Method for predicting motion state of manually-driven vehicle in mixed traffic environment

Abstract

The invention relates to a method for predicting the state of a manually-driven vehicle in a mixed traffic environment, and belongs to the field of intelligent traffic. The method comprises the steps that firstly, an open source vehicle trajectory data set is primarily processed, the standard deviation of speed data of the open source vehicle trajectory data set is calculated, sorting and splicing are carried out according to the standard deviation, and it is guaranteed that two pieces of data of an original trajectory do not excessively oscillate and better conform to the real situation; secondly, Kalman filtering is used for eliminating noise, caused by a sensor, of original data, and a more reliable data basis is provided; thirdly, considering the estimation error of the space clearance and the relative speed, improving on the basis of the IDM car-following model to obtain an IDMEST car-following model, and reflecting the behavior heterogeneity of the manually-driven vehicle; and finally, a deep learning algorithm is used to predict the state information of the manual driving vehicle between the intelligent network connection vehicles. According to the invention, the number and the positions of the manually driven vehicles in the mixed traffic environment and the speed difference and the acceleration difference between the manually driven vehicles and the rear intelligent network-connected vehicles can be predicted.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent transportation and relates to a method for using deep learning to predict the state of manually driven vehicles between networked autonomous driving vehicles in a mixed traffic environment. Background Art

[0002] The deep integration of advanced technologies in the fields of automobiles, transportation, information and communications has promoted the rapid development of intelligent transportation systems. In the next few decades, connected autonomous vehicles will be on the road together with human-driven vehicles, and the characteristics of traffic flow will change accordingly. Predicting the speed, acceleration and position status information of human-driven vehicles has significant benefits for transportation. Predicting speed, acceleration and position information can help vehicles in traffic to better cooperate and maintain stable vehicle spacing and speed; reduce speed fluctuations and unnecessary acceleration and deceleration between vehicles, thereby improving the overall stability of traffic; connected and autonomous vehicles (CAV) can more effectively respond to traffic changes such as traffic congestion and waiting at intersections. Prediction information helps drivers or autonomous driving systems identify potential dangerous situations in advance, such as sudden deceleration or emergency braking of the vehicle in front. By responding in time, dangerous situations such as rear-end collisions can be avoided or reduced, improving traffic safety. Predicting the status information of human-driven vehicles has multiple benefits for traffic, such as improving stability, optimizing efficiency, enhancing safety, reducing energy consumption and emissions, supporting collaborative control, improving driver comfort, and promoting the development of intelligent transportation systems. These benefits will help improve the overall performance and efficiency of vehicles, making a positive contribution to safer and smoother road traffic.

[0003] In this context, the state prediction of human-driven vehicles has many significances, such as improving traffic safety, improving traffic efficiency, promoting environmental protection and energy saving, improving driving experience, and promoting the development of intelligent transportation systems. The research on the operation state prediction method of human-driven vehicles (HDV) in mixed traffic environments mainly focuses on integrating multi-source traffic data and using advanced information technology and data analysis technology to achieve accurate prediction of vehicle speed, acceleration, position and other state information. Although significant progress has been made, there are still some shortcomings, which need further exploration and improvement in data collection and processing, traffic flow modeling, and prediction models. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method for predicting the operating status of manually driven vehicles in a mixed traffic environment. By constructing a model that describes the heterogeneity of human-driven vehicles, a traffic simulation environment is established and the real speed curve is loaded into the leading vehicle. A deep model is used for information prediction, thereby enabling CAV to obtain more vehicle status information in mixed traffic, supporting the needs of safe, efficient, energy-saving and comfortable driving of networked autonomous driving vehicles, and achieving the optimal development goal of the transportation system.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for predicting the motion state of a manually driven vehicle in a mixed traffic environment, the method comprising:

[0007] Construct a mixed traffic environment simulation model in which intelligent connected vehicles and manually driven vehicles drive stably on the road. Intelligent connected vehicles obtain information about themselves and neighboring vehicles through on-board sensors.

[0008] In the mixed traffic environment simulation model, the IDM_SET model which introduces the spatial gap and relative speed estimation error is used to control the following vehicle to improve the heterogeneity of the artificially driven vehicles in the simulation model.

[0009] In the mixed traffic environment simulation model, the leading intelligent connected vehicle travels according to the input driving speed data, and the following vehicles are controlled by the IDM_SET model to follow the driving; the leading intelligent connected vehicle obtains information about itself and the adjacent vehicles behind it, and the following intelligent connected vehicles that are not adjacent to the leading intelligent connected vehicle obtain information about itself and the adjacent vehicles in front of it, and use this vehicle information to train the state prediction model of the manually driven vehicle, so that the state prediction model of the manually driven vehicle can predict the information of the manually driven vehicle between the leading intelligent connected vehicle and the following intelligent connected vehicle that cannot be perceived;

[0010] The trained manually driven vehicle state prediction model is deployed to the cloud server or intelligent connected vehicle. The intelligent connected vehicle in the real mixed traffic flow will input its own vehicle information and the perceived manually driven vehicle information into the manually driven vehicle state prediction model to realize the prediction of the motion state of unknown manually driven vehicles between non-adjacent intelligent connected vehicles in the mixed traffic flow.

[0011] Furthermore, the IDM_SET model is expressed as:

[0012]

[0013] In the formula, represents the acceleration of the following vehicle, a represents the maximum acceleration, v0 represents the expected speed; s *represents the safety gap, s0 represents the minimum distance between the two vehicles, T represents the time interval with the preceding vehicle, and b represents the expected deceleration; s n , Δv n They represent the gap and speed difference between vehicle n and the preceding vehicle, v n represents the speed of vehicle n; They represent the spatial gap and velocity difference with estimation error at time t, represents the speed of vehicle n with estimation error at time t, ξ and σ represent the standard deviation of gap and speed respectively; ω l (t) and ω s (t) are used to describe the time evolution of relative velocity and spatial gap error, Δt represents the time step, Ψ represents the correlation time, and η i (t) represents a random number.

[0014] Furthermore, the driving speed data of the leading intelligent networked vehicle is obtained by extracting multiple vehicle speed trajectories from a public data set and splicing them. Specifically, for the extracted multiple vehicle speed trajectories, the speed standard deviation of each trajectory is first calculated, and the vehicle speed trajectories are sorted according to the size of the standard deviation; a number of vehicle speed trajectories with similar standard deviations after sorting are selected, and Kalman filtering is performed on the selected vehicle speed trajectories; for the filtered vehicle speed trajectories, the vehicle speed trajectories with even numbers are reversed, and then connected in sequence in the sorting order to form a vehicle speed trajectory as the driving speed data of the leading intelligent networked vehicle.

[0015] Furthermore, the artificially driven vehicle prediction model includes a CNN network, a BGRU network and an output layer connected in sequence. The CNN network is used to extract the features of the input vehicle information; the extracted features enter the BGRU network to capture the bidirectional dependencies of the sequence data and improve the model's ability to understand the sequence data; the output layer converts the output of the BGRU network into a predicted value of the state of the artificially driven vehicle. The output layer has four-dimensional output nodes, each of which corresponds to the prediction results of the four dimensions output by the artificially driven vehicle state prediction model.

[0016] Furthermore, the data output by the manually driven vehicle prediction model includes the number of manually driven vehicles that cannot perceive vehicle information between two non-adjacent intelligent connected vehicles, and the speed difference, acceleration difference and spatial distance between each manually driven vehicle and the intelligent connected vehicle behind it.

[0017] The beneficial effects of the present invention are as follows: the present invention constructs an IDM_SET model that describes the heterogeneity of human-driven vehicles, establishes a traffic simulation environment and loads the real speed curve into the leading vehicle in the simulation environment, and uses a deep learning network model to predict HDV information, thereby enabling CAV to obtain more vehicle status information in mixed traffic, supporting the needs of safe, efficient, energy-saving and comfortable driving of networked autonomous driving vehicles, so as to achieve the optimization development goal of the transportation system.

[0018] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0020] Figure 1 It is a traffic model based on SUMO;

[0021] Figure 2 is the speed curve of the leading vehicle;

[0022] Figure 3 This is a schematic diagram of the state prediction scenario;

[0023] Figure 4 This is a deep learning network model diagram;

[0024] Figure 5 It is the speed curve of IDM_EST model. DETAILED DESCRIPTION

[0025] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0026] An embodiment of the present invention provides a method for predicting the motion state of a manually driven vehicle in a mixed traffic environment. The method is mainly divided into two parts, namely, a traffic environment based on SUMO and training of a prediction model, wherein the simulation environment is controlled through a Traci interface. First, the constructed prediction model is trained offline using an existing public data set, and then deployed to the cloud or an intelligent connected vehicle for online operation, wherein the intelligent connected vehicle can obtain the status information of the intelligent connected vehicle in front of it through communication technology.

[0027] This example is based on the following assumptions:

[0028] (1) Within the communication range, the status of the CAV can be transmitted through messages between the on-board communication equipment and the road test communication unit or 5G base station.

[0029] (2) CAV vehicles can obtain the speed and acceleration of the vehicles in front and behind them through on-board sensors.

[0030] 1. Traffic environment based on SUMO

[0031] Consider a typical mixed traffic scenario. Use SUMO to construct a one-way road. Add 12 vehicles to form a traffic environment. The maximum vehicle speed is 120 km / h. CAV and HDV are driving steadily on the road. Figure 1 The speed data of the leading vehicle in the traffic is the vehicle speed splicing trajectory extracted from the NGSIM dataset. All vehicle trajectory data on the US-101 road in the time period of "0750am-0805am" are selected, and the data are preprocessed, including:

[0032] First, the standard deviation of each vehicle is calculated using the STD calculation formula below, and the vehicles are sorted according to this standard deviation.

[0033]

[0034] Among them, v t is the speed of the i-th vehicle at time frame t, is the average speed of vehicle i, and n is the total number of data.

[0035] Then, ten vehicle trajectory data with similar STD after sorting are selected, and the vehicle trajectories are reconstructed according to the alternating positive and reverse order splicing logic, that is, the trajectory data of the first vehicle is kept in positive order, and the trajectory data of the second vehicle is arranged in reverse order, and the subsequent vehicle trajectories are arranged alternately in positive order and reverse order. Among them, the conditions for connecting two vehicle trajectories are that the STD error of the two vehicles is small (that is, the two connected trajectories must be adjacent in the sorting), and the absolute value of the difference between the acceleration of the front vehicle at the end and the acceleration of the rear vehicle at the beginning is less than a certain threshold (set to 0.5m / s in this embodiment). 2). For these ten trajectories, Kalman filtering is needed first. Filtering can improve the accuracy of model prediction. Figure 2 After connecting these ten trajectories, the resulting trajectory is about 6000 frames with a frequency of 10 Hz, which is injected into the SUMO model through the Traci interface as the trajectory of the leading vehicle.

[0036] Among them, the state equation of Kalman filter is expressed as:

[0037] x k =F k x k-1 +B k u k +w k

[0038] The observation equation is expressed as:

[0039] z k =H k x k +v k

[0040] In the formula, x k is the state at the kth moment, F k is the state transfer matrix, B k is the control input matrix, u k is the control input; w k is the process noise, which is a Gaussian white noise, w k ~N(0,Q),Q=1×10 -5 .z k is the observation at the kth moment, H k is the observation matrix; v k is the observation noise, which is a Gaussian white noise, v k ~N(0,R), R=0.1.

[0041] In the prediction step, the state and covariance estimates are expressed as:

[0042]

[0043] Where F is the identity matrix, and there is no control input, that is, B k and u k is a 0 matrix, so we can deduce:

[0044]

[0045] P k|k-1 =P k-1|k-1 +Q

[0046] In the update step, the state estimate is adjusted according to the new observations:

[0047]

[0048] P k|k =(1-K k H k ) k|k -1

[0049] Among them, H k is the identity matrix.

[0050] An improved vehicle following model, denoted as IDM_EST, is constructed. This model introduces the estimation errors of spatial gap and relative speed based on the IDM controller to more accurately reflect the operating status of HDVs in mixed traffic environments. The trajectory of the leading vehicle in the traffic flow adopts the real vehicle trajectory of the data set, and the following vehicle is controlled using IDM_EST. Two non-adjacent vehicles are randomly selected as CAVs. The HDVs before and after the CAV can be obtained by the on-board sensors to predict the status information of the HDV that cannot be obtained between the two CAVs.

[0051] In IDM_EST, the vehicle acceleration is expressed as:

[0052]

[0053] Where a is the maximum acceleration, v0 is the desired velocity, s n =x n-1 -x n -l n-1 , Δv n =v n -v n-1 are the gap and speed difference between the current vehicle n and the preceding vehicle respectively. n represents the current position of vehicle n, l n Indicates the length of the current vehicle n. s * For safety clearance:

[0054]

[0055] Among them, s0 is the minimum distance between two vehicles, T is the time interval with the preceding vehicle, and b is the expected deceleration. and are the spatial gap and velocity difference with estimation error at time t, is the speed of vehicle n with estimation error at time t. Introducing the errors in estimating the gaps and relative speeds between vehicles in the traffic flow can better reflect the heterogeneity of HDVs. These estimation errors are expressed as follows:

[0056]

[0057] Where ξ and σ are the standard deviations of gap and velocity, respectively, and ω l (t) and ω s (t) describe the time evolution of relative velocity and spatial gap error, ω l (t) and ω s (t) can be expressed by the following formula:

[0058]

[0059] Where Δt is the time step, Ψ is the correlation time, η i (t) is a random number that follows a uniform distribution, has an expected value of zero, and has a unit variance. The following Table 1 shows the specific values ​​of each parameter in the IDM_SET model:

[0060] Table 1

[0061]

[0062]

[0063] In order to further intuitively illustrate the effectiveness of the heterogeneity of the proposed IDM_EST model, the model is tested. Specifically, all vehicles run at an initial speed of 15 m / s in a steady state, and their acceleration fluctuations are:

[0064]

[0065] Its speed fluctuation curve is as follows Figure 5 As shown, the speed curves of each HDV can be clearly distinguished, that is, the IDM_EST model can fully reflect the heterogeneity of HDVs, thereby simulating the traffic shock of HDV vehicles.

[0066] 2. Build a BGRU-based HDV state prediction model to predict the state of unknown HDV

[0067] The model includes CNN, BGRU and output layer. CNN is used to extract features from the front and rear vehicle information obtained by CAV, and the extracted features are input into BGRU for sequence prediction. The output layer converts the output of BGRU into the final classification or regression result. By analyzing the speed difference, acceleration difference and spatial position difference between the leading vehicle and the following vehicles, the BGRU model can predict the number and position of HDVs between CAVs, as well as the speed difference and acceleration difference between CAV and the HDVs in front.

[0068] The training process of the model includes: in the SUMO simulation model, two non-adjacent vehicles in the driving fleet are randomly selected as CAVs and establish communication to share the vehicle status information of the two. With the help of the on-board sensors, the CAV can obtain the speed and position information of the vehicles in front and behind it. For example, during the model training period, the vehicle numbered 5 is selected as the CAV that can communicate with the lead vehicle. With the help of the on-board sensors, the speed and position information of the vehicles numbered 1 and 4 between the two CAVs can be obtained. The information of the vehicles numbered 2 and 3 is in an unobtainable state. Figure 3 As shown. Therefore, the model training input state can be expressed as:

[0069]

[0070] Among them, T s is the length of the time series.

[0071] Table 2 shows the relevant training parameters of the HDV state prediction model:

[0072] Table 2

[0073]

[0074]

[0075] The traffic data is extracted and transferred to the HDV state prediction model through the Traci interface. Figure 4 In this model, the input data first passes through three convolutional layers to extract features from the information of vehicles No. 0, 1, 4, and 5. The three convolutional layers extract local spatiotemporal features respectively. The number of filters in each convolutional layer is 32, 64, and 128 respectively. Each layer uses Tanh activation function, batch normalization, and Dropout regularization. The extracted features enter the BGRU to capture the bidirectional dependency of sequence data and improve the model's ability to understand sequence data.

[0076] Among them, the calculation of GRU's update and reset gates is as follows:

[0077] z t =σ(W z ·[h t-1 ,x t ]+b z )

[0078] r t =σ(W r ·[h t-1 ,x t ]+b r )

[0079] In the formula, h t-1 is the hidden state of the previous time step, xt is the input of the current time step, W z , W r 、b z 、b r are the model parameters used to calculate the weight matrix and bias term of the update gate and reset gate. The parameters are continuously optimized during the model training process and determine how the model updates its internal state at each time step. σ(·) is the sigmoid activation function.

[0080] The update equation for the candidate hidden state using GRU is:

[0081]

[0082] Among them, W h 、b h are model parameters, is the candidate hidden state, h t is the hidden state at the current time step.

[0083] According to the above GRU model, BGRU can be derived. For the forward GRU, it is expressed as:

[0084]

[0085] For the reverse GRU, it is expressed as:

[0086]

[0087] in, is the hidden state of the forward GRU, is the hidden state of the reverse GRU.

[0088] The output of BGRU is concatenated from the hidden states of the forward and reverse GRUs:

[0089]

[0090] Finally, the output of BGRU is converted into the final classification or regression result through the output layer with four-dimensional output nodes. The four-dimensional output nodes correspond to the number of unknown HDVs, and the speed difference, acceleration difference and spatial distance between each unknown HDV and the rear CAV. Figure 3 The vehicle information collected by the two CAVs in the prediction model is the number of unknown HDVs between vehicles No. 5 and No. 0 (corresponding to Figure 3That is, the position difference, speed difference and acceleration difference between the vehicle 0, 1, 4 and 5 and the unknown HDVs in front of it. For example, during the model operation, the input is the structured data of vehicles 0, 1, 4 and 5 at the historical moment 3s (i.e. 30 frames) The number and status of unknown HDVs between the two CAVs output at this moment are shown in Table 3:

[0091] Table 3

[0092]

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the motion state of a manually driven vehicle in a mixed traffic environment, wherein the mixed traffic environment includes intelligent networked vehicles and manually driven vehicles, characterized in that: The method includes: Construct a mixed traffic environment simulation model in which intelligent connected vehicles and manually driven vehicles drive stably on the road. Intelligent connected vehicles obtain information about themselves and neighboring vehicles through on-board sensors. In the mixed traffic environment simulation model, the IDM_SET model which introduces the spatial gap and relative speed estimation error is used to control the following vehicle to improve the heterogeneity of the artificially driven vehicles in the simulation model. In the mixed traffic environment simulation model, the leading intelligent connected vehicle travels according to the input driving speed data, and the following vehicles are controlled by the IDM_SET model to follow; the leading intelligent connected vehicle obtains information about itself and the adjacent vehicles behind it, and the following intelligent connected vehicles that are not adjacent to the leading intelligent connected vehicle obtain information about itself and the adjacent vehicles in front of it, and use this vehicle information to train the state prediction model of the manually driven vehicle, so that the state prediction model of the manually driven vehicle can predict the information of the manually driven vehicles that are not perceived between the leading intelligent connected vehicle and the following intelligent connected vehicle; The trained manually driven vehicle state prediction model is deployed to the cloud server or intelligent connected vehicle. The intelligent connected vehicle in the real mixed traffic flow will input its own vehicle information and the perceived manually driven vehicle information into the manually driven vehicle state prediction model to realize the prediction of the motion state of unknown manually driven vehicles between non-adjacent intelligent connected vehicles in the mixed traffic flow.

2. The method according to claim 1, characterized in that: The IDM_SET model is represented as: In the formula, represents the acceleration of the following vehicle, a represents the maximum acceleration, v0 represents the expected speed; s * represents the safety gap, s0 represents the minimum distance between the two vehicles, T represents the time interval with the preceding vehicle, and b represents the expected deceleration; s n , Δv n They represent the gap and speed difference between vehicle n and the preceding vehicle, v n represents the speed of vehicle n; and They represent the spatial gap and velocity difference with estimation error at time t, represents the speed of vehicle n with estimation error at time t, ξ and σ represent the standard deviation of gap and speed respectively; ω l (t) and ω s (t) are used to describe the time evolution of relative velocity and spatial gap error, Δt represents the time step, Ψ represents the correlation time, η i (t) represents a random number.

3. The method according to claim 1, characterized in that In the mixed traffic environment simulation model, the driving speed data of the leading intelligent connected vehicle is obtained by extracting and splicing multiple vehicle speed trajectories from a public data set; wherein, for the multiple extracted vehicle speed trajectories, the speed standard deviation of each trajectory is first calculated, and the vehicle speed trajectories are sorted according to the size of the standard deviation; a number of vehicle speed trajectories with similar standard deviations after sorting are selected, and Kalman filtering is performed on the selected vehicle speed trajectories; for the filtered vehicle speed trajectories, the vehicle speed trajectories with even numbers are reversed, and then connected in sequence in the sorting order to form a vehicle speed trajectory as the driving speed data of the leading intelligent connected vehicle.

4. The method according to claim 1, characterized in that: The manually driven vehicle prediction model includes a CNN network, a BGRU network and an output layer connected in sequence; the CNN network is used to extract features of input vehicle information; the extracted features enter the BGRU network to capture the bidirectional dependency of sequence data and improve the model's ability to understand sequence data; the output layer converts the output of the BGRU network into a predicted value of the state of the manually driven vehicle.

5. The method according to claim 4, characterized in that The output layer has four-dimensional output nodes, and each output node corresponds to the prediction results of four dimensions output by the artificial driving vehicle state prediction model.

6. The method according to claim 1, 4 or 5, characterized in that: The data output by the manually driven vehicle state prediction model includes the number of manually driven vehicles that cannot perceive vehicle information between two non-adjacent intelligent connected vehicles, as well as the speed difference, acceleration difference and spatial distance between each manually driven vehicle and the intelligent connected vehicle behind it.

Images