longitudinal position, vehicle lateral velocity, longitudinal velocity, vehicle longitudinal acceleration and vehicle heading angle.

[0021] In combination with the first aspect, in some optional implementations, the LSTM-MLP model is trained using the preprocessed training data to obtain a trained LSTM-MLP model, including:

[0022] The preprocessed training data is used to train the LSTM model with the initialization parameters preset, and the training is stopped until the last training iteration ends or the preset gradient error is met, so as to obtain a trained LSTM-MLP model.

[0023] In combination with the first aspect, in some optional implementations, the method further includes:

[0024] The input feature to be identified is input into the tested LSTM-MLP model to obtain the prediction result output by the tested LSTM-MLP model according to the input feature to be identified.

[0025] In a second aspect, an embodiment of the present application further provides a vehicle behavior prediction device, comprising:

[0026] A building unit, building an LSTM-MLP model under a specified architecture, wherein the LSTM-MLP model includes a long short-term memory network LSTM model and a multi-layer perceptron MLP model;

[0027] An acquisition unit, extracting input features required for training and testing through the open source data set NGSIM, wherein the input features include training data and test data;

[0028] The processing unit, the input feature is used to classify each vehicle trajectory sequence into M categories through the open source data set NGSIM, and the vehicle trajectory sequences of the M categories are sorted into a matrix form according to time to obtain pre-processed training data and test data, the training data and the test data both have corresponding data labels, and M is an integer greater than 1;

[0029] A training unit, using the preprocessed training data to train the LSTM-MLP model to obtain a trained LSTM-MLP model;

[0030] A prediction unit tests the trained LSTM-MLP model using the preprocessed test data to obtain a tested LSTM-MLP model, wherein the tested LSTM-MLP model is used to output a prediction result according to input features to be identified of the vehicle, wherein the prediction result includes a driving mode of the vehicle, and the driving mode includes any one of going straight, changing lanes to the left, and changing lanes to the right.

[0031] In a third aspect, an embodiment of the present application further provides an electronic device, comprising a processor and a memory coupled to each other, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the electronic device executes the above method.

[0032] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium. When the computer program is executed on a computer, the computer executes the above method.

[0033] The present invention has the following advantages:

[0034] First, an LSTM-MLP model is built under a specified architecture. Secondly, an electronic device is used to extract input features required for training and testing through the open source data set NGSIM, where the input features include training data and test data. Then, the training data is divided into M categories for each vehicle trajectory sequence through the open source data set NGSIM, and the M categories of vehicle trajectory sequences are organized into a matrix form according to time to obtain preprocessed training data. Then, the preprocessed training data is used to train the LSTM-MLP model to obtain the trained LSTM-MLP model, thereby completing the training of the LSTM-MLP model.

[0035] Finally, the prediction data is divided into M categories for each vehicle trajectory sequence through the open source data set NGSIM, and the M category vehicle trajectory sequences are organized into a matrix form according to time to obtain preprocessed prediction data, and then the trained LSTM-MLP model is tested using the preprocessed test data to obtain a tested LSTM-MLP model. The tested LSTM-MLP model is used to output a prediction result according to the input features to be identified of the vehicle, and the prediction result includes the driving mode of the vehicle, and the driving mode includes any one of going straight, changing lanes to the left, and changing lanes to the right.

[0036] By predicting vehicle behavior through the LSTM-MLP model, the problems of low precision and accuracy of existing prediction methods can be solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The present invention can be further illustrated by the non-limiting embodiments given in the accompanying drawings:

[0038] Figure 1 It is a flow chart of a vehicle behavior prediction method provided by the present invention;

[0039] Figure 2is a block diagram of a vehicle behavior prediction device provided by the present invention;

[0040] Figure 3 It is a flow chart of LSTM model prediction in a vehicle behavior prediction method provided by the present invention;

[0041] Figure 4 It is a structural schematic diagram of an MLP model in a vehicle behavior prediction method provided by the implementation of the present invention.

[0042] Icon: 200 - building unit; 210 - acquisition unit; 220 - processing unit; 230 - training unit; 240 - prediction unit. DETAILED DESCRIPTION

[0043] The present application will be described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that in the drawings or descriptions, similar or identical parts use the same figure numbers, and the implementation methods not shown or described in the drawings are forms known to ordinary technicians in the relevant technical field. In the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0044] The embodiment of the present application provides an electronic device. The electronic device includes a processor and a memory coupled to each other, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the electronic device can execute corresponding steps in the following method for detecting vehicle path planning in a tunnel.

