Expressway short-time vehicle speed prediction method considering dynamic traffic scene
By considering dynamic traffic scenarios and key scenario vehicles in vehicle speed prediction, and using DBN to build a short-term vehicle speed prediction model, the problem of low prediction accuracy in the prior art is solved, and more accurate vehicle speed prediction is achieved.
Patent Information
- Application Number
- CN202410947726.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing vehicle speed prediction technology takes into account the driving conditions of the target vehicle itself or the impact of the leading vehicle on its speed, and ignores the impact of the traffic scene around the target vehicle, resulting in low prediction accuracy.
By obtaining vehicle trajectory data around the target vehicle, dividing different traffic scenarios, defining key scenario vehicles, and using Deep Confidence Network (DBN) to build and train a short-time vehicle speed prediction model, considering the relationship between the target vehicle and the key scenario vehicles to perform speed prediction.
It improves the accuracy and pertinence of vehicle speed prediction, makes the prediction results closer to the real road environment, and enhances the accuracy and practicality of the prediction.
Smart Images

Figure CN118968783A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a vehicle speed prediction technology, in particular to a method for predicting short-term vehicle speed on an expressway taking dynamic traffic scenarios into consideration. Background Art
[0002] With the development of connected autonomous driving technology, micro-vehicle control has become possible. By predicting the vehicle's future speed, the intelligent driving vehicle control strategy can be optimized and the driving plan can be adjusted in advance. As a research hotspot in the field of intelligent transportation and intelligent driving, vehicle speed prediction is widely used in vehicle collision warning, driving path planning and advanced driving assistance.
[0003] Existing vehicle speed prediction technologies mostly consider the target vehicle's own driving conditions or the impact of the leading vehicle on its speed, and use modeling, statistical analysis or intelligent prediction algorithms to predict vehicle speed, but rarely consider the relationship between the traffic scene and speed around the target vehicle. However, the traffic scene changes dynamically during the driving of the target vehicle, and the driver will pay attention to the surrounding key scene vehicles and adjust the current driving speed according to the behavior of the key scene vehicles. Only considering the target vehicle's own driving conditions or the leading vehicle for speed prediction faces the problem of low speed prediction accuracy. To solve this problem, we provide a method for predicting short-term vehicle speed on expressways that considers dynamic traffic scenes. Summary of the invention
[0004] The purpose of the present invention is to provide a method for predicting short-term vehicle speed on expressways taking into account dynamic traffic scenarios, extract key scene vehicles based on real-time traffic scenarios around the target vehicle, and accurately restore the source of vehicle speed stimulation. According to the speed relationship between key scene vehicles and target vehicles, a target vehicle speed prediction model input is constructed, and the prediction model is solved using a deep belief network DBN to improve the prediction accuracy of the target vehicle speed.
[0005] The technical problem solved by the present invention is that the existing vehicle speed prediction technology mostly considers the driving conditions of the target vehicle itself or the influence of the leading vehicle on its speed, and ignores the influencing factors such as the relationship between the traffic scene around the target vehicle and the speed, thus resulting in the problem of low prediction accuracy.
[0006] The present invention can be implemented by the following technical solution: A method for predicting short-term vehicle speed on an expressway considering dynamic traffic scenarios comprises the following steps:
[0007] Step 1: Based on the road traffic flow monitoring equipment, the vehicle trajectory data on the corresponding road section is obtained and a vehicle trajectory database is generated, wherein the vehicle trajectory data includes the vehicle position and the vehicle speed;
[0008] Step 2: Divide the surrounding road area of the target vehicle into the area in front of the left lane, the area in front of the current lane, and the area in front of the right lane, construct an area set A = {fl, f, fr}, and assign a state attribute S to each area;
[0009] Step 3: Distinguish the four traffic scenarios faced by the target vehicle based on the different state attribute combinations of each area;
[0010] Step 4: Define the key scene vehicles that the target vehicle driver needs to pay attention to for different traffic scenarios;
[0011] Step 5: Based on the DBN deep belief network, the target vehicle and key scene vehicle data within a certain period of time are used as input, and the instantaneous speed of the target vehicle in the future time period is used as output to build and train a short-term vehicle speed prediction model;
[0012] Step 6: Use the trained short-term vehicle speed prediction model to make real-time predictions on the short-term vehicle speed of the target vehicle.
[0013] A further technical improvement of the present invention is that the state attribute represents the occupancy of the corresponding area. When there is a car in the area, the state attribute S=1; otherwise, S=0.
[0014] A further technical improvement of the present invention is that the corresponding state attributes of different areas in the four traffic scenes and the corresponding relationship between the vehicles in the key scenes include:
[0015] Traffic scenario 1: The area in front of the current lane is occupied, and the area in front of the left lane and the area in front of the right lane are not occupied, that is, S fl =0, S f =1, S fr =0, the key scene vehicle is the leading vehicle n+1;
[0016] Traffic scenario 2: The area in front of the current lane and the area in front of the left lane are occupied, and the area in front of the right lane is not occupied, that is, S fl =1, S f =1, S fr =0; the key scene vehicles are the leading vehicle n+1 and the left front vehicle n+2;
[0017] Traffic scenario 3: The area in front of the current lane and the area in front of the right lane are occupied, and the area in front of the left lane is not occupied, that is, S fl =0, S f =1, S fr =1; the key scene vehicles are the leading vehicle n+1 and the right front vehicle n+3;
[0018] Traffic scenario 4: The area in front of the current lane, the area in front of the left lane, and the area in front of the right lane are all occupied, that is, S fl=1, S f =1, S fr =1; the key scene vehicles are the leading vehicle n+1, the left front vehicle n+2 and the right front vehicle n+3.
[0019] Among them, S fl ,S f and S fr They are the occupancy status of the area in front of the left lane, the area in front of the current lane, and the area in front of the right lane.
[0020] A further technical improvement of the present invention is that the input of the short-term vehicle speed prediction model is specifically: H = {h t-τ ,h t -τ+1 ,...,h t}, where τ is the driver’s reaction time, h t It is the key scene vehicle data at time t.
[0021] A further technical improvement of the present invention is that: in different traffic scenarios, h t The at least includes the longitudinal distance difference, relative speed and relative acceleration between the target vehicle and the key scene vehicle;
[0022] For the case where there are critical scene vehicles on both sides of the target vehicle’s current lane, h t It also includes the lateral distance difference between the target vehicle and the key scene vehicle.
[0023] A further technical improvement of the present invention is that the output of the short-term vehicle speed prediction model is specifically: in, is the speed of the target vehicle at time t, and T is the prediction step length.
[0024] A further technical improvement of the present invention is that the training step of the short-term vehicle speed prediction model includes:
[0025] Randomly select a set of data and use the minimum and maximum normalization method to preprocess the model input and output data;
[0026] Calculate the neuron states of the hidden layer and the neuron states of the visible layer;
[0027] Repeat the above steps until all data are selected;
[0028] Train the first layer of RBM until the maximum number of iterations is reached;
[0029] The output of the first layer of trained RBM is used as the input of the next layer of RBM, and the process is repeated until all RBM layers are trained.
[0030] Compared with the prior art, the present invention has the following beneficial effects:
[0031] 1. The present invention divides the traffic scenes existing in the surrounding area where the target vehicle is traveling based on the spatial relative position of the target vehicle and the surrounding scene vehicles, thereby clarifying the state attributes of the corresponding areas in different traffic scenes and the relationship between the key scene vehicles and the target vehicle, laying a foundation for the subsequent data input for constructing a short-term speed prediction model. In addition, by clarifying the vehicle relationship under different traffic scenes, the accuracy and pertinence of subsequent predictions can be improved according to the distinction of scenes.
[0032] 2. A vehicle speed prediction model considering dynamic traffic scenarios and key scenario vehicles was constructed. This model is closer to the actual road environment and can predict the short-term vehicle speed in different scenarios based on historical vehicle trajectory data, making the prediction results closer to reality. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.
[0034] Figure 1 It is a schematic diagram of the process of the present invention;
[0035] Figure 2 A schematic diagram of the area division around the target vehicle of the present invention;
[0036] Figure 3 Schematic diagrams of different traffic scenarios of the present invention;
[0037] Figure 4 Schematic diagram of the DBN deep belief network structure of the present invention. DETAILED DESCRIPTION
[0038] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.
[0039] See also Figure 1 As shown, a method for predicting short-term vehicle speed on an expressway considering dynamic traffic scenarios mainly includes the following steps:
[0040] Step 1: Obtain the trajectory data of the road section vehicle
[0041] Based on road-side traffic flow monitoring equipment such as radar, video surveillance equipment, and integrated radar and vision devices, image recognition and tracking technology are used to identify traffic flow parameters such as the position and speed of vehicles on the road. Through long-term observation, a vehicle trajectory database is obtained.
[0042] Step 2: Divide the area around the target vehicle
[0043] According to the positional relationship between the target vehicle and the leading vehicle, the road around the target vehicle is divided into three areas using the grid method: the area in front of the left lane, the area in front of the current lane, and the area in front of the right lane. Figure 2 As shown. Then the area set around the target vehicle A = {fl, f, fr}, where fl, f, fr are the area in front of the left lane, the area in front of the current lane, and the area in front of the right lane, respectively.
[0044] Each area has a state attribute S. When there is a car in the area, the state attribute is S = 1; otherwise, S = 0. Among them, n is the target vehicle and n+1 is the leading vehicle.
[0045] Step 3: Divide traffic scenarios
[0046] like Figure 3 As shown in the figure, according to the state attributes of the area around the target vehicle, the traffic scene is divided into four categories, n+2 and n+3 are the front vehicles on the left lane and the front vehicles on the right lane of the target vehicle, respectively, S fl ,S f and S fr They are the occupancy status of the area in front of the left lane, the area in front of the current lane, and the area in front of the right lane;
[0047] Traffic scenario 1: The area in front of the current lane is occupied, and the area in front of the left lane and the area in front of the right lane are not occupied, that is, S fl =0, S f =1, S fr =0;
[0048] Traffic scenario 2: The area in front of the current lane and the area in front of the left lane are occupied, and the area in front of the right lane is not occupied, that is, S fl =1, S f =1, S fr =0;
[0049] Traffic scenario 3: The area in front of the current lane and the area in front of the right lane are occupied, and the area in front of the left lane is not occupied, that is, S fl =0, S f =1, S fr =1;
[0050] Traffic scenario 4: The area in front of the current lane, the area in front of the left lane, and the area in front of the right lane are all occupied, that is, S fl =1, S f =1, S fr =1.
[0051] Step 4: Select key scene vehicles in different traffic scenarios
[0052] For traffic scenario 1, the driver of target vehicle n mainly focuses on the leading vehicle, that is, the key scenario vehicle is the leading vehicle;
[0053] For traffic scenario 2, the driver of target vehicle n needs to pay attention to the leading vehicle as well as the side vehicle to avoid the risk of collision caused by the side vehicle changing lanes. Therefore, the key scene vehicles in traffic scenario 2 are the leading vehicle and the left front vehicle.
[0054] Similarly, similar to traffic scenario 2, for traffic scenario 3, the key scene vehicles are the leading vehicle and the vehicle in front on the right;
[0055] For traffic scenario four, the key scenario vehicles are the leading vehicle, the left front vehicle, and the right front vehicle.
[0056] Step 5: Build and train a short-term vehicle speed prediction model
[0057] (1) Constructing model input
[0058] Based on the vehicle historical trajectory database, the relevant traffic data of the target vehicle and key scene vehicles within a certain period of time are selected as input. Take the target vehicle n as an example to illustrate:
[0059] Based on the target vehicle n and the trajectory data of surrounding vehicles, the relevant traffic data in the time period from t-τ to t is selected, that is: H = {h t-τ ,h t-τ+1 ,...,h t}, where τ is the driver’s reaction time, h t is the key scene vehicle data at time t.
[0060] According to different traffic scene types, different key scene vehicle data are matched as input:
[0061] ①For traffic scenario 1, i.e. S fl =0, S f =1, S fr = 0,
[0062]
[0063] ②For traffic scenario 2, namely S fl =1, S f =1, S fr = 0,
[0064]
[0065] ③For traffic scenario three, namely S fl =0, S f =1, S fr =1,
[0066]
[0067] ④For traffic scenario 4, namely S fl =1, S f =1, S fr =1,
[0068]
[0069] in, They represent the longitudinal distance difference, relative speed and relative acceleration of the target vehicle n and the leading vehicle n+1, respectively, that is,
[0070] Similarly, represent the longitudinal distance difference, relative speed, relative acceleration and lateral distance difference between the target vehicle n and the vehicle n+2 in front of the left lane, respectively, and
[0071] They respectively represent the longitudinal distance difference, relative speed, relative acceleration and lateral distance difference between the target vehicle n and the vehicle n+3 in front of the left lane.
[0072] (2) Constructing model output
[0073] The instantaneous speed of the target vehicle n in the time period from t to t+T is used as the model output data, that is: in, is the speed of the target vehicle at time t, and T is the prediction step length.
[0074] (3) Construct and train a short-term vehicle speed prediction model based on DBN deep belief network
[0075] like Figure 4 As shown in Figure 1, DBN consists of multiple layers of neurons, which are divided into two categories: visible neurons that accept input and hidden neurons responsible for extracting features. The key component of DBN is RBM, each of which consists of a visible layer and a hidden layer.
[0076] The construction and training steps of the short-term vehicle speed prediction model are as follows:
[0077] S1: Data Preparation
[0078] Import input data H = {h t-τ ,h t-τ+1 ,...,h t}, output data Set the random initialization parameter θ = {w ij ,a i ,b j} and the maximum number of iterations.
[0079] S2: Data preprocessing
[0080] The input and output data are preprocessed using the minimum and maximum normalization method, specifically:
[0081] enter:
[0082] Output:
[0083] in, is the normalized target vehicle speed prediction model input at time t, min(H) and max(H) are the minimum and maximum values in the input vector, respectively. is the normalized speed of the target vehicle n at time t+j, min(O) and max(O) are the minimum and maximum values in the output vector respectively;
[0084] S3: Calculate the neuron states of the hidden layer and the neuron states of the visible layer
[0085] Randomly select input data H and input it into the visible layer of the first layer RBM to calculate the neuron state of the hidden layer, P(v i =1h,θ)=sig(∑ i ω ij h i +a i ). Among them, the visible layer state vector v = (v 1 ,v 2 ,...,v i ), i is the number of neurons in the visible layer, v j ={0,1};
[0086] Calculate the neuron state of the visible layer, P(h i =1v,θ)=sig(∑ j v j ω ij +b j ), where the hidden layer state vector h = (h 1 ,h 2 ,...,h j ), j is the number of neurons in the hidden layer, h j ={0,1};
[0087] S4: Randomly input the next set of data and repeat step S3 until all sample data are selected;
[0088] S5: Train the first layer of RBM until the number of iterations is reached;
[0089] S6: Use the output of the first layer of trained RBM as the input of the next layer of RBM, and repeat this process until all RBM layers are trained.
[0090] Step 6: Use the short-term vehicle speed prediction model to make real-time predictions of short-term vehicle speed
[0091] The historical trajectory data of the vehicle to be predicted in the time period from t-τ to t is input into the trained short-term vehicle speed prediction model, and the speed of the vehicle to be predicted in the time period from t to t+T can be output.
[0092] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.
Claims
1. A method for predicting short-term vehicle speed on an expressway considering dynamic traffic scenarios, characterized in that: The steps include: Step 1: Based on the road traffic flow monitoring equipment, the vehicle trajectory data on the corresponding road section is obtained and a vehicle trajectory database is generated, wherein the vehicle trajectory data includes the vehicle position and the vehicle speed; Step 2: Divide the surrounding road area of the target vehicle into the area in front of the left lane, the area in front of the current lane, and the area in front of the right lane, construct an area set A = {fl, f, fr}, and assign a state attribute S to each area; Step 3: Distinguish the four traffic scenarios faced by the target vehicle based on the different state attribute combinations of each area; Step 4: Define the key scene vehicles that the target vehicle driver needs to pay attention to for different traffic scenarios; Step 5: Based on the DBN deep belief network, the target vehicle and key scene vehicle data within a certain period of time are used as input, and the instantaneous speed of the target vehicle in the future time period is used as output to build and train a short-term vehicle speed prediction model; Step 6: Use the trained short-term vehicle speed prediction model to make real-time predictions on the short-term vehicle speed of the target vehicle.
2. The method for predicting short-term vehicle speed on expressways considering dynamic traffic scenarios according to claim 1, characterized in that: The state attribute represents the occupancy of the corresponding area. When there is a car in the area, the state attribute S=1; Otherwise, S=0.
3. The method for predicting short-term vehicle speed on expressways considering dynamic traffic scenarios according to claim 2, characterized in that: The corresponding state attributes of different areas in the four traffic scenes and the corresponding relationships between vehicles in key scenes include: Traffic scenario 1: The area in front of the current lane is occupied, and the area in front of the left lane and the area in front of the right lane are not occupied, that is, S fl =0, S f =1, S fr =0, the key scene vehicle is the leading vehicle n+1; Traffic scenario 2: The area in front of the current lane and the area in front of the left lane are occupied, and the area in front of the right lane is not occupied, that is, S fl =1, S f =1, S fr =0; the key scene vehicles are the leading vehicle n+1 and the left front vehicle n+2; Traffic scenario 3: The area in front of the current lane and the area in front of the right lane are occupied, and the area in front of the left lane is not occupied, that is, S fl =0, S f =1, S fr =1; the key scene vehicles are the leading vehicle n+1 and the right front vehicle n+3; Traffic scenario 4: The area in front of the current lane, the area in front of the left lane, and the area in front of the right lane are all occupied, that is, S fl =1, S f =1, S fr =1; the key scene vehicles are the leading vehicle n+1, the left front vehicle n+2 and the right front vehicle n+3; Among them, S fl ,S f and S fr They are the occupancy status of the area in front of the left lane, the area in front of the current lane, and the area in front of the right lane.
4. The method for predicting short-term vehicle speed on expressways considering dynamic traffic scenarios according to claim 3 is characterized in that: The input of the short-term vehicle speed prediction model is specifically: H = {h t-τ ,h t-τ+1 ,...,h t }, where τ is the driver’s reaction time, h t is the key scene vehicle data at time t.
5. The method for predicting short-term vehicle speed on expressways considering dynamic traffic scenarios according to claim 4, characterized in that: In different traffic scenarios, h t The at least includes the longitudinal distance difference, relative speed and relative acceleration between the target vehicle and the key scene vehicle; For the case where there are critical scene vehicles on both sides of the target vehicle’s current lane, h t It also includes the lateral distance difference between the target vehicle and the key scene vehicle.
6. The method for predicting short-term vehicle speed on expressways considering dynamic traffic scenarios according to claim 4, characterized in that: The output of the short-term vehicle speed prediction model is specifically: in, is the speed of the target vehicle at time t, and T is the prediction step length.
7. The method for predicting short-term vehicle speed on expressways considering dynamic traffic scenarios according to claim 1, characterized in that: The training steps of the short-term vehicle speed prediction model include: Randomly select a set of data and use the minimum and maximum normalization method to preprocess the model input and output data; Calculate the neuron states of the hidden layer and the neuron states of the visible layer; Repeat the above steps until all data are selected; Train the first layer of RBM until the maximum number of iterations is reached; The output of the first layer of trained RBM is used as the input of the next layer of RBM, and the process is repeated until all RBM layers are trained.
Citation Information
Patent Citations
Hybrid traffic flow automobile cooperative self-adaptive cruise control method based on automobile-to-automobile communication
CN114394092A
Vehicle trajectory prediction method and system, computer equipment and storage medium
CN114881339A
Vehicle speed prediction method and system in self-vehicle expressway scene
CN116882505A
Training method for a convolutional neural network for predicting a driving maneuver of a traffic participant
EP3798912A1
