Highway dynamic information issuing method based on risk propagation prediction

Abstract

The invention discloses a highway dynamic information publishing method based on risk propagation prediction, and relates to the technical field of traffic information publishing. Comprising the steps of actively identifying risk types, constructing a risk propagation database and a traffic risk influence range prediction model, predicting a risk influence range caused by a real-time risk event of a road section, and constructing an accident early warning grade prediction model based on a GBDT (Gradient Broken Decision Tree) to predict an accident early warning grade. And issuing traffic risk induction information in real time and the like. According to the method, information release and information early warning are performed in a targeted manner according to the real-time position of the vehicle and the variable information board, information release under the time granularity is switched along with time migration, accurate prediction and timely release of highway traffic risks are realized, the timeliness and reliability of information release are improved, and the traffic risk of the highway is prevented from being damaged. And the driving safety and the risk management and control capability are improved.

Description

Technical Field

[0001] The invention relates to the technical field of traffic information release, and in particular to a method for releasing dynamic information on a highway based on risk propagation prediction. Background Art

[0002] With the rapid development of highways, traffic safety issues have become increasingly prominent. Existing highway traffic safety risk warnings mainly use variable information boards, loudspeakers, etc. to warn travelers of risks, and information release has a time lag. However, after a traffic risk occurs, it will spread to adjacent sections over time, resulting in risk hazards in adjacent sections. Warning drivers based solely on real-time traffic risk detection results may result in untimely warnings and failure to provide drivers with accurate long-term travel route planning and suggestions. To this end, we propose a method for dynamic highway information release based on risk propagation prediction. Summary of the invention

[0003] The purpose of the present invention is to provide a method for publishing dynamic information on highways based on risk propagation prediction, taking into account the propagation of various traffic risks such as congestion, accidents, and dangerous driving behaviors on the road section, and providing long-term risk warning information to drivers based on the real-time position of the vehicle and surrounding variable information boards, on-board navigation and other equipment, so as to achieve accurate prediction and timely release of highway traffic risks, and provide more comprehensive and effective protection for traffic safety.

[0004] The technical problem solved by the present invention is that the existing method for publishing highway traffic risk information does not take into account the real-time propagation of risks and cannot dynamically update risk warning information according to the real-time propagation of risks.

[0005] The present invention can be implemented by the following technical solution: A method for publishing dynamic information of a highway based on risk propagation prediction, comprising the following steps:

[0006] Step 1: Actively identify risk types based on roadside traffic flow detection equipment, including traffic congestion, traffic accidents, and dangerous driving behaviors;

[0007] Step 2: Build a risk propagation database and traffic risk impact range prediction model;

[0008] The risk propagation database is constructed based on the location, type, level and impact range of historical accident risks at different time scales. The risk propagation database is used to construct a traffic risk impact range prediction model based on the support vector regression (SVR) algorithm.

[0009] Step 3: Use the traffic risk impact range prediction model to predict the risk impact range caused by the real-time risk events on the road section;

[0010] Step 4: Based on historical accident risks, an accident warning level prediction model based on the GBDT gradient boosting decision tree is constructed, and the accident warning level is predicted using the model;

[0011] Step 5: Distinguish different types of risk events and publish traffic risk induction information in real time.

[0012] A further technical improvement of the present invention is that the specific method for determining the risk impact range includes:

[0013] Discretize the highway section into multiple sub-sections;

[0014] The cross-sectional speed of each sub-section is counted at a set time interval, and the risk impact condition is defined to determine whether the sub-section is affected by the risk;

[0015] Generate a set of road section numbers affected by the risk at a certain moment, and extract a set of road sections that meet spatial continuity from the set of road section numbers. The risk impact range at that moment is the interval range where the road section set is located.

[0016] A further technical improvement of the present invention is that the data model of the risk impact condition is expressed as:

[0017]

[0018] in, is the state attribute of sub-section i at time t, It means that sub-section i is affected by risk at time t; otherwise, Indicates that sub-section i is not affected by the risk at time t; and are the cross-sectional velocities of sub-section i at time t and time t-1, respectively. is the design speed of sub-section i, and ε is the speed change threshold.

