Traffic conflict assessment method and system based on multivariate and Bayesian network

Abstract

The invention relates to the technical field of traffic control systems, and in particular to a traffic conflict assessment method and system based on multivariate and Bayesian networks, wherein the method comprises: obtaining traffic conflict history data; constructing a data set based on the traffic conflict history data; constructing a directed acyclic graph according to the directed connection relationship of each variable in the traffic conflict history data; the directed acyclic graph comprises: a node set and an edge set, wherein the node set is the data in the data set; obtaining a Bayesian network based on the directed acyclic graph; training and optimizing the Bayesian network based on a training set, wherein the input value of the network is the data of the training set, and the output value of the network is the predicted probability value of the traffic conflict; testing the Bayesian network based on a test set; inputting the traffic data to be predicted into the trained Bayesian network to obtain the predicted probability value of the traffic conflict. The accuracy and precision of assessing the severity of the traffic conflict are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic control systems, and in particular to a traffic conflict assessment method and system based on multivariate and Bayesian networks. Background Art

[0002] Traffic conflicts arise from the complex interactions and mutual influences of vehicles on the road, and these conflicts are further exacerbated in merging areas. In the initial stage, merging vehicles need to merge from parallel acceleration lanes into the main lane, which may be occupied by high-speed vehicles. The speed difference and position relationship between merging vehicles and main lane vehicles can cause various problems. As merging vehicles try to synchronize with main lane vehicles and eventually merge into the traffic flow, conflicts gradually decrease, but if the operation is not done properly or the traffic volume in the merging area is heavy, conflicts may still develop into accidents.

[0003] Conflicts in the merging zone can be defined as various dangerous traffic situations that occur between merging vehicles and mainline vehicles, or between merging vehicles and surrounding vehicles. According to the nature of the conflict, conflicts in the merging zone can be divided into conflicts between merging vehicles and the vehicles ahead in the mainline lane, and conflicts between merging vehicles and the vehicles behind in the mainline lane. Driver behavior, traffic flow, road design, and traffic management may all affect the occurrence and development of conflicts.

[0004] In terms of traffic conflict assessment, there are mainly sensor-based, vehicle-to-vehicle communication-based and simulation-based traffic conflict assessments. These research results provide important theoretical and technical support for the assessment of traffic conflicts in this study. By identifying, evaluating and predicting traffic conflicts, it is beneficial to early warning of traffic accidents, reduce the number of traffic accidents, and improve road traffic safety.

[0005] At present, most of the research backgrounds on traffic conflicts are concentrated in the intersections and main sections of the merging area, and there are few studies on other parts, especially on the conflict in the merging area perceived by the roadside. Therefore, the existing research results cannot guarantee whether they are applicable to the assessment of traffic conflicts in all areas of the merging area, and the results need to be verified at a deeper level. This makes it difficult for most current studies to serve the development of the overall intelligent management of the merging area. The authenticity of the research results produced in the merging area needs to be improved, and there are still challenges and difficulties in successful application in real road environments. Summary of the invention

[0006] In order to solve the deficiencies of the prior art, the present invention provides a traffic conflict assessment method and system based on multivariate and Bayesian networks.

[0007] On the one hand, a traffic conflict assessment method based on multivariate and Bayesian network is provided, including: obtaining historical traffic conflict data; constructing a data set based on the historical traffic conflict data; in the process of constructing the data set, classifying the data set and supplementing the missing data in the classified data set; dividing the data set into a training set and a test set; constructing a directed acyclic graph according to the directed connection relationship of each variable in the historical traffic conflict data; the directed acyclic graph includes: a node set and an edge set, and the node set is the data in the data set; based on the directed acyclic graph, a Bayesian network is obtained; based on the training set, the Bayesian network is trained and optimized, the input value of the network is the data of the training set, and the output value of the network is the predicted probability value of the traffic conflict; based on the test set, the Bayesian network is tested, when all the test indicators meet the requirements, it means that the trained Bayesian network meets the requirements, otherwise, the training set is replaced and trained again; the traffic data to be predicted is input into the trained Bayesian network to obtain the predicted probability value of the traffic conflict.

[0008] On the other hand, a traffic conflict assessment system based on multivariate and Bayesian network is provided, including: an acquisition module, which is configured to: acquire traffic conflict history data; construct a data set based on the traffic conflict history data; in the process of constructing the data set, classify the data set and supplement the missing data in the classified data set; divide the data set into a training set and a test set; a network construction module, which is configured to: construct a directed acyclic graph according to the directed connection relationship of each variable in the traffic conflict history data; the directed acyclic graph includes: a node set and an edge set, the node set is the data in the data set According to; based on the directed acyclic graph, a Bayesian network is obtained; a training module is configured to: based on the training set, train and optimize the Bayesian network, the input value of the network is the data of the training set, and the output value of the network is the predicted probability value of the traffic conflict; a testing module is configured to: based on the test set, test the Bayesian network, when the test indicators all meet the requirements, it means that the trained Bayesian network meets the requirements, otherwise, the training set is replaced and trained again; a prediction module is configured to: input the traffic data to be predicted into the trained Bayesian network to obtain the predicted probability value of the traffic conflict.

[0009] The above technical solution has the following advantages or beneficial effects: (1) In order to improve the evaluation accuracy of the model, the data layer classifies the data, uses the relationship between the target node and the observation node to establish a Bayesian network, and performs parameter estimation. Bayesian reasoning is carried out based on the Bayesian network and parameter estimation results. The possibility of traffic conflicts is predicted, and a traffic conflict evaluation method based on a multivariate Bayesian hierarchical network is formed, which is conducive to improving the accuracy and precision of evaluating the severity of traffic conflicts. (2) The accuracy of data classification processing reaches 84%, and the MAPE is 3.5%, which effectively improves the evaluation performance of the model. The evaluation accuracy is improved by about 15% compared with the original. In addition, the model realizes the variable-level traffic conflict evaluation in the merging area, which provides strong support for identifying potential conflict factors in the traffic system. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0011] Figure 1 This is a schematic diagram of post-invasion time calculation in Example 1.

[0012] Figure 2 This is a Bayesian network topology diagram of Example 1.

[0013] Figure 3 Schematic diagram of the Bayesian network vehicle risk assessment results of Example 1.

[0014] Figure 4 Schematic diagram of the Bayesian network model of Example 1.

[0015] Figure 5 This is a schematic diagram of a highway merging area in Example 1. DETAILED DESCRIPTION

[0016] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0017] Terminology explanation: Post-entry time (PET): The time difference between the rear end of the leading vehicle leaving the conflict zone and the front end of the trailing vehicle entering the conflict zone. Figure 1 As shown, it is assumed that the area between the first invasion line and the second invasion line is the conflict area, wherein the first invasion line represents the line where the conflict first occurs, and the second invasion line represents the line where the conflict finally ends.

[0018] For the conflict between the merging vehicle and the vehicle in front of the target lane, the moment when the rear end of the vehicle A in front of the target lane leaves the first invasion line is recorded as T1, and the moment when the head of the merging vehicle B enters the first invasion line is recorded as T2, then PET front = T2-T1, and the larger the PET value is, the smaller the possibility of conflict.

[0019] For the conflict between the merging vehicle and the vehicle behind the target lane, let the time the merging vehicle travels in the conflict area be T1, and the time the vehicle in the target lane travels in the sudden area be T2, then PET = T1-T2. The value is inversely proportional to the degree of conflict. The larger the value, the smaller the possibility of conflict.

[0020] In the present invention, the length of the acceleration lane is recorded as the conflict area. In the same time interval, the time difference between the tail or head of the front or rear vehicle in the target lane leaving or entering the conflict area and the head or tail of the merging vehicle entering or leaving the conflict area. The overall conflict between the merging vehicle and the front vehicle in the target lane and the overall conflict between the merging vehicle and the rear vehicle in the target lane is low when IPET>3.

[0021] After calculating the value of PET, the present invention divides the conflict determination into two parts, namely, the conflict between the merging vehicle and the vehicle in front of the target lane and the conflict between the merging vehicle and the vehicle behind the target lane.

[0022] 2. Inverse PET (IPET): The travel time difference between the target lane vehicle and the merging vehicle in the conflict area. The value is inversely proportional to the degree of conflict. The larger the value, the smaller the probability of vehicle conflict.

[0023] Embodiment 1: This embodiment provides a traffic conflict assessment method based on multivariate and Bayesian network, including: S101: obtaining traffic conflict history data; constructing a data set based on the traffic conflict history data; in the process of constructing the data set, classifying the data set and supplementing the missing data in the classified data set; dividing the data set into a training set and a test set; S102: constructing a directed acyclic graph according to the directed connection relationship of each variable in the traffic conflict history data; the directed acyclic graph includes: a node set and an edge set, and the node set is a data set. concentrated data; based on the directed acyclic graph, a Bayesian network is obtained; S103: based on the training set, the Bayesian network is trained and optimized, the input value of the network is the data of the training set, and the output value of the network is the predicted probability value of the traffic conflict; S104: based on the test set, the Bayesian network is tested, when the test indicators all meet the requirements, it means that the trained Bayesian network meets the requirements, otherwise, the training set is replaced and trained again; S105: the traffic data to be predicted is input into the trained Bayesian network to obtain the predicted probability value of the traffic conflict.

[0024] Further, S101: Obtain traffic conflict data, which is data collected based on roadside laser radar, and conflict data is collected on the highway. According to the actual traffic flow characteristics, 11:00-12:00 in the morning and 17:00-18:00 in the afternoon on weekdays are selected as data collection periods, and a total of 6 hours of traffic flow data are collected.

[0025] Based on the extracted vehicle trajectory data, the vehicle trajectory database of the merging area is constructed. The disconnected trajectory data is manually processed through the speed, acceleration, timestamp and other information of the target vehicle.

[0026] For the target vehicle speed solution, since the laser radar can collect the three-dimensional information data of the target, after trajectory tracking, the three-dimensional information of the target speed in each frame in the laser radar coordinate system can be obtained. Therefore, the target vehicle's speed can be obtained by the following formula: ;in, represents the speed of the target vehicle, Indicates that the vehicle is following The speed of the axis, Indicates that the vehicle is following The speed of the axis, Indicates that the vehicle is following The speed at which the axis advances.

[0027] The timestamp between two adjacent frames of the LiDAR is 0.1s. The same vehicle will appear in different frames continuously. Therefore, the acceleration of the vehicle in each frame can be obtained by taking the speed difference between two adjacent frames of the vehicle. ;in, is the acceleration of the target vehicle, For the target vehicle The speed of time, is the speed of the target vehicle at time t-1. Based on the complete trajectory data obtained, a vehicle trajectory database based on the roadside lidar is constructed, including: timestamp, frame number, target type, and laser radar coordinate system. axis, axis, Axle, vehicle height, width, distance, speed and other information.

[0028] Further, S101: construct a data set based on historical traffic conflict data; in the process of constructing the data set, classify the data set and supplement the missing data in the classified data set, wherein the classification process refers to: classifying each factor parameter into a traffic conflict generating category and a traffic conflict not generating category.

[0029] Furthermore, the classification process includes: decision tree classification algorithm C4.5.

[0030] Furthermore, the classification process includes: S101-1: calculating information entropy; S101-2: calculating information gain based on information entropy; S101-3: selecting a point with the largest information gain value as a segmentation point to classify the data.

[0031] The possibility of merging conflict is selected as the target node, and the merging speed, merging distance, relative speed and relative distance related to the merging area conflict are selected as the observation node. The target node and observation node data are discretized and the discretization results are shown in Table 1.

[0032] Furthermore, the S101-1: calculating information entropy includes: the definition formula of information entropy is: ;in: Represents the current sample data set, Indicates the sample data category, Indicates that in the current sample data set Middle range The number of internal values, Representation interval Inside The number of values. The smaller the value, the higher the purity of the partition interval. The C4.5 algorithm information gain is used to perform decision tree attribute partitioning. Assume that there is a set of discrete attribute variables , the current data set will be split, and There are t-1 values, namely , at this time, the current data set can be divided into t samples, each sample is recorded as .

[0033] Table 1 Data classification results

[0034]

[0035] It should be understood that information entropy is used to measure the feature purity of the classification interval. When processing discretization, this property is used to select the best split point to ensure that the data in each interval has high similarity and uniformity.

[0036] The calculation of information entropy refers to calculating the information entropy of the four data of inflow speed, inflow distance, relative speed and relative distance. Specifically, it refers to discretizing the data based on the information entropy. Table 1 shows the data classification results.

[0037] In order to compare the binning performance of the C4.5 algorithm, the k-means algorithm was used to bin the data and the results were compared. In terms of precision, the C4.5 algorithm is approximately 84%, and the K-Means algorithm is approximately 79%. The accuracy of the C4.5 data binning is higher than that of the K-Means algorithm. In terms of recall, the C4.5 algorithm is higher than the K-Means algorithm, and the MAPE value of the C4.5 algorithm is smaller than that of the K-Means algorithm. This shows that the C4.5 algorithm has better classification results and can improve the evaluation accuracy of the model to a greater extent.

[0038] Furthermore, the S101-2: calculating information gain based on information entropy includes: ;in, represents information gain.

[0039] Given the current sample data set D and discrete attribute q, the discrete attribute q has t-1 different values, and these values ​​are arranged in order, recorded as , then randomly select a split point c from the set q to divide the current data set. At this time, the data set can be divided into two parts, and the part larger than the split point c is recorded as The part smaller than the split point c is recorded as At this time, select The midpoint of is taken as the pre-segmentation point, that is, , ( ) . Calculate the information gain of each segmentation point and select The split point with the largest value is taken as the final split point.

[0040] It should be understood that classifying the Bayesian network node data can not only effectively improve its evaluation performance, but also when the data itself has a true probability distribution different from the Gaussian distribution, after classification, the same probability estimation result can be obtained during the Bayesian network estimation. In addition, classifying and processing continuous data can also significantly improve the learning efficiency of the Bayesian network model and greatly reduce the time complexity. Therefore, in order to better meet the modeling needs, the node data will be classified.

[0041] Furthermore, the missing data in the classified data set are supplemented by interpolation.

[0042] Furthermore, dividing the data set into a training set and a test set is to divide the data according to a set ratio.

[0043] Further, the S102: construct a directed acyclic graph according to the directed connection relationship of each variable in the traffic conflict history data, the directed acyclic graph includes: a node set and an edge set, the node set is the data in the data set; the nodes include: a relative speed node between the merging vehicle and the front vehicle, a relative speed node between the merging vehicle and the rear vehicle, a first difference node, a speed node of the merging vehicle, a distance node between the merging vehicle and the front vehicle, a distance node between the merging vehicle and the rear vehicle, a merging time node of the merging vehicle, a relative distance node between the merging vehicle and the front vehicle, a relative distance node between the merging vehicle and the rear vehicle, and a second difference node; wherein the first difference refers to: the difference between the relative speed of the merging vehicle and the front vehicle and the relative speed of the merging vehicle and the rear vehicle; the second difference refers to: the difference between the relative distance between the merging vehicle and the front vehicle and the relative distance between the merging vehicle and the rear vehicle.

[0044] Further, S102: the edge set refers to whether there is a relationship between nodes. If there is a relationship, there is an edge connecting the two nodes. Otherwise, there is no edge connecting the two nodes. The edge set includes: a connecting edge between the relative speed node of the merging vehicle and the front vehicle and the traffic conflict node, a connecting edge between the relative speed node of the merging vehicle and the rear vehicle and the traffic conflict node, a connecting edge between the first difference node and the traffic conflict node, a connecting edge between the speed node of the merging vehicle and the traffic conflict node, a connecting edge between the distance node of the merging vehicle and the front vehicle and the traffic conflict node, and a connecting edge between the distance node of the merging vehicle and the rear vehicle and the traffic conflict node. The connecting edges between the conflict nodes, the connecting edges between the merging time node of the merging vehicle and the traffic conflict node, the connecting edges between the relative distance node between the merging vehicle and the leading vehicle and the traffic conflict node, the connecting edges between the relative distance node between the merging vehicle and the trailing vehicle and the traffic conflict node, the connecting edges between the second difference node and the traffic conflict node, the connecting edges between the relative speed node between the merging vehicle and the leading vehicle and the first difference node, the connecting edges between the relative speed node between the merging vehicle and the trailing vehicle and the first difference node, the connecting edges between the relative distance node between the merging vehicle and the leading vehicle and the second difference node, and the connecting edges between the relative distance node between the merging vehicle and the trailing vehicle and the second difference node.

[0045] Further, the S102: based on the directed acyclic graph, obtaining a Bayesian network includes: the basic formula of the Bayesian network is as follows: ;in, In the observed data back The posterior probability, that is, the problem to be solved, is the data described in S101-3. Indicated in the parameter Observed data The probability of , also called the likelihood function. express The prior probability of . Represents observation data The probability of , also called marginal probability.

[0046] Table 2 Conditional probability table of road aggregate linear design

[0047]

[0048] Among them, P represents probability, S represents that the road is a curve, and Z represents that the road is a straight line.

[0049] Table 3 Vehicle acceleration conditional probability table

[0050]

[0051] in, Indicates that the acceleration is large. Indicates a small acceleration, measured in meters per second squared.

[0052] Table 4 Vehicle speed conditional probability table

[0053]

[0054] in, Indicates a higher speed. Indicates a smaller speed, measured in meters per second.

[0055] Table 5 Vehicle risk conditional probability table

[0056]

[0057] in, Indicates that the acceleration is large. Indicates a small acceleration, measured in meters per second squared.

[0058] The Bayesian network is used to establish points and edges, and the multiple factors affecting traffic conflicts are assigned to each point. The edges connect traffic conflicts with various influencing factors, thereby constructing a traffic conflict network.

[0059] Bayesian shares qualitative and quantitative descriptions. The qualitative description refers to the Bayesian network being a directed acyclic graph, such as Figure 2 As shown in Figure 1. The network consists of two parts: directed edges and nodes. The quantitative description mainly refers to the Bayesian principle, which is mainly based on two types of information: prior information and observation information. These two types of information are used to infer the posterior probability distribution of unknown parameters or variables. The merging speed, merging distance, relative speed and relative distance are selected as variables for traffic conflict assessment. The variable selection is based on the fact that the ego vehicle and the merging process are most affected by the merging speed, merging distance and relative speed. At the same time, the relative distance also has a certain impact on the merging vehicle.

[0060] Bayesian network determination and parameter estimation, by analyzing and determining the dependency between node variables, and then determining the Bayesian network. The present invention adopts parameter estimation mainly by calculating the conditional probability between nodes to sequentially explain the correlation between the target node and the observation node.

[0061] Further, the S103: based on the training set, the Bayesian network is trained and optimized, the input value of the network is the data of the training set, and the output value of the network is the predicted probability value of the traffic conflict, specifically including: training and optimizing the Bayesian network, the data of the training set refers to the merging speed, merging distance, relative speed and relative distance.

[0062] Use the likelihood function to represent the impact of the current data set, and the parameters that need to be estimated are , is recorded as a random variable, , then, the Bayesian formula is used to calculate the parameters The posterior distribution of : .

[0063] Among them, the current dataset , represents the likelihood function, Indicates the given parameter Post-variable The probability of , that is, the influence of the current data set, Indicates known Under the conditions The conditional distribution of express The prior distribution of Representation parameters probability.

[0064] when When the estimated parameter .

[0065] The calculation process of the parameters to be estimated is: ; ; ; ;in, represents the parameters that need to be estimated, Indicates different , , value, express The maximum value of express The maximum value of express The maximum value of , , represents the number of different conditions. represents the parameters to be estimated after solving the problem, Under the condition that C The conditional distribution of Indicates conditions, Indicates the condition The size of the case, represents probability; it will be solved based on three assumptions :(1)Global parameter independence assumption: The parameters of each node are independent of the parameters of other nodes;(2)Local parameter independence assumption: Given the value state of the target node, the parameters of each variable are also independent of the parameters of other variables;(3)Parameters obey Dirichlet distribution: Dirichlet distribution is a priori distribution used to describe the parameters of multinomial distribution. Assume that the parameters It obeys the Dirichlet distribution, which indicates the diversity and sparsity of parameters, making the estimation of parameters accurate and robust.

[0066] Based on the above three assumptions, the parameters to be estimated are: ;in, Represents the prior distribution The hyperparameters of Indicates that in data set C, the target node is satisfied , The number of values ​​of represents the parameter to be estimated.

[0067] Furthermore, S104: based on the test set, the Bayesian network is tested. When all the test indicators meet the requirements, it means that the trained Bayesian network meets the requirements. Otherwise, the training set is replaced and trained again, which specifically includes: selecting precision, recall and MAPE indicators for evaluation.

[0068] Precision( ): The ratio of the predicted positive samples to the actual positive samples. The higher the accuracy, the better: . Recall rate ( ): The proportion of predicted positive samples among the actual positive samples. The larger the recall rate, the better: ;in, Indicates the number of predicted values ​​that are positive and the true value that is positive, indicates the number of predicted values ​​that are negative and true values ​​that are positive, represents the number of predicted values ​​that are positive and true values ​​that are negative, It indicates the number of predicted values ​​that are negative and the true values ​​that are negative. Table 6 is the confusion matrix of the samples divided into four categories.

[0069] Table 6 Confusion matrix

[0070]

[0071] The calculation of APE and MAPE can be expressed as: ; .

[0072] APE is the ratio of the number of correctly predicted events to the number of events involved in the prediction, while MAPE is the mean absolute percentage error. represents the number of correctly predicted events, Indicates the number of events involved in the prediction, Represents the total number of events. Indicates The value of APE.

[0073] Furthermore, after S104 and before S105, the process also includes: S104-5: determining whether a conflict occurs between the merging vehicle and the vehicle in front of the target lane, or whether a conflict occurs between the merging vehicle and the vehicle behind the target lane; if a conflict occurs between the merging vehicle and the vehicle in front of the target lane, or a conflict occurs between the merging vehicle and the vehicle behind the target lane, the process proceeds to S105; if not, the process does not proceed to S105.

[0074] Furthermore, whether a conflict occurs between the merging vehicle and the vehicle ahead in the target lane is determined by the following criteria: assuming that the expressway is provided with a first section, a second section and a third section connected in sequence; the first section has three parallel lanes, the second section has three parallel lanes, and the third section has three parallel lanes; the first section is upstream of the merging area, the second section is the conflict area, and the third section is downstream of the merging area; the connecting line between the first section and the second section is the first intrusion line, and the connecting line between the second section and the third section is the second intrusion line; an acceleration lane is provided at the outer edge of the highway of the second section, the acceleration lane is arranged in parallel with the three lanes of the second section, the starting end of the acceleration lane is connected to the ending end of the ramp, and the ending end of the acceleration lane is connected to the conflict area; the merging vehicle enters the acceleration lane from the ramp; after accelerating in the acceleration lane, the merging vehicle enters the conflict area from the acceleration lane according to the driving instruction of the driver, and finally, the merging vehicle exits the conflict area and enters the downstream of the merging area.

[0075] In the process of entering the conflict area, for the conflict between the merging vehicle and the vehicle in front of the target lane, the time when the tail of the vehicle A in front of the target lane leaves the first invasion line is recorded as T1, and the time when the head of the merging vehicle B enters the first invasion line is recorded as T2, then PET1=T2-T1. If PET1 is less than the set threshold, it means that a conflict will occur between the merging vehicle and the vehicle in front of the target lane.

[0076] Furthermore, the judgment standard for whether a conflict occurs between the merging vehicle and the rear vehicle in the target lane is: in the process of exiting the conflict area, for the conflict between the merging vehicle D and the rear vehicle C in the target lane, the time when the tail of the merging vehicle D leaves the second invasion line is recorded as T3, and the time when the head of the rear vehicle C in the target lane enters the second invasion line is recorded as T4, then PET2=T4-T3, if PET2 is less than the set threshold, it means that a conflict will occur between the merging vehicle and the rear vehicle in the target lane.

[0077] Furthermore, the S105: inputting the traffic data to be predicted into the trained Bayesian network to obtain the predicted probability value of traffic conflict occurrence, specifically including: when the merging speed of the merging vehicle is less than 17.78 m / s, the probability of conflict occurrence increases from 19% to 31%, indicating that when the merging speed of the merging vehicle is less than this value or smaller, traffic conflict is more likely to occur. The reason is that when the merging vehicle merges at a lower speed, the speed difference between it and the vehicles traveling on the highway is large, and traffic conflict is likely to occur.

[0078] When the merging speed of the merging vehicles is greater than 17.78 m / s, the probability of traffic conflict decreases, which means that the merging speed of some merging vehicles is equivalent to the speed of the target lane vehicles.

[0079] When the relative speed between the merging vehicle and the preceding vehicle is less than 3.55 m / s, the probability of traffic conflict increases from 19% to 34%, indicating that traffic conflict is more likely to occur when the speed difference between the preceding vehicle and the merging vehicle is small or very close. However, when the relative speed with the preceding vehicle gradually increases, the probability of traffic conflict decreases.

[0080] When the speed of the merging vehicle and the vehicle behind the target lane is less than 3.66 m / s, the probability of traffic conflict increases from 19% to 33%. This shows that when the speed of the vehicle behind the merging vehicle is relatively small, traffic conflict is also likely to occur. However, as the relative speed to the vehicle behind gradually increases, the probability of traffic conflict decreases. This shows that the relative speed of the merging vehicle to the vehicle ahead and the vehicle behind the target lane has a greater impact on whether a traffic conflict occurs.

[0081] When the distance between the merging vehicle and the vehicle in front of the target lane is less than 8.45m, the probability of traffic conflict at this time is 47% at most compared with other factors, indicating that the relative distance between the merging vehicle and the vehicle in front of the target lane must be maintained above this value. From the reasoning results, it can be seen that when the relative distance to the vehicle in front is greater than 8.45m, the probability of traffic conflict is reduced from 47% to 21%.

[0082] When the relative distance between the merging vehicle and the vehicle behind the target lane is less than 9.29m, the probability of traffic conflict is 43%. When the distance between the two vehicles gradually increases, the probability of traffic conflict decreases from 43% to 20%, indicating that as the distance between the two vehicles increases, the probability of vehicle conflict decreases.

[0083] When the distance between the merging vehicle and the end of the acceleration lane is less than 48.47m, the possibility of traffic conflict increases from 32%. As the merging distance increases, the possibility of conflict decreases to 22%. This shows that if the merging vehicle merges into the target lane too late, the risk of traffic conflict will increase.

[0084] Bayesian network reasoning is expressed as inferring the probability of related nodes, i.e., the posterior probability, based on the Bayesian network given some node values.

[0085] Figure 3 is an example of using Bayesian network reasoning. In the road line type, Z indicates that the road is a straight line, and S indicates that the road is a curve; in the vehicle acceleration, Indicates that the acceleration is large. Indicates that the acceleration is small; in vehicle speed, It means the speed is greater than the first set threshold, which is 17.78 m / s. Indicates that the acceleration is less than the first set threshold; 0 indicates that the target vehicle has no traffic risk, and 1 indicates that the target vehicle has a traffic risk.

[0086] The model data was processed. The possibility of merging conflicts was selected as the target node, and the merging speed, merging distance, relative speed, and relative distance related to the merging area conflict were selected as the observation nodes. The target node and observation node data were discretized and reversed using the above-mentioned data discretization processing method based on information entropy. In order to compare the classification performance of the C4.5 algorithm, the k-means algorithm was used to classify the data.

[0087] Perform modeling and parameter estimation operations. By analyzing the dependency between the target node and the prediction node, establish Figure 4 The Bayesian network is constructed. Then the parameters are estimated. On this basis, the above Bayesian parameter estimation method is used.

[0088] By setting the value of the prediction node and using the conditional probability table between the prediction node and the target node, the posterior probability of the target node is inferred and Bayesian network reasoning is performed.

[0089] Traffic conflict metrics can be divided into two categories, namely time-related and non-time-related. Among the non-time-related metrics, they can be further divided into metrics related to distance, deceleration and other attributes. Based on driving behavior analysis, the present invention proposes an improved traffic conflict identification index IPET, which effectively identifies traffic conflicts in merging areas. For merging areas with traffic conflicts, a multivariate Bayesian hierarchical merging area conflict assessment model is established, which effectively improves the evaluation performance of the model and reduces the potential conflict factors in the identified traffic system.

[0090] Among the conflict indicators for conflict identification in merging areas, IPET comprehensively considers the changes in spatial proximity and time. It has been widely recognized both in theory and in practical applications, has high reliability, and is easy to calculate. Therefore, IPET is selected as the indicator for measuring merging conflicts.

[0091] like Figure 5 As shown, it is assumed that the expressway is provided with a first section, a second section and a third section connected in sequence; the first section has three parallel lanes, the second section has three parallel lanes, and the third section has three parallel lanes; the first section is upstream of the merging area, the second section is a conflict area, and the third section is downstream of the merging area, the connecting line between the first section and the second section is a first intrusion line, and the connecting line between the second section and the third section is a second intrusion line; an acceleration lane is provided at the outer edge of the highway of the second section, the acceleration lane is arranged in parallel with the three lanes of the second section, the starting end of the acceleration lane is connected to the ending end of the ramp, and the ending end of the acceleration lane is connected to the conflict area; the merging vehicle enters the acceleration lane from the ramp; after accelerating in the acceleration lane, the merging vehicle enters the conflict area from the acceleration lane according to the driving instruction of the driver, and finally, the merging vehicle exits the conflict area and enters the downstream of the merging area.

[0092] The present invention uses Bayesian networks to describe the dependencies between random variables, and uses directed edges and nodes to respectively describe the multiple factors that affect traffic conflicts and traffic conflicts and various influencing factors, thereby constructing a traffic conflict network. The possibility of traffic conflicts is predicted. The possibility of merging conflicts is selected as the target node, and the merging speed, merging distance, relative speed, and relative distance related to the merging area conflict are used as observation nodes, and the target node and observation node data and the reverse are discretized through a data discretization processing method based on information entropy.

[0093] Embodiment 2: This embodiment provides a traffic conflict assessment system based on multivariate and Bayesian network, including: an acquisition module, which is configured to: acquire traffic conflict history data; construct a data set based on the traffic conflict history data; in the process of constructing the data set, classify the data set and supplement the missing data in the classified data set; divide the data set into a training set and a test set; a network construction module, which is configured to: construct a directed acyclic graph according to the directed connection relationship of each variable in the traffic conflict history data; the directed acyclic graph includes: a node set and an edge set, and the node set is a node in the data set. data; based on the directed acyclic graph, a Bayesian network is obtained; a training module is configured to: based on the training set, train and optimize the Bayesian network, the input value of the network is the data of the training set, and the output value of the network is the predicted probability value of the traffic conflict; a testing module is configured to: based on the test set, test the Bayesian network, when the test indicators all meet the requirements, it means that the trained Bayesian network meets the requirements, otherwise, the training set is replaced and trained again; a prediction module is configured to: input the traffic data to be predicted into the trained Bayesian network to obtain the predicted probability value of the traffic conflict.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Traffic conflict assessment method based on multivariate and Bayesian network, characterized by: include: Obtain historical data on traffic conflicts; Construct a dataset based on historical traffic conflict data; In the process of constructing the data set, the data set is classified and processed, and the missing data in the classified data set is supplemented; the data set is divided into a training set and a test set; wherein the classification process refers to: classifying each factor parameter into a traffic conflict generating category and a traffic conflict non-generating category; the classification process includes: calculating information entropy; calculating information gain based on information entropy; selecting the point with the largest information gain value as the split point to classify the data; According to the directed connection relationship of each variable in the historical data of traffic conflicts, a directed acyclic graph is formed; the directed acyclic graph includes: a node set and an edge set, and the node set is the data in the data set; based on the directed acyclic graph, a Bayesian network is obtained; Based on the training set, the Bayesian network is trained and optimized. The input value of the network is the data of the training set, and the output value of the network is the predicted probability value of the traffic conflict. Based on the test set, the Bayesian network is tested. If all the test indicators meet the requirements, it means that the trained Bayesian network meets the requirements. Otherwise, the training set is changed and trained again. The traffic data to be predicted is input into the trained Bayesian network to obtain the predicted probability value of traffic conflict.

2. The traffic conflict assessment method based on multivariate and Bayesian network as claimed in claim 1, characterized in that: Calculate information entropy, including: The definition formula of information entropy is: ； in: Represents the current sample data set, Indicates the sample data category, Indicates that in the current sample data set Middle range The number of internal values, Representation interval Inside The number of values; The smaller the value, the higher the purity of the partition interval. The C4.5 algorithm information gain is used to perform decision tree attribute partitioning. Assuming that there is a set of discrete attribute variables , the current data set will be split, and have A value, , divide the current data set into samples, each sample is recorded as .

3. The traffic conflict assessment method based on multivariate and Bayesian network as claimed in claim 2, characterized in that: Based on information entropy, information gain is calculated, including: ； in, represents information gain; Given the current sample dataset and discrete attributes , discrete attributes have different values, and arrange these values ​​in order, recorded as , then, from the set Randomly select split points , divide the current data set, and divide the data set into two parts, which is greater than the split point The part is recorded as Less than split point The part is recorded as ; At this time, select The midpoint of is taken as the pre-segmentation point. , ; Calculate the information gain of each segmentation point and select The split point with the largest value is taken as the final split point.

4. The traffic conflict assessment method based on multivariate and Bayesian network as claimed in claim 1, characterized in that: According to the directed connection relationship of each variable in the traffic conflict history data, a directed acyclic graph is formed, and the directed acyclic graph includes: a node set and an edge set, and the node set is the data in the data set; the nodes include: a relative speed node between the merging vehicle and the front vehicle, a relative speed node between the merging vehicle and the rear vehicle, a first difference node, a speed node of the merging vehicle, a distance node between the merging vehicle and the front vehicle, a distance node between the merging vehicle and the rear vehicle, a merging time node of the merging vehicle, a relative distance node between the merging vehicle and the front vehicle, a relative distance node between the merging vehicle and the rear vehicle, and a second difference node; wherein the first difference refers to: the difference between the relative speed of the merging vehicle and the front vehicle and the relative speed of the merging vehicle and the rear vehicle; the second difference refers to: the difference between the relative distance between the merging vehicle and the front vehicle and the relative distance between the merging vehicle and the rear vehicle.

5. The traffic conflict assessment method based on multivariate and Bayesian network as claimed in claim 4, characterized in that: The edge set refers to whether there is a relationship between nodes. If there is a relationship, there is an edge connecting the two nodes. Otherwise, there is no edge connecting the two nodes. The edge set includes: a connection edge between a relative speed node of a merging vehicle and a preceding vehicle and a traffic conflict node, a connection edge between a relative speed node of a merging vehicle and a following vehicle and a traffic conflict node, a connection edge between a first difference node and a traffic conflict node, a connection edge between a speed node of the merging vehicle and a traffic conflict node, a connection edge between a distance node of the merging vehicle and a preceding vehicle and a traffic conflict node, a connection edge between a distance node of the merging vehicle and a following vehicle and a traffic conflict node, a connection edge between a merging time node of the merging vehicle and a traffic conflict node, a connection edge between a relative distance node of the merging vehicle and a preceding vehicle and a traffic conflict node, a connection edge between a relative distance node of the merging vehicle and a following vehicle and a traffic conflict node, a connection edge between a second difference node and a traffic conflict node, a connection edge between a relative speed node of the merging vehicle and a preceding vehicle and a first difference node, a connection edge between a relative speed node of the merging vehicle and a following vehicle and a first difference node, a connection edge between a relative distance node of the merging vehicle and a preceding vehicle and a second difference node, and a connection edge between a relative distance node of the merging vehicle and a following vehicle and a second difference node.

6. The traffic conflict assessment method based on multivariate and Bayesian network as claimed in claim 1, characterized in that: Based on the training set, the Bayesian network is trained and optimized. The input value of the network is the data of the training set, and the output value of the network is the predicted probability value of the traffic conflict. Specifically, the Bayesian network is trained and optimized. The data of the training set refers to the merging speed, merging distance, relative speed and relative distance; Use the likelihood function to represent the impact of the current data set, and the parameters that need to be estimated are , is recorded as a random variable, , then, the Bayesian formula is used to calculate the parameters The posterior distribution of : ; Among them, the current data set , represents the likelihood function, Indicates the given parameter Post-variable The probability of the current data set, Indicates known Under the conditions The conditional distribution of express The prior distribution of Representation parameters The probability of when When the estimated parameter , the calculation method of the estimated parameters is: ; ; ; ;in, represents the parameters that need to be estimated, Indicates different , , value, express The maximum value of express The maximum value of express The maximum value of , , represents the number of different conditions. represents the parameters to be estimated after solving the problem, Under the condition that C The conditional distribution of Indicates conditions, Indicates the condition The size of the case, represents probability; The solution is based on the three assumptions of global parameter independence, local parameter independence, and parameter Dirichlet distribution. ; Based on three assumptions, the parameters to be estimated are: ;in, Represents the prior distribution The hyperparameters of Indicates that in data set C, the target node is satisfied , The number of values ​​of represents the parameter to be estimated.

7. The traffic conflict assessment method based on multivariate and Bayesian network as claimed in claim 1, characterized in that: After the Bayesian network is tested, before the traffic data to be predicted is input into the trained Bayesian network, the following steps are further included: Determine whether a conflict occurs between the merging vehicle and the vehicle in front of the target lane, or whether a conflict occurs between the merging vehicle and the vehicle behind the target lane; if a conflict occurs between the merging vehicle and the vehicle in front of the target lane, or a conflict occurs between the merging vehicle and the vehicle behind the target lane, input the traffic data to be predicted into the trained Bayesian network; if not, do not input the traffic data to be predicted into the trained Bayesian network.

8. The traffic conflict assessment method based on multivariate and Bayesian network as claimed in claim 7, characterized in that: The criteria for judging whether a conflict occurs between the merging vehicle and the vehicle ahead in the target lane are: In the process of entering the conflict area, for the conflict between the merging vehicle and the vehicle in front of the target lane, the time when the tail of the vehicle A in front of the target lane leaves the first invasion line is recorded as T1, and the time when the head of the merging vehicle B enters the first invasion line is recorded as T2, then PET1=T2-T1. If PET1 is less than the set threshold, it means that there will be a conflict between the merging vehicle and the vehicle in front of the target lane; The criterion for judging whether a conflict occurs between the merging vehicle and the rear vehicle in the target lane is as follows: in the process of exiting the conflict area, for the conflict between the merging vehicle D and the rear vehicle C in the target lane, the time when the tail of the merging vehicle D leaves the second invasion line is recorded as T3, and the time when the head of the rear vehicle C in the target lane enters the second invasion line is recorded as T4, then PET2=T4-T3. If PET2 is less than the set threshold, it means that a conflict will occur between the merging vehicle and the rear vehicle in the target lane.

9. Traffic conflict assessment system based on multivariate and Bayesian network, characterized by: include: An acquisition module is configured to: acquire traffic conflict history data; Construct a dataset based on historical traffic conflict data; In the process of constructing the data set, the data set is classified and processed, and the missing data in the classified data set is supplemented; the data set is divided into a training set and a test set; wherein the classification process refers to: classifying each factor parameter into a traffic conflict generating category and a traffic conflict non-generating category; the classification process includes: calculating information entropy; calculating information gain based on information entropy; selecting the point with the largest information gain value as the split point to classify the data; A network construction module is configured to: construct a directed acyclic graph according to the directed connection relationship of each variable in the traffic conflict history data; the directed acyclic graph includes: a node set and an edge set, and the node set is the data in the data set; based on the directed acyclic graph, a Bayesian network is obtained; A training module is configured to: perform training optimization on the Bayesian network based on a training set, wherein the input value of the network is the data of the training set, and the output value of the network is the predicted probability value of the traffic conflict; The testing module is configured to: test the Bayesian network based on the test set. When all the test indicators meet the requirements, it means that the trained Bayesian network meets the requirements. Otherwise, the training set is replaced and trained again. The prediction module is configured to: input the traffic data to be predicted into the trained Bayesian network to obtain the predicted probability value of the traffic conflict.

Images