Road section-vehicle group accident risk combined prediction and analysis method and system

Abstract

The invention relates to a road section-vehicle group accident risk combined prediction and analysis method and system, and the method comprises the steps: carrying out the vehicle group division based on vehicle track data, obtaining the geometric features of a road section, carrying out the road section division, and obtaining a risk impact feature data set of each vehicle group and road section; vehicle group accident risk prediction and road section accident risk prediction are carried out based on the risk influence feature data set of each vehicle group, and sensitivity analysis is carried out to obtain a risk visualization result; based on the risk influence characteristic data set of the vehicle group and the road section and the prediction result of the vehicle group accident risk and the road section accident risk, combined prediction is carried out, sensitivity analysis is carried out, the risk visualization result is verified and corrected based on the sensitivity analysis result, and the system is used for realizing the method. Compared with the prior art, the interaction between the road section and the vehicle group is considered, and the prediction accuracy of the road section and vehicle group accident risk prediction model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of continuous traffic accident risk prediction, and particularly to a method and system for jointly predicting and analyzing the accident risks of road sections and vehicle groups. Background Art

[0003] In the modern traffic system, continuous traffic flow facilities, as the main arteries of the national transportation system, are characterized by high driving speeds and complex traffic compositions. Once a traffic accident occurs on such facilities, it will lead to a significant decline in traffic capacity and may trigger secondary accident risks. Active safety control, as the core means to improve traffic safety, highly depends on the accurate analysis of accident risk mechanisms. However, existing research has obvious limitations in the fineness and accuracy of accident risk mechanism analysis, making it difficult to meet the real-time dynamic control requirements and becoming the key bottleneck restricting the improvement of the active safety efficiency of traffic flow. Traditional accident risk research mainly relies on section detectors to obtain traffic flow parameters, and the data has significant defects of time aggregation and space discontinuity. Such data can only reflect the characteristics of local sections and cannot fully depict the spatio-temporal continuity of vehicle running states, resulting in insufficient accuracy in identifying accident risk influencing factors. For example, section data is difficult to capture key risk characteristics such as the dynamic interaction behavior of vehicle groups and the acceleration change pattern, nor can it effectively represent the coupling relationship between road section geometric characteristics and vehicle group behavior. With the development of sensing technologies, a fusion sensing system based on radar, video surveillance, and trajectory fitting software has been able to achieve full-time and full-domain continuous sensing of traffic parameters, providing technical support for obtaining full-sample data such as flow density, speed distribution, and vehicle trajectories. However, the existing research still confines the utilization of new data resources to a single analysis dimension, lacking both the joint analysis of the micro-behavior of vehicle groups and the macro-situation of road sections and the establishment of a risk interaction mechanism between the two, resulting in insufficient pertinence of control measures. Moreover, most of the existing technologies regard the prediction model based on deep learning as a black box without further analyzing the internal mechanism, making the current accident risk research still stay at the macro-situation evaluation level and unable to support the refined safety control requirements of "vehicle-road" coordination.

[0004] Therefore, it is a technical problem to be solved to provide a method for analyzing and predicting the accident risks of road sections and vehicles that can meet the refined safety control requirements of "vehicle-road" coordination. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for jointly predicting and analyzing the accident risks of road sections and vehicle groups to overcome the defects of the existing technologies that cannot achieve refined safety management of "vehicle-road" coordination.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to a first aspect of the present invention, there is provided a method for jointly predicting and analyzing the accident risk of road sections and vehicle groups, the method comprising:

[0008] Obtain continuous-flow vehicle trajectory data, perform vehicle group division based on the vehicle trajectory data, obtain road section geometric features for road section division, and obtain a risk impact feature data set for each vehicle group and road section; wherein, the vehicle group risk impact feature data set includes vehicle group internal features, road section geometric features, and road section accident risk values; the road section risk impact feature data set includes the speed and acceleration of the vehicle group and the traffic and geometric features of the corresponding road section;

[0009] Respectively construct a vehicle group accident risk prediction model and a road section accident risk prediction model, perform vehicle group accident risk prediction and road section accident risk prediction respectively based on the risk impact feature data set of each vehicle group, and obtain a risk visualization result by using global or local sensitivity analysis;

[0010] Based on the risk impact feature data set of the vehicle group and the road section, the results of the vehicle group accident risk and road section accident risk prediction, combine the vehicle group accident risk prediction model and the road section accident risk prediction model, use the combined model for joint prediction of the road section-vehicle group accident risk, and perform sensitivity analysis, and verify and correct the risk visualization result based on the sensitivity analysis result.