[0045] The electronic device may be, but is not limited to, a personal computer, a server, etc. The processor is electrically connected to the memory, and the processor is used to execute the computer program in the memory. Both the processor and the memory are existing hardware modules, so they are not described in detail here.

[0046] like Figure 1 As shown, the present application also provides a vehicle behavior prediction method, which can be applied to the above-mentioned electronic device, and each step of the method is executed or implemented by the electronic device. Among them, the vehicle behavior prediction method can include the following steps:

[0047] Step 100, building an LSTM-MLP model under a specified architecture, wherein the LSTM-MLP model includes a long short-term memory network LSTM model and a multi-layer perceptron MLP model;

[0048] Step 110, extracting input features required for training and testing through the open source data set NGSIM, wherein the input features include training data and test data;

[0049] Step 120, the input feature is used to classify each vehicle trajectory sequence into M categories through the open source data set NGSIM, and the vehicle trajectory sequences of the M categories are sorted into a matrix form according to time to obtain pre-processed training data and test data, wherein the training data and the test data both have corresponding data labels, and M is an integer greater than 1;

[0050] Step 130, training the LSTM-MLP model using the preprocessed training data to obtain a trained LSTM-MLP model;

[0051] Step 140, using the preprocessed test data to test the trained LSTM-MLP model to obtain a tested LSTM-MLP model, wherein the tested LSTM-MLP model is used to output a prediction result according to the input features to be identified of the vehicle, wherein the prediction result includes the driving mode of the vehicle, and the driving mode includes any one of going straight, changing lanes to the left, and changing lanes to the right.

[0052] The following will describe the various steps of the vehicle behavior prediction method in detail as follows:

[0053] In step 100, engineers can use electronic devices to build LSTM-MLP models under the tensorflow-keras architecture (an example of a specified architecture), and the LSTM-MLP model includes a long short-term memory network LSTM model and a multi-layer perceptron MLP model. In this embodiment, TensorFlow is an open source machine learning library for research and production. TensorFlow (open source software library) provides various APIs (Application Program Interface) for beginners and experts to develop in desktop, mobile, network and cloud environments. Keras is an advanced neural network API written in Python (computer programming language), which can run with TensorFlow, CNTK (Computational Network Toolkit, Microsoft cognitive tool set), or Theano (a Python library) as the backend.

[0054] When building the LSTM-MLP model, the most important thing is to set up the LSTM neural network under the specified architecture, and you can choose the adam (Adaptive Moment Estimation) gradient optimization algorithm to update the learning rate, and select the ReLU (Rectified Linear Unit) function as the activation function; set the initial learning rate, initialize the weights, and set the gradient threshold and forgetting bias to prevent gradient overflow. In this embodiment, the LSTM model is an end-to-end training model for outputting the vehicle prediction trajectory information within the first specified time period. Among them, the first specified time period can be flexibly set according to the actual situation. For example, the first specified time period can refer to 3 to 5 seconds in the future (i.e. after the current moment).

[0055] For example, training data or test data of a historical time period is input into the LSTM model. After the training data or test data is processed by the LSTM model, the vehicle predicted trajectory information, including vehicle position, vehicle speed, vehicle acceleration and vehicle heading angle, will be output.

[0056] According to the features extracted from the LSTM neural network output, the number of input layer neurons of the MLP network is determined. For example, if the LSTM neural network outputs 6 features, the number of input neurons of the MLP network is 6.

[0057] The MLP model is provided with a classification network, and the classification network is used to classify and predict the predicted trajectory output by the LSTM neural network. The classification network includes an input layer, a hidden layer and an output layer. The input layer is provided with 6 neurons, representing 6 input features to be predicted, including the vehicle's lateral position, longitudinal position, vehicle lateral speed, longitudinal speed, vehicle longitudinal acceleration and vehicle heading angle. The hidden layer has 1 layer, and the number of neurons in each layer can be adjusted according to the specific actual situation. For example, Figure 4 As shown, in this embodiment, the number of neurons in the hidden layer is 4, and the output layer has 3 neurons, which respectively represent the output of three classification results: going straight, changing lanes to the left, and changing lanes to the right.

[0058] In step 110, an electronic device is used to extract input features required for training and testing through the open source data set NGSIM, and the input features include training data and test data. The input features include the position, speed, acceleration, heading angle and lane position information of the vehicle in the historical time period. The training data is used to train the LSTM-MLP model, and the test data is the data required for predicting vehicle behavior in this embodiment.

[0059] In step 120, the input features are divided into M categories for each vehicle trajectory sequence through the open source dataset NGSIM, and the M categories of vehicle trajectory sequences are organized into a matrix form by time to obtain preprocessed training data and test data, wherein the training data and the test data both have corresponding data labels, and M is an integer greater than 1.

[0060] In this embodiment, a sequence-sequence mode is used for input and output, the training data and the test data are subjected to feature extraction, and then the training data and the test data are arranged in a matrix along the row direction of time and input into the LSTM neural network, and X is used to extract the training data and the test data. Train represents the input vector, and the training output sequence corresponding to the input vector is denoted by Y train express.

[0061] For example, the X Train is a 1-by-N cell array, where a cell is a sequence, which is actually the data of a segment of trajectory, and N is the number of segments of trajectory; Train Each unit of Y is a 6-by-n matrix, where 6 is the number of features and n is the time. The n of each unit can be different, and the actual meaning is represented by the different number of frames obtained. train is a 1-by-N class unit sequence, a vector, which is the vehicle test data, where N is the same as X Train The feature input and output data of the N segments of preprocessed vehicle data are stored in the X Train and the Y train A cell array where X Train There are 6 input features in each unit of , including vehicle lateral position, longitudinal position, vehicle lateral velocity, longitudinal velocity, vehicle longitudinal acceleration and vehicle heading angle. After preprocessing, the training input and output data of the LSTM neural network are obtained.

[0062] In step 130, the LSTM-MLP model is trained using the preprocessed training data to obtain a trained LSTM-MLP model. Figure 3As shown, the LSTM model with the initialization parameters pre-set is trained using the pre-processed training data, that is, the pre-processed training data is input into the LSTM network, and the LSTM network determines whether the gradient error reaches the requirement or the iterator reaches the maximum value. If the gradient error reaches the requirement, the next step is performed to determine whether the model training is completed. If it is completed, the prediction sequence is output, the error index is calculated, and the work is completed; if the iterator reaches the maximum value, or it is determined that the model training is not completed, it is re-determined whether the gradient error reaches the requirement or the iterator reaches the maximum value, until the last training iteration ends or the preset gradient error is met, the training is stopped to obtain the trained LSTM-MLP model. In this embodiment, the number of trainings can be adjusted according to the actual situation, such as selecting 50, that is, training 50 times to obtain the network parameters; the model batch size is selected as 100, that is, 100 groups of data in the set are selected each time for a round of iteration to update the network parameters.

[0063] The most important thing about the LSTM model is to perform gradient training to determine the weights. For example, during the training process, the LSTM model uses a gradient descent algorithm to continuously iterate and update the connection weights of artificial neuron nodes, that is, to find the gradient of the mathematical expression and obtain the appropriate weights according to the verification error rate requirements, so that the LSTM model reaches the optimal state, thereby obtaining a prediction result with high precision and accuracy.

[0064] In step 140, the trained LSTM-MLP model is tested using the preprocessed test data to obtain a tested LSTM-MLP model, and the tested LSTM-MLP model is used to output a prediction result based on the input features to be identified of the vehicle, and the prediction result includes the driving mode of the vehicle, and the driving mode includes any one of going straight, changing lanes to the left, and changing lanes to the right.

[0065] For example, the input features to be identified are input into the tested LSTM-MLP model, and the input features to be identified include the vehicle position, vehicle speed, vehicle acceleration, vehicle heading angle, and lane position information within a historical time period. Finally, the prediction result output by the tested LSTM-MLP model based on the input features to be identified is obtained.

[0066] In this embodiment, when performing the prediction step, the LSTM model first determines the predicted track information of the vehicle through the vehicle position, vehicle speed, vehicle acceleration and vehicle heading angle in the historical time period, and then the MLP model determines the prediction result through the predicted track information of the vehicle and the position information of the lane. The prediction result includes the driving mode of the vehicle, and the driving mode includes any one of straight driving, changing lanes to the left and changing lanes to the right.

[0067] The present application also provides a vehicle behavior prediction device, which includes at least one software function module stored in a storage module or fixed in an operating system in the form of software or firmware. The processor is used to execute executable modules stored in the storage module, such as the software function modules and computer programs included in the vehicle behavior prediction device.

[0068] like Figure 2 As shown, the vehicle behavior prediction device includes a construction unit 200, an acquisition unit 210, a processing unit 220, a training unit 230 and a prediction unit 240, and the functions of each unit may be as follows:

[0069] A building unit 200 is used to build an LSTM-MLP model under a specified architecture, wherein the LSTM-MLP model includes a long short-term memory network LSTM model and a multi-layer perceptron MLP model;

[0070] An acquisition unit 210 extracts input features required for training and testing through an open source data set NGSIM, wherein the input features include training data and test data;

[0071] Processing unit 220, the input feature is used to classify each vehicle trajectory sequence into M categories through the open source data set NGSIM, and the vehicle trajectory sequences of the M categories are sorted into a matrix form according to time to obtain pre-processed training data and test data, the training data and the test data both have corresponding data labels, and M is an integer greater than 1;

[0072] A training unit 230 is used to train the LSTM-MLP model using the preprocessed training data to obtain a trained LSTM-MLP model;

[0073] The prediction unit 240 tests the trained LSTM-MLP model using the preprocessed test data to obtain a tested LSTM-MLP model, wherein the tested LSTM-MLP model is used to output a prediction result according to the input features to be identified of the vehicle, wherein the prediction result includes the driving mode of the vehicle, and the driving mode includes any one of going straight, changing lanes to the left, and changing lanes to the right.

[0074] Optionally, the building unit 200 can be used to: set up an LSTM neural network under a specified architecture, select an adam gradient optimization algorithm, a ReLU activation function, set an initial learning rate, initialize weights, and set a gradient threshold and a forgetting bias; and determine the number of input layer neurons of the MLP network based on the features extracted from the output of the LSTM neural network.

[0075] Optionally, the acquisition unit 210 may be used to: the input features include the position of the vehicle, the vehicle speed, the vehicle acceleration, the vehicle heading angle and the position information of the lane in the historical time period, wherein the input vector input to the LSTM neural network is represented by X Train Indicates that the training output sequence corresponding to the input vector is denoted by Y train Said X Train Each unit of includes 6 input features, and the 6 input features are:

[0076] Vehicle lateral position, longitudinal position, vehicle lateral velocity, longitudinal velocity, vehicle longitudinal acceleration and vehicle heading angle.

[0077] The LSTM model is used to output vehicle predicted trajectory information within a first specified time period.

[0078] Optionally, the training unit 230 can be used to: use the preprocessed training data to train the LSTM model with pre-set initialization parameters, until the last training iteration ends or the preset gradient error is met, and then stop the training to obtain a trained LSTM-MLP model.

[0079] Optionally, the prediction unit 240 can be used to: input the input feature to be identified into the tested LSTM-MLP model to obtain the prediction result output by the tested LSTM-MLP model according to the input feature to be identified.

[0080] In this embodiment, when predicting the result, the LSTM model first determines the predicted track information of the vehicle through the vehicle position, vehicle speed, vehicle acceleration and vehicle heading angle in the historical time period, and then the MLP model determines the prediction result through the predicted track information of the vehicle and the position information of the lane. The prediction result includes the driving mode of the vehicle, and the driving mode includes any one of straight driving, changing lanes to the left and changing lanes to the right.

[0081] An embodiment of the present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed on a computer, the computer executes the vehicle behavior prediction method as described in the above embodiment.

[0082] Through the description of the above implementation methods, technical personnel in this field can clearly understand that the present application can be implemented by hardware, and can also be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, a control device, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0083] In summary, the embodiments of the present application provide a vehicle behavior prediction method, device and electronic device. In this solution, when it is necessary to predict vehicle behavior, first, an electronic device is used to build an LSTM-MLP model under the tensorflow-keras architecture, and the LSTM-MLP model includes a long short-term memory network LSTM model and a multi-layer perceptron MLP model. Secondly, the electronic device is used to extract the input features required for training and testing through the open source data set NGSIM, and the input features include training data and test data; the training data is then divided into M categories for each vehicle trajectory sequence through the open source data set NGSIM, and the M categories of vehicle trajectory sequences are organized into a matrix form according to time to obtain preprocessed training data, and then the preprocessed training data is used to train the LSTM-MLP model to obtain the trained LSTM-MLP model, that is, the training of the LSTM-MLP model is completed.

[0084] Finally, the prediction data is divided into M categories for each vehicle trajectory sequence through the open source data set NGSIM, and the M category vehicle trajectory sequences are organized into a matrix form according to time to obtain preprocessed prediction data, and then the trained LSTM-MLP model is tested using the preprocessed test data to obtain a tested LSTM-MLP model. The tested LSTM-MLP model is used to output a prediction result according to the input features to be identified of the vehicle, and the prediction result includes the driving mode of the vehicle, and the driving mode includes any one of going straight, changing lanes to the left, and changing lanes to the right.

[0085] In the embodiments provided in the present application, it should be understood that the disclosed devices, equipment and methods can also be implemented in other ways. The above-described device, device and method embodiments are merely schematic, for example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a part of a module, a program segment or a code, and a part of the module, program segment or code includes one or more executable instructions for implementing the specified logical function. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of the boxes in the block diagram and / or the flowchart can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. In addition, each functional module in each embodiment of the present application can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part.

[0086] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A vehicle behavior prediction method, characterized in that: include: Building an LSTM-MLP model under a specified architecture, wherein the LSTM-MLP model includes a long short-term memory network LSTM model and a multi-layer perceptron MLP model; Extract input features required for training and testing through the open source data set NGSIM, where the input features include training data and test data; The input feature divides each vehicle trajectory sequence into M categories through the open source data set NGSIM, and organizes the M categories of vehicle trajectory sequences into a matrix form according to time to obtain preprocessed training data and test data, wherein the training data and the test data both have corresponding data labels, and M is an integer greater than 1; Using the preprocessed training data to train the LSTM-MLP model to obtain a trained LSTM-MLP model; Testing the trained LSTM-MLP model using the preprocessed test data to obtain a tested LSTM-MLP model, wherein the tested LSTM-MLP model is used to output a prediction result according to the input features to be identified of the vehicle, wherein the prediction result includes a driving mode of the vehicle, and the driving mode includes any one of going straight, changing lanes to the left, and changing lanes to the right; When performing the prediction step, the LSTM model first determines the vehicle's predicted trajectory information through the vehicle's position, vehicle speed, vehicle acceleration and vehicle heading angle in the historical time period, and then the MLP model determines the prediction result through the vehicle's predicted trajectory information and lane position information.

2. The method according to claim 1, characterized in that Build an LSTM-MLP model under the specified architecture, including: Set up the LSTM neural network under the specified architecture, select the adam gradient optimization algorithm, the ReLU activation function, set the initial learning rate, initialize the weights, and set the gradient threshold and forgetting bias; According to the characteristics of the LSTM neural network output, the number of neurons in the input layer of the MLP network is determined.

3. The method according to claim 1, characterized in that The LSTM model is used to output vehicle predicted trajectory information within a first specified time period.

4. The method according to claim 1, characterized in that: The input features include the position, speed, acceleration, heading and lane information of the vehicle in the historical time period, wherein the input vector input to the LSTM neural network is represented by X Train Indicates that the training output sequence corresponding to the input vector is denoted by Y train express.

5. The method according to claim 4, characterized in that The X Train Each unit of includes 6 input features, and the 6 input features are: Vehicle lateral position, longitudinal position, vehicle lateral velocity, longitudinal velocity, vehicle longitudinal acceleration and vehicle heading angle.

6. The method according to claim 1, characterized in that The LSTM-MLP model is trained using the preprocessed training data to obtain a trained LSTM-MLP model, including: The preprocessed training data is used to train the LSTM model with the initialization parameters preset, and the training is stopped until the last training iteration ends or the preset gradient error is met, so as to obtain a trained LSTM-MLP model.

7. The method according to any one of claims 1, characterized in that The method further comprises: The input feature to be identified is input into the tested LSTM-MLP model to obtain the prediction result output by the tested LSTM-MLP model according to the input feature to be identified.

8. A vehicle behavior prediction device, characterized in that: include: A building unit, building an LSTM-MLP model under a specified architecture, wherein the LSTM-MLP model includes a long short-term memory network LSTM model and a multi-layer perceptron MLP model; An acquisition unit, extracting input features required for training and testing through the open source data set NGSIM, wherein the input features include training data and test data; The processing unit, the input feature is used to classify each vehicle trajectory sequence into M categories through the open source data set NGSIM, and the vehicle trajectory sequences of the M categories are sorted into a matrix form according to time to obtain pre-processed training data and test data, the training data and the test data both have corresponding data labels, and M is an integer greater than 1; A training unit, using the preprocessed training data to train the LSTM-MLP model to obtain a trained LSTM-MLP model; A prediction unit, using the preprocessed test data to test the trained LSTM-MLP model to obtain a tested LSTM-MLP model, wherein the tested LSTM-MLP model is used to output a prediction result according to the input features to be identified of the vehicle, wherein the prediction result includes a driving mode of the vehicle, and the driving mode includes any one of going straight, changing lanes to the left, and changing lanes to the right; When performing the prediction step, the LSTM model first determines the vehicle's predicted trajectory information through the vehicle's position, vehicle speed, vehicle acceleration and vehicle heading angle in the historical time period, and then the MLP model determines the prediction result through the vehicle's predicted trajectory information and lane position information.

9. An electronic device, characterized in that: The electronic device comprises a processor and a memory coupled to each other, wherein the memory stores a computer program. When the computer program is executed by the processor, the electronic device executes the 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 on a computer, the computer is enabled to execute the method according to any one of claims 1 to 7.

Images