[0019] A further technical improvement of the present invention is that the specific process of constructing a traffic risk impact range prediction model includes:

[0020] (1) Define input and output:

[0021] Assume that the training set is T = {(x1,y1),(x2,y2),...,(x k ,y k ),...,(x N ,y N )},in

[0022] x k ={p k ,s k ,rk ,q k ,v k} as input, y k ={y k1 ,y k2 ,y k3 ,y k4} as output;

[0023] p k is the location where the kth risk event occurs, s k is the risk type of the kth risk event, r k is the risk level of the kth risk event; s k and v k are the flow and speed information of the road section where the k-th risk event is located; y k1 ,y k2 ,y k3 ,y k4 They are the risk impact scopes at different time granularities;

[0024] (2) Select a penalty function C, where C>0, and construct and solve a convex quadratic programming problem:

[0025]

[0026] Among them, ω and b are the normal vector and intercept of the hyperplane respectively, ξ i and They are the mapping from input space to features, α i is the Lagrange multiplier vector, φ(x i ) is the mapping relationship from input space to feature space, (x i ,y i ) are sample point coordinates;

[0027] (3) Solve the optimization problem and obtain its optimal solution And select α * Satisfied The weight of the condition ,calculate Among them, K(x i ,x j ) is the kernel function;

[0028] (4) Save the trained traffic risk impact range prediction model.

[0029] A further technical improvement of the present invention is that the time granularity is divided into 5 minutes, 10 minutes, 15 minutes and 30 minutes.

[0030] A further technical improvement of the present invention is that the process of constructing the accident warning level prediction model includes:

[0031] Determine the input and output:

[0032] Select highway traffic volume / number of lanes / weather / large vehicle ratio / wind force level / whether it occupies the emergency lane / whether it occupies the driving lane / whether it occupies the overtaking lane / whether it is a bridge / whether it is a hub as input; accident risk warning level as output;

[0033] Design parameter configuration of GBDT gradient boosting decision tree;

[0034] The accident warning level prediction model is obtained by using the data in the accident history database as training samples;

[0035] The real-time value of an input indicator of a certain accident risk is used as input for prediction.

[0036] A further technical improvement of the present invention is that the real-time information release in step 5 includes the following aspects:

[0037] For traffic congestion risk event types: track the real-time location of vehicles, search for variable information boards within the congestion propagation range based on the real-time location of vehicles, and publish congestion time and space information on the variable information boards;

[0038] For traffic accident risk event types: based on the real-time location and speed information of the vehicle and the spatial location of the variable information board, the accident impact range and warning level information are dynamically updated and published on the variable information board and the vehicle navigation terminal;

[0039] For risk event types of dangerous driving behavior: based on the real-time location and driving direction of the vehicle, search for fixed variable information boards and tweeters upstream of the vehicle. The variable information boards and tweeters display the vehicle's license plate number and warning information, and provide continuous accompanying information warnings to drivers of dangerously driving vehicles.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention utilizes roadside traffic flow detection equipment and accident history database to construct a risk propagation database, and predicts the traffic risk impact range at different time granularities according to the location, type, level of risk events, and road section flow and speed information, and predicts the accident warning level based on the real-time value of the accident influencing factor indicator as input; in combination with the accident risk type, accident warning level and risk impact range, information is released and information warning is carried out in a targeted manner according to the real-time position of the vehicle and in conjunction with the variable information board, and the information release at the time granularity is switched as time passes, so as to achieve accurate prediction and timely release of highway traffic risks, improve the timeliness and reliability of information release, and improve driving safety and risk management capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0043] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0044] 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.

[0045] See also Figure 1 As shown, a method for publishing dynamic information of a highway based on risk propagation prediction includes the following steps:

[0046] Step 1: Identify traffic risk events

[0047] Actively identify risk event types based on roadside traffic flow detection equipment such as video cameras, vehicle-mounted terminals and radars; risk event types include congestion, accidents and dangerous driving behaviors (overloading, speeding and fatigue driving);

[0048] Step 2: Build a risk propagation database and traffic risk impact range prediction model

[0049] Based on the traffic data obtained by the roadside detection equipment, the data on the speed change over time of the road sections within 10km upstream and downstream of the congestion or accident location are extracted respectively, so as to determine the traffic risk impact range at the time granularity of 5min, 10min, 15min and 30min after the traffic congestion or accident occurs;

[0050] The risk propagation database is constructed using the risk location, risk type, risk level, and risk impact range obtained from road surface detection devices and accident history database, and the traffic risk impact range prediction model is constructed using the support vector regression SVR algorithm.

[0051] The methods for determining the impact scope of the above-mentioned traffic risks specifically include:

[0052] S11: discretizing the highway section into n sub-sections according to the spatial layout positions of the roadside detection equipment arranged on the highway;

[0053] S12: Count the cross-sectional speed of each sub-section at a time granularity of 1 minute, and define the risk impact conditions based on the speed changes of each section.

[0054] When the cross-sectional speed of a sub-section is greater than 50% of the design speed of the section, and the cross-sectional speed change value at adjacent moments does not exceed ε, the section is considered to be unaffected by the risk;

[0055] On the contrary, when the section speed of a sub-section does not exceed 50% of the design speed of the section, and the section speed change value at adjacent moments is greater than ε, the section is considered to be affected by risk;

[0056] The data model is expressed as follows:

[0057] in, is the state attribute of sub-section i at time t, It means that sub-section i is affected by risk at time t; otherwise, Indicates that sub-section i is not affected by the risk at time t. and are the cross-sectional velocities of sub-section i at time t and time t-1, respectively. is the design speed of sub-section i, ε is the speed change threshold, which can be defined according to the actual needs of the expressway and is generally taken as 0.01.

[0058] S13: Let the set of road section numbers affected by the risk at time t be A t =[k1,k2,......,k j ] t ,in, satisfy

[0059] S14: Extract the road segment set B that satisfies spatial continuity from the set A, then the risk impact range at time t is , where l max{Β} and l min{Β} They are the locations of the largest and smallest road section numbers in set B respectively.

[0060] The above spatial continuity means that several sections are continuous in space, forming a continuous section. The section set B contains a continuous set of sub-section changes, for example, B = {4, 5, 6, 7, 8}. According to the number set in set B, the risk impact range can be obtained, that is, the location of sub-section 8 minus the location of sub-section 4. Assuming that the section is from the starting point to the end point, the section length is measured from 0.

[0061] The specific process of constructing the traffic risk impact range prediction model using the support vector regression (SVR) algorithm mentioned above includes:

[0062] S21: Define inputs and outputs

[0063] Define input: training set T = {(x1,y1),(x2,y2),...,(x k ,y k ),...,(x N ,y N )}, where x k ={p k ,s k ,r k ,q k ,v k}, p k is the location where the kth risk event occurs, including: straight sections, curves, downhill sections, etc.; s k is the risk type of the kth risk event, including congestion, rear-end collision, side collision, etc.; r k is the risk level of the kth risk event; s k and v k are the flow and speed information of the road section where the k-th risk event is located.

[0064] Define output: y k represents the risk impact range after the k-th risk event occurs, y k ={y k1 ,y k2 ,y k3 ,y k4},y k1 ,y k2 ,y k3 ,y k4 The risk impact ranges are 5min, 10min, 15min and 30min time granularities respectively. The specific time granularity can be selected according to actual application requirements.

[0065] S22: Select a penalty function C, where C>0, and construct and solve a convex quadratic programming problem:

[0066]

[0067] Among them, ω and b are the normal vector and intercept of the hyperplane respectively, ξ i and They are the mapping from input space to features, α i is the Lagrange multiplier vector, φ(x i ) is the mapping relationship from input space to feature space, (x i ,y i ) are the sample point coordinates.

[0068] S23: Solving the optimization problem, we can get its optimal solution And select α * Satisfied The weight of the condition calculate Among them, K(x i ,x j ) is the kernel function. Commonly used kernel functions include linear kernel function, polynomial kernel function and Gaussian kernel function. The kernel function with good prediction effect is selected as the final kernel function used in the traffic risk impact range prediction model.

[0069] S24: Save the trained traffic risk impact range prediction model.

[0070] Step 3: Real-time prediction of traffic risk impact range

[0071] Based on the traffic flow operation data (flow, speed, etc.) and risk identification results obtained in real time by the front-end roadside detection equipment of the expressway, the real-time risk event factor set such as the real-time risk location, risk type, risk level, flow rate of the risk section, speed of the section, etc. is input. Based on the trained traffic risk impact range prediction model, the risk impact range can be predicted according to time granularities such as 5min, 10min, 15min and 30min. The specific time granularity can be selected according to actual application needs.

[0072] Step 4: Construct an accident warning level prediction model based on the GBDT gradient boosting decision tree

[0073] S31: Determine the input and output of the prediction model

[0074] Based on the accident history database of expressways, expressway flow / number of lanes / weather / large vehicle ratio / wind force level / whether the emergency lane is occupied / whether the driving lane is occupied / whether the overtaking lane is occupied / whether it is a bridge / whether it is a hub are selected as the input of the accident warning level prediction model, and the accident risk warning level is used as the output;

[0075] S32: Design the key parameters of the GBDT gradient boosting decision tree, including the learning rate, number of trees, maximum depth, minimum number of leaf node samples, feature selection ratio, minimum number of split samples, subsampling ratio, loss function, and regularization term. The learning rate is usually 0.01, and the remaining parameters can be adaptively optimized through parameter optimization methods such as Bayesian optimizer.

[0076] S33: Use the training samples to train the network and obtain a trained GBDT gradient boosting decision tree;

[0077] S34: Using the real-time value of the accident influencing factor indicator mentioned in step S31 as the model input, the accident warning level is predicted by the trained accident warning level prediction model.

[0078] Step 5: Release traffic risk information in real time

[0079] S41: For traffic congestion risk event types, based on the real-time location of the vehicle detected by the checkpoint and the driver's travel route, the vehicle license plate number is used as the tracking basis, and the roadside detection equipment deployed on the highway is used to track the real-time location of the vehicle; based on the real-time location of the vehicle and the location of the variable information board, the variable information board within the congestion propagation range is searched to publish differentiated and accurate congestion time and space information;

[0080] Based on the travel path and real-time location of individual vehicles, the congestion duration of the road section where the vehicle is located is dynamically displayed on the navigation terminal device at a time granularity of 5 minutes, 10 minutes, 15 minutes and 30 minutes;

[0081] S42: For traffic accidents, based on the real-time position of vehicles on the road, the spatial position of fixed variable information boards, the position and speed information of navigation vehicles, and the accident warning level prediction model are used to make real-time predictions of the accident level, and the accident impact range and warning level information are dynamically updated and published on the variable information board and the vehicle navigation terminal;

[0082] S43: For dangerous driving behaviors, the license plate number is used as the tracking basis. The vehicle is dynamically tracked according to the video checkpoints along the highway. Based on the vehicle's real-time location and driving direction, the variable information board and tweeter upstream of the vehicle are searched. The variable information board and tweeter display the vehicle's license plate number and warning information, and continuous accompanying information warnings are provided to the driver of the dangerous driving vehicle.

[0083] 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 publishing dynamic information on highways based on risk propagation prediction, characterized in that: The steps include: Step 1: Actively identify risk types based on roadside traffic flow detection equipment, including traffic congestion, traffic accidents, and dangerous driving behaviors; Step 2: Build a risk propagation database and traffic risk impact range prediction model; The risk propagation database is constructed based on the location, type, level and impact range of historical accident risks at different time scales. The risk propagation database is used to construct a traffic risk impact range prediction model based on the support vector regression (SVR) algorithm. Step 3: Use the traffic risk impact range prediction model to predict the risk impact range caused by the real-time risk events on the road section; Step 4: Based on historical accident risks, an accident warning level prediction model based on the GBDT gradient boosting decision tree is constructed, and the accident warning level is predicted using the model; Step 5: Distinguish different types of risk events and publish traffic risk induction information in real time.

2. A highway dynamic information release method based on risk propagation prediction according to claim 1, characterized in that: The specific methods for determining the scope of risk impact include: Discretize the highway section into multiple sub-sections; The cross-sectional speed of each sub-section is counted at a set time interval, and the risk impact condition is defined to determine whether the sub-section is affected by the risk; Generate a set of road section numbers affected by the risk at a certain moment, and extract a set of road sections that meet spatial continuity from the set of road section numbers. The risk impact range at that moment is the interval range where the road section set is located.

3. A highway dynamic information release method based on risk propagation prediction according to claim 2, characterized in that: The data model of the risk impact condition is expressed as: in, is the state attribute of sub-segment i at time t, It means that sub-section i is affected by risk at time t; otherwise, Indicates that sub-section i is not affected by the risk at time t; and are the cross-sectional velocities of sub-section i at time t and time t-1, respectively. is the design speed of sub-section i, and ε is the speed change threshold.

4. The method for publishing highway dynamic information based on risk propagation prediction according to claim 1 is characterized in that: The specific process of constructing the traffic risk impact range prediction model includes: (1) Define input and output: Assume that the training set is T = {(x1,y1),(x2,y2),...,(x k ,y k ),...,(x N ,y N )}, where x k ={p k ,s k ,r k ,q k ,v k } as input, y k ={y k1 ,y k2 ,y k3 ,y k4 } as output; p k is the location where the kth risk event occurs, s k is the risk type of the kth risk event, r k is the risk level of the kth risk event; s k and v k are the flow and speed information of the road section where the k-th risk event is located; y k1 ,y k2 ,y k3 ,y k4 They are the risk impact scopes at different time granularities; (2) Select a penalty function C, where C>0, and construct and solve a convex quadratic programming problem: Among them, ω and b are the normal vector and intercept of the hyperplane respectively, ξ i and They are the mapping from input space to features, α i is the Lagrange multiplier vector, φ(x i ) is the mapping relationship from input space to feature space, (x i ,y i ) are sample point coordinates; (3) Solve the optimization problem and obtain its optimal solution α * Satisfied The weight of the condition calculate Among them, K(x i ,x j ) is the kernel function; (4) Save the trained traffic risk impact range prediction model.

5. A highway dynamic information release method based on risk propagation prediction according to claim 4, characterized in that: The time granularity is divided into 5min, 10min, 15min and 30min.

6. The method for publishing highway dynamic information based on risk propagation prediction according to claim 1 is characterized in that: The process of constructing the accident warning level prediction model includes: Determine the input and output: Select highway traffic volume / number of lanes / weather / large vehicle ratio / wind force level / whether it occupies the emergency lane / whether it occupies the driving lane / whether it occupies the overtaking lane / whether it is a bridge / whether it is a hub as input; accident risk warning level as output; Design parameter configuration of GBDT gradient boosting decision tree; The accident warning level prediction model is obtained by using the data in the accident history database as training samples; The real-time value of an input indicator of a certain accident risk is used as input for prediction.

7. The method for publishing highway dynamic information based on risk propagation prediction according to claim 1 is characterized in that: The real-time information release in step 5 includes the following aspects: For traffic congestion risk event types: track the real-time location of vehicles, search for variable information boards within the congestion propagation range based on the real-time location of vehicles, and publish congestion time and space information on the variable information boards; For the risk event type of traffic accidents: based on the real-time location and speed information of the vehicle and the spatial location of the variable information board, the accident impact scope and warning level information are dynamically updated and published on the variable information board and the vehicle navigation terminal; For risk event types of dangerous driving behavior: based on the real-time location and driving direction of the vehicle, search for fixed variable information boards and tweeters upstream of the vehicle. The variable information boards and tweeters display the vehicle's license plate number and warning information, and provide continuous accompanying information warnings to drivers of dangerously driven vehicles.

Images