[0011] As a preferred technical solution, the method for vehicle group division is: calculate the TTC value of each vehicle based on the vehicle trajectory data, and select the vehicles with TTC values less than a first preset value at the same moment and divide them into a high-risk vehicle group; select the vehicles with TTC values less than a second preset value and greater than or equal to the first preset value at the same moment and divide them into a medium-risk vehicle group; select the vehicles with TTC values greater than or equal to the first preset value at the same moment and divide them into a low-risk vehicle group.

[0012] As a preferred technical solution, the method for road section division is: use the change points of the road section geometric features as road section segmentation points, and the geometric features include the number of lanes, road section type, and curvature. Among them, performing road section segmentation based on the number of lanes means selecting the entrance and exit ramps where the number of lanes changes as segmentation points; performing road section segmentation based on the road section type means selecting the entrance and exit ramps where the road section type changes as segmentation points; performing road section segmentation based on curvature means taking the entire curve and smooth curve as the same road section.

[0013] As a preferred technical solution, the method for predicting the accident risk of vehicle groups is:

[0014] Select the road section geometric features and the internal geometric features of the target vehicle group in the downstream road section of the target road section, and calculate the accident risk of the downstream road section; the internal geometric features of the vehicle group include the average speed of the vehicle group, the average acceleration of the vehicle group, the standard deviation of the vehicle group speed, the extreme speed of the vehicle group, the extreme acceleration of the vehicle group, the standard deviation of the vehicle group acceleration, the proportion of large vehicles in the vehicle group, and the lane-changing rate of the vehicle group; the road section combination features include the type of the road section where it is located, the type of the downstream road section, the current number of lanes, and the number of downstream lanes.

[0015] Use the road section geometric features, the internal geometric features of the vehicle group, and the accident risk of the downstream road section of the downstream road section as the input of the vehicle group accident risk prediction model, and output the accident risk value of the target vehicle group at the next moment.

[0016] As a preferred technical solution, the method for calculating the accident risk of the downstream road section is:

[0017] Calculate the TTC value of each vehicle in the downstream road section, and its expression is:

[0018]

[0019] where x a,b represents the spatial distance between two vehicles, v a (t) represents the speed of the target vehicle, and v b (t) represents the speed of the vehicle in front of the target vehicle;

[0020] Based on the TTC value, calculate the accident risk of the downstream road section, and its expression is:

[0021]

[0022] where TTC a,b (j) represents the TTC value between vehicle a and vehicle b at the j-th moment.

[0023] As a preferred technical solution, the method for predicting the road section accident risk is:

[0024] Obtain the internal geometric features of the high-risk vehicle group and the medium-risk vehicle group in the downstream road section of the target road section, as well as the road section traffic features, road section geometric features, and accident risk sequence; among them, the internal geometric features of the vehicle group also include the risk value of the high-risk vehicle group and the propagation distance of the high-risk vehicle group; the road section traffic features include the average speed of the road section, the standard deviation of the road section acceleration, the extreme speed of the road section, the average acceleration of the road section, and the extreme acceleration of the road section; the accident risk sequence includes the accident risk at each moment in multiple time periods.

[0025] The internal geometric features of the high-risk vehicle group and the medium-risk vehicle group, as well as the road section traffic features, road section geometric features, and accident risk sequence are used as the inputs of the road section accident risk prediction model; the road section accident risk prediction model includes two LSTM layers, two fully connected layers, and a Dense layer;

[0026] The accident risk sequence is input into the road section accident risk prediction model from the first LSTM layer and processed by the first LSTM layer and the second LSTM layer. After the internal geometric features of the high-risk vehicle group and the medium-risk vehicle group, and the road section traffic features and road section geometric features are input into the road section accident risk prediction model through the first fully connected layer and processed using the Relu function, they are fused with the output of the second LSTM layer;

[0027] The Dense layer receives the fused result and outputs the target road section accident risk value through the second fully connected layer.

[0028] As a preferred technical solution, the accident risk sequence input into the first LSTM layer also needs to be processed by Kalman filtering.

[0029] As a preferred technical solution, the method for joint prediction of road section-vehicle group accident risk includes accident risk prediction and parallel prediction of vehicle group-road section-vehicle group;

[0030] Among them, the steps of the accident risk prediction of vehicle group-road section-vehicle group include:

[0031] The accident risk value of the downstream road section of the road section where the vehicle group is located, the road section geometric features of the downstream road section of the road section where the vehicle group is located, and the internal geometric features of the vehicle group are used as the inputs of the vehicle group accident risk prediction model to predict the accident risk value of the vehicle group at the next moment;

[0032] If the accident risk value of the vehicle group at the next moment is greater than the threshold, the road section accident risk prediction model is used to predict the accident risk value of the road section where the vehicle group is located to obtain the accident risk prediction value of the road section where the vehicle group is located;

[0033] The predicted accident risk prediction value of the road section where the vehicle group is located, the internal geometric features of the high-risk vehicle group and the medium-risk vehicle group, as well as the road section traffic features and road section geometric features are used as the inputs of the vehicle group accident risk prediction model of the upstream road section of the road section where the vehicle group is located to predict the accident risk value of the vehicle group at the next moment of the upstream vehicle group.

[0034] As a preferred technical solution, the method of parallel prediction is:

[0035] The output layer of the road section accident risk prediction model and the accident risk prediction model are connected using a fully connected layer to obtain a parallel model;

[0036] Taking the internal characteristics of the vehicle group, the geometric characteristics of the road section, the traffic characteristics of the road section, the accident risk of the downstream road section, and the accident risk sequence as the inputs of the parallel model, and outputting the accident risk value of the target vehicle group arriving at the target road section.

[0037] According to the second aspect of the present invention, a road section-vehicle group accident risk joint prediction and analysis system is provided, and the system is used to implement the above method.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] 1) This application considers the interaction between the road section and the vehicle group. When taking the road section as the research unit, it considers the risk propagation of the downstream high-risk vehicle group, and when taking the vehicle group as the research unit, it considers the risk propagation of the downstream road section. It can not only improve the prediction accuracy of the road section and vehicle group accident risk prediction models, but also deeply analyze the evolution path of the road section-vehicle group risk, the dissipation and propagation law, and the interaction and coupling relationship between the accident risks of both, significantly improving the two-dimensional risk prediction accuracy; by establishing a computable model of the risk propagation path, the risk cascade analysis can be accurately obtained from the traffic flow disturbance to the road section response, providing a decision support system for the active safety prevention and control of the vehicle group-road section, including risk source tracing and positioning, propagation intensity quantification, and dissipation time scheduling.

[0040] 2) The present invention also provides a combined prediction model, which innovatively realizes the spatio-temporal coordination of risk prediction, conducts closed-loop feedback analysis in the form of "vehicle group → road section → vehicle group", enabling the accident risk prediction model to have the ability to dynamically adapt to complex traffic scenarios; it also conducts visual analysis on the prediction process through sensitivity analysis to ensure the interpretability of the model; this combined prediction method enables the early warning response of the upstream and downstream road sections to be triggered synchronously when a high-risk vehicle group is detected; and the road section risk early warning can reversely optimize the vehicle group control strategy. Compared with the single-model application scenario, this coordination mechanism makes the control measures take effect faster.

[0041] 3) The present invention uses feature importance analysis and SHAP value analysis to reverse-deconstruct the model decision logic, transforming the complex non-linear relationship between traffic flow disturbance, road section characteristics, and risk propagation into quantifiable and interpretable association rules, which can not only realize risk prediction but also analyze the evolution logic of risk formation. Brief Description of the Drawings

[0042] Figure 1 It is the flowchart of the method of the present invention;

[0043] Figure 2 It is the technical roadmap of the present invention;

[0044] Figure 3 It is the first-order weight of the vehicle group accident risk prediction model of the present invention;

[0045] Figure 4 is the total benefit of the vehicle group accident risk prediction model of the present invention;

[0046] Figure 5 is the SHAP macro importance diagram of the road section accident risk prediction model of the present invention;

[0047] Figure 6 is the SHAP macro force diagram of the road section accident risk prediction model of the present invention. Specific Embodiments

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the ordinary meanings understood by those with ordinary skills in the technical field to which this application belongs. The terms "a", "one", "kind", "the" and the like involved in this application do not indicate a quantity limit and can represent a single or plural number. The terms "including", "comprising", "having" and any variations thereof involved in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may further include steps or units not listed, or may further include other steps or units inherent to these processes, methods, products or devices. The terms "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, "A and / or B" may represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after. The terms "first", "second", "third" and the like involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0050] Embodiment 1

[0051] Active safety control is the key to improving traffic safety, and active safety control requires accurate analysis of accident risk mechanisms as a support. The risk of consecutive-flow accidents has been widely studied by domestic and foreign scholars. Previous studies mainly relied on data from section detectors, and the influencing factors considered were limited to traffic parameters and geometric features on the target road section. The continuous movement of vehicles and the influence of interactions with nearby vehicles were usually ignored, and the accuracy and interpretability of predictions were not taken into account. Currently, the research on accident risk with road sections as the research unit is relatively mature, but the research with vehicle groups as the research unit is still in its infancy, and there is almost no research on the interactive influence of accident risks between road sections and vehicle groups. In addition, although some studies on the risk of consecutive-flow accidents have begun to consider the spread of accident risks, the current research is mostly limited to a single research unit and does not consider integrating vehicle groups and road sections for research. And the prediction model based on deep learning is regarded as a black box without further analyzing the internal mechanism. Based on considering the spread of accident risks, this invention conducts an upstream study on the accident risks of vehicle groups and road sections, uses machine learning and deep learning methods, and at the same time uses algorithms to analyze the black box model, aiming to achieve the prediction and analysis of the accident risk situations of road sections and vehicle groups. Therefore, this invention takes vehicle groups and road sections as research units respectively, establishes machine learning and deep learning models for prediction and index comparison and conducts visual analysis, determines the influencing factors affecting the accident risks of vehicle groups and road sections. Specifically, this invention conducts vehicle group-road section accident risk analysis according to the flow chart as shown in Figure 1 and the framework diagram as shown in Figure 2 , which specifically includes:

[0052] S1. Division of road sections and vehicle groups.

[0053] S11. Obtain consecutive-flow vehicle trajectory data, and conduct vehicle group division based on the vehicle trajectory data. Among them, the vehicle group risk impact feature data set includes internal features of the vehicle group, road section geometric features, and road section accident risk values.

[0054] Specifically, calculate the TTC value of each vehicle based on the vehicle trajectory data, and select the vehicles with TTC values less than the first preset value at the same moment and divide them into high-risk vehicle groups. In this embodiment, the first preset value is set to 1.5 s. Select the vehicles with TTC values less than the second preset value and greater than or equal to the first preset value at the same moment and divide them into medium-risk vehicle groups. Select the vehicles with TTC values greater than or equal to the first preset value at the same moment and divide them into low-risk vehicle groups.

[0055] S12. Obtain road section geometric features for road section division, and obtain the risk impact feature data set of each vehicle group and road section. Among them, the road section risk impact feature data set includes the speed and acceleration of the vehicle group and the traffic features and geometric features of the corresponding road section.

[0056] Specifically, through high-precision map point marking combined with satellite map observation, the coordinates of the road centerline in the UTM coordinate system were obtained, and the coordinate points of the change in the geometric features of the road section were used as the road section segmentation points. The geometric features include the number of lanes, road section type, and curvature. Among them, segmenting the road section based on the number of lanes means selecting the entrance and exit ramps where the number of lanes changes as the segmentation points. Segmenting the road section based on the road section type means selecting the entrance and exit ramps where the road section type changes as the segmentation points. Segmenting the road section based on curvature means taking the entire curve and smooth curve as the same road section.

[0057] S2. Model construction and accident risk prediction.

[0058] S21. Model construction:

[0059] S211. Respectively construct a vehicle group accident risk prediction model and a road section accident risk prediction model. The selected base model types for the vehicle group accident risk prediction model include: random forest model, support vector machine regression model, convolutional neural network model, and deep neural network model; the base model types for the road section accident risk prediction model include: support vector machine regression model, convolutional neural network model, and long short-term memory recurrent neural network.

[0060] S212. Obtain a historical risk impact feature dataset of multiple vehicle group-road combinations, and generate output labels corresponding to the vehicle group accident risk prediction model and the road section accident risk prediction model based on this historical risk impact feature dataset.

[0061] The process of obtaining the output label of the vehicle group accident risk prediction model is as follows:

[0062]

[0063] Among them, TTC i represents the collision time value between the i-th vehicle and the (i - 1)-th vehicle within the target vehicle group, that is, the TTC value; count(vehicle groupe) represents the total number of vehicles in the target vehicle group; count(high risk) represents the number of vehicles with high collision risk in the target vehicle group.

[0064] The process of obtaining the output label of the road section accident risk prediction model is as follows:

[0065]

[0066] Among them, TTC i (j) represents the TTC value between the i-th vehicle and the (i - 1)-th vehicle at the j-th moment on the road section.

[0067] S213. Based on the historical risk impact feature dataset and the corresponding output labels, 80% of the data is used as the training set and the remaining 20% is used as the validation set. Among them, the accident risk event sequence data is not processed, and the traffic impact factor data is normalized by removing the mean and normalizing the variance, so as to improve the generalization ability of the model. Perform performance screening on the above base model, and use the mean absolute error, mean absolute percentage error, and determination coefficient as evaluation indicators for performance evaluation. The detailed results are shown in Table 1 and Table 2.

[0068] Table 1 Performance Screening Results of Vehicle Group Accident Risk Prediction Model

[0069] Model MAE MAPE <![CDATA[R 2 > Random Forest 0.22 19.87％ 0.78 Support Vector Machine Regression 0.25 27.93％ 0.71 Convolutional Neural Network 0.22 18.99％ 0.78 Deep Neural Network 0.20 16.29％ 0.81

[0070] As can be seen from Table 1, when the base model is a deep neural network, the corresponding model has the best performance in predicting the accident risk of vehicle groups. Therefore, the deep neural network is selected as the base model of the vehicle group accident risk prediction model.

[0071] Table 2 Performance Screening Results of Road Section Accident Risk Prediction Model

[0072]

[0073] As can be seen from Table 2, when the base model is a long short-term memory recurrent neural network, the corresponding model has the best performance in predicting the accident risk of road sections. Therefore, the long short-term memory recurrent neural network is selected as the base model of the vehicle group accident risk prediction model.

[0074] S22. Accident risk prediction:

[0075] Based on the risk impact feature dataset of each vehicle group, the accident risk of the vehicle group is predicted. The detailed steps are as follows:

[0076] S221. Select the road section geometric features of the downstream road section of the target vehicle group and the internal geometric features of the target vehicle group, and calculate the accident risk of the downstream road section. The steps for calculating the accident risk are as follows:

[0077] Calculate the TTC value of each vehicle in the downstream road section. Its expression is:

[0078]

[0079] where, x a,b represents the spatial distance between two vehicles, v a (t) represents the speed of the target vehicle, and v b (t) represents the speed of the vehicle in front of the target vehicle;

[0080] Based on the TTC value, calculate the accident risk of the downstream road section. Its expression is:

[0081]

[0082] Among them, TTC a,b (j) represents the TTC value between vehicle a and vehicle b at the j-th moment.

[0083] Among them, the internal geometric features of the vehicle group include the average speed of the vehicle group, the average acceleration of the vehicle group, the standard deviation of the vehicle group speed, the extreme speed of the vehicle group, the extreme acceleration of the vehicle group, the standard deviation of the vehicle group acceleration, the proportion of large and medium-sized vehicles in the vehicle group, and the lane-changing rate of the vehicle group.

[0084] The combined features of the road section include the type of the current road section, the type of the downstream road section, the number of current lanes, and the number of downstream lanes. The details are shown in Table 3.

[0085] Table 3 Dataset of risk impact features of vehicle groups

[0086] Variable Symbol Meaning speed_mean Average Speed of Vehicle Group speed_std Standard Deviation of Vehicle Group Speed speed_max Extreme Speed of Vehicle Group tan_acc_mean Average Acceleration of Vehicle Group tan_acc_std Standard Deviation of Vehicle Group Acceleration tan_acc_max Extreme Acceleration of Vehicle Group Acceleration type_ratio Ratio of Large Vehicles in Vehicle Group change_lane_ratio Lane Changing Rate of Vehicle Group road_type_1 Current Road Section Type road_type_2 Downstream Road Section Type lane_num_1 Number of Lanes in Current Road Section lane_num_2 Number of Lanes in Downstream Road Section road_10s_risk_1 Accident Risk of Downstream Road Section in 10s road_5s_risk_1 Accident Risk of Downstream Road Section within 5s

[0087] S222. Use the road section geometric features, the internal geometric features of the vehicle group, and the accident risk of the downstream road section as the input of the vehicle group accident risk prediction model, and output the vehicle group accident risk value at the next moment of the target vehicle group.

[0088] S223. Use Sobol to conduct a global sensitivity analysis with all the risk impact features in Table 3 and Table 4 as the objects of sensitivity analysis. The Sobol algorithm can obtain the overall impact after the input variables and the interaction of variables, so as to obtain the first-order effect index S1 and the T total effect index. Its expression is:

[0089]

[0090] Among them, X i represents the i-th risk impact feature; E~i(Y) represents the expectation of the output variable under the condition of fixing all other risk impact features; Var Xi represents the variance on the i-th risk impact feature.

[0091]

[0092] Among them, V ～i represents the variance of the predicted value under the condition of fixing all other input variables.

[0093] S224. Use LIME to conduct a local sensitivity analysis with all the risk impact features in Table 3 as the objects of sensitivity analysis. Lime constructs a locally interpretable model (linear model) to approximate the behavior of the complex model around a specific sample, so as to obtain the approximate influence coefficient. Its calculation expression is:

[0094]

[0095] Among them, x represents the sample to be explained (i.e., the risk impact feature); x ′ represents the neighborhood samples of x, generating a specific number of new values at x using the standard normal distribution; g represents the interpretable model; G represents the hypothesis space of the interpretable model; Ω(g) represents a regularization term used to control the complexity of the model and prevent overfitting.

[0096] After steps S223 - S224, the results as shown in Figure 3 and Figure 4 can be obtained. The results show that the most significant influencing factor of the accident risk of the vehicle group is the standard deviation of speed, and it is related to both the accident risk and geometric characteristics of the downstream section, that is, the type of section where the vehicle group is located and the accident risk value have a relatively significant impact on the accident risk of the vehicle group. Therefore, it can be seen that the characteristics of the section will affect the accident risk of the vehicle group.

[0097] Based on the risk impact feature dataset of each section, the accident risk prediction of the section is carried out. The detailed steps are as follows:

[0098] S223. Obtain the internal geometric characteristics of the vehicle group of high - risk vehicle groups and medium - risk vehicle groups in the downstream section of the target section, as well as the section traffic characteristics, section geometric characteristics, and accident risk sequence.

[0099] Among them, the internal geometric characteristics of the vehicle group also include the risk value of the high - risk vehicle group and the propagation distance of the high - risk vehicle group; the section traffic characteristics include the section average speed, section acceleration standard deviation, section extreme speed, section average acceleration, and section extreme acceleration; the accident risk sequence includes the accident risk at each moment in multiple time periods, and the detailed information is shown in Table 4.

[0100] Table 4 Section risk impact feature dataset

[0101]

[0102]

[0103] S224. Take the internal geometric characteristics of the high - risk vehicle group and medium - risk vehicle group, and the section traffic characteristics, section geometric characteristics, and accident risk sequence as the input of the section accident risk prediction model.

[0104] The above accident risk prediction model for the road section is a long short-term memory recurrent neural network created using the Sequential class, including two LSTM layers, two fully connected layers, and a Dense layer. For the first LSTM layer, return_sequences = True is set to maintain the output in the form of a time series for subsequent layers to continue processing. The second LSTM layer no longer retains the time series output and outputs a single vector.

[0105] S225. After processing the accident risk sequence through the Kalman filter, it is input into the accident risk prediction model for the road section from the first LSTM layer and processed by the first LSTM layer and the second LSTM layer. The internal geometric features of the vehicle group, the road traffic features, and the road geometric features of the high-risk vehicle group and the medium-risk vehicle group are input into the accident risk prediction model for the road section through the first fully connected layer and processed using the Relu function, and then fused with the output of the second LSTM layer.

[0106] S226. The Dense layer receives the fused result and outputs the accident risk value of the target road section through the second fully connected layer (this second fully connected layer only includes one output unit).

[0107] After completing the accident risk prediction for the vehicle group and the road section risk prediction, global or local sensitivity analysis is used to obtain the risk visualization result. Specifically, the process is as follows:

[0108] S229. Calculate the Shapley value of each risk impact feature in Table 4, and its expression is:

[0109]

[0110] Among them, φ i represents the Shapley value of the risk impact feature i; N represents the set of all risk impact features; S represents a subset of risk impact features that does not include i; v(S) represents the predicted value of the model under the feature subset S.

[0111] After being processed by step S229, the results can be obtained as shown in Figure 5 and Figure 6 It can be seen from Figure 5 and Figure 6 that the accident risk of the road section is related to the traffic features on the road section, and is also related to the accident risk and traffic features of the downstream vehicle group. Moreover, the high-risk vehicle group in the downstream road section has a significant impact on the accident risk of the target road section, which will make the stability of the overall traffic flow low and increase the accident risk of the road section.

[0112] Whether taking the vehicle group as a unit or the road section as a unit, the greater the accident risk of the downstream, the more significant the impact on the accident risk of the upstream.

[0113] S3. Joint prediction of road section - vehicle group accident risks.

[0114] Based on the risk impact feature datasets of vehicle groups and road sections, and the results of vehicle group accident risk and road section accident risk predictions, the vehicle group accident risk prediction model and the road section accident risk prediction model are combined. The combined model is used for joint prediction of road section - vehicle group accident risks, and sensitivity analysis is carried out. Based on the results of the sensitivity analysis, the risk visualization results are verified and corrected. The method of this joint prediction includes accident risk prediction and parallel prediction of vehicle group - road section - vehicle group. The detailed process is as follows:

[0115] S31. Accident risk prediction of vehicle group - road section - vehicle group:

[0116] S311. Take the accident risk value of the downstream road section of the road section where the vehicle group is located, the road geometry features of the downstream road section of the road section where the vehicle group is located, and the internal geometry features of the vehicle group as the input of the vehicle group accident risk prediction model, and predict the vehicle group accident risk value at the next moment.

[0117] S312. If the vehicle group accident risk value at the next moment is greater than the threshold, use the road section accident risk prediction model to predict the road section accident risk value of the road section where the vehicle group is located.

[0118] S313. Take the predicted road section accident risk prediction value of the road section where the vehicle group is located, the internal geometry features of high - risk vehicle groups and medium - risk vehicle groups, and the road traffic features and road geometry features as the input of the vehicle group accident risk prediction model of the upstream road section of the road section where the vehicle group is located, and predict the vehicle group accident risk value at the next moment of the upstream vehicle group.

[0119] S32. Parallel prediction.

[0120] S321. Connect the output layers of the road section accident risk prediction model and the accident risk prediction model using a fully - connected layer to obtain a parallel model.

[0121] S322. Take the internal features of the vehicle group, road geometry features, road traffic features, downstream road section accident risks, and accident risk sequences as the input of the parallel model, and output the accident risk value of the target vehicle group arriving at the target road section.

[0122] Embodiment 2

[0123] The above is the introduction of the method embodiment. The following further illustrates the solution of the present invention through the embodiment of the road section - vehicle group accident risk joint prediction and analysis system. Those skilled in the art can clearly understand that for the sake of convenience and brevity of description, the specific working process described can refer to the corresponding process in the foregoing method embodiment, and will not be elaborated here.

[0124] The road-vehicle group accident risk joint prediction and analysis system provided by the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or computer program instructions loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0125] Multiple components in the device are connected to the I / O interface, including: an input unit, such as a keyboard, mouse, etc.; an output unit, such as various types of displays, speakers, etc.; a storage unit, such as a disk, optical disc, etc.; and a communication unit, such as a network card, modem, wireless communication transceiver, etc. The communication unit allows the device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0126] The processing unit executes the various methods and processes described above, such as methods S1 to S3. For example, in some embodiments, methods S1 to S3 can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device via the ROM and / or the communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S3 described above can be executed. Alternatively, in other embodiments, the CPU can be configured to execute methods S1 to S3 in any other suitable manner (e.g., by means of firmware).

[0127] The functions described above herein can be at least partially executed by one or more hardware logic components. For example, by way of non-limitation, exemplary types of hardware logic components that can be used include: field programmable gate arrays (FPGA), application specific integrated circuits (ASIC), application specific standard products (ASSP), system on a chip (SOC), complex programmable logic devices (CPLD), and so on.

[0128] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing devices, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowchart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, executed partially on the machine and partially on a remote machine as an independent software package, or executed entirely on a remote machine or server.

[0129] In the context of the present invention, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0130] As described above, only the specific embodiments of the present invention are provided, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily conceive of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for jointly predicting and analyzing the accident risk of road sections - vehicle groups, characterized in that The method described above includes: Obtaining continuous-flow vehicle trajectory data, dividing vehicle groups based on the vehicle trajectory data, obtaining road section geometric features for road section division, and obtaining a risk impact feature data set for each vehicle group and road section; among them, the vehicle group risk impact feature data set includes vehicle group internal features, road section geometric features, and road section accident risk values; the road section risk impact feature data set includes the speed and acceleration of the vehicle group and the traffic and geometric features of the corresponding road section; Constructing a vehicle group accident risk prediction model and a road section accident risk prediction model respectively, performing vehicle group accident risk prediction and road section accident risk prediction based on the risk impact feature data set of each vehicle group respectively, and obtaining a risk visualization result by using global or local sensitivity analysis; Based on the risk impact feature data set of the vehicle group and road section, the results of vehicle group accident risk and road section accident risk prediction, combining the vehicle group accident risk prediction model and the road section accident risk prediction model, using the combined model for joint prediction of road section-vehicle group accident risk, and performing sensitivity analysis, verifying and correcting the risk visualization result based on the sensitivity analysis result.

2. A method for jointly predicting and analyzing the accident risks of road sections and vehicle groups according to claim 1, characterized in that, The method for dividing vehicle groups is as follows: calculating the TTC value of each vehicle based on the vehicle trajectory data, and selecting the vehicles with TTC values less than a first preset value at the same moment to be divided into a high-risk vehicle group; Selecting the vehicles with TTC values less than a second preset value and greater than or equal to the first preset value at the same moment to be divided into a medium-risk vehicle group; Selecting the vehicles with TTC values greater than or equal to the first preset value at the same moment to be divided into a low-risk vehicle group.

3. The joint prediction and analysis method for road section - vehicle group accident risks according to claim 1, characterized in that, The method for road section division is as follows: taking the change points of road section geometric features as road section segmentation points, where the geometric features include the number of lanes, road section type, and curvature. Among them, dividing the road section based on the number of lanes means selecting the entrance and exit ramps with a change in the number of lanes as segmentation points; dividing the road section based on the road section type means selecting the entrance and exit ramps with a change in the road section type as segmentation points; dividing the road section based on the curvature means taking the entire curve and smooth curve as the same road section.

4. A method for jointly predicting and analyzing road section - vehicle group accident risks according to claim 1, characterized in that, The method for predicting vehicle group accident risk is as follows: Selecting the road section geometric features of the downstream road section of the road section where the target vehicle group is located and the internal geometric features of the target vehicle group, and calculating the accident risk of the downstream road section; The internal geometric features of the vehicle group include the average speed of the vehicle group, the average acceleration of the vehicle group, the standard deviation of the vehicle group speed, the extreme speed of the vehicle group, the extreme acceleration of the vehicle group, the standard deviation of the vehicle group acceleration, the proportion of large and medium-sized vehicles in the vehicle group, and the lane-changing rate of the vehicle group; the road section combination features include the road section type where it is located, the road section type of the downstream road section, the current number of lanes, and the number of lanes of the downstream road section; Taking the road section geometric features of the downstream road section, the internal geometric features of the vehicle group, and the accident risk of the downstream road section as the input of the vehicle group accident risk prediction model, and outputting the vehicle group accident risk value at the next moment of the target vehicle group.

5. A method for jointly predicting and analyzing the accident risks of road sections and vehicle groups according to claim 4, characterized in that, The method for calculating the accident risk of the downstream road section is as follows: Calculating the TTC value of each vehicle in the downstream road section, and its expression is: where x a,b represents the spatial distance between two vehicles, v a (t) represents the speed of the target vehicle, v b (t) represents the speed of the vehicle in front of the target vehicle; Based on the TTC value, calculating the accident risk of the downstream road section, and its expression is: Among them, TTC a,b (j) represents the TTC value between vehicle a and vehicle b at the j-th moment.

6. The method for jointly predicting and analyzing road section - vehicle group accident risks according to claim 5, characterized in that, The method for predicting road section accident risk is as follows: Obtain the internal geometric characteristics of high-risk vehicle groups and medium-risk vehicle groups in the downstream section of the target section, as well as the section traffic characteristics, section geometric characteristics, and accident risk sequence; wherein, the internal geometric characteristics of the vehicle group also include the risk value of the high-risk vehicle group and the propagation distance of the high-risk vehicle group; the section traffic characteristics include the average speed of the section, the standard deviation of the section acceleration, the extreme speed of the section, the average acceleration of the section, and the extreme acceleration of the section; the accident risk sequence includes the accident risk at each moment in multiple time periods. Use the internal geometric characteristics of the high-risk vehicle group and the medium-risk vehicle group, the section traffic characteristics, the section geometric characteristics, and the accident risk sequence as the input of the section accident risk prediction model; the section accident risk prediction model includes two LSTM layers, two fully connected layers, and a Dense layer. The accident risk sequence is input into the section accident risk prediction model from the first LSTM layer and processed by the first LSTM layer and the second LSTM layer. After the internal geometric characteristics of the high-risk vehicle group and the medium-risk vehicle group, the section traffic characteristics, and the section geometric characteristics are input into the section accident risk prediction model through the first fully connected layer and processed using the Relu function, they are fused with the output of the second LSTM layer. The Dense layer receives the fused result and outputs the accident risk value of the target section through the second fully connected layer.

7. A method for jointly predicting and analyzing the accident risk of a road section - vehicle group according to claim 6, characterized in that, The accident risk sequence input into the first LSTM layer also needs to be processed by Kalman filtering.

8. A method for jointly predicting and analyzing the accident risks of road sections and vehicle groups according to claim 1, characterized in that The method for joint prediction of section-vehicle group accident risk includes accident risk prediction and parallel prediction of vehicle group-section-vehicle group. Among them, the steps of the accident risk prediction of vehicle group-section-vehicle group include: Use the accident risk value of the downstream section of the section where the vehicle group is located, the section geometric characteristics of the downstream section of the section where the vehicle group is located, and the internal geometric characteristics of the vehicle group as the input of the vehicle group accident risk prediction model to predict the accident risk value of the vehicle group at the next moment. If the accident risk value of the vehicle group at the next moment is greater than the threshold, use the section accident risk prediction model to predict the accident risk of the section where the vehicle group is located to obtain the accident risk prediction value of the section where the vehicle group is located. Use the predicted accident risk prediction value of the section where the vehicle group is located, the internal geometric characteristics of the high-risk vehicle group and the medium-risk vehicle group, the section traffic characteristics, and the section geometric characteristics as the input of the vehicle group accident risk prediction model of the upstream section of the section where the vehicle group is located to predict the accident risk value of the vehicle group at the next moment of the upstream vehicle group.

9. A method for jointly predicting and analyzing the accident risk of a road section - vehicle group according to claim 8, characterized in that, The method of parallel prediction is: Connect the output layer of the section accident risk prediction model and the accident risk prediction model using a fully connected layer to obtain a parallel model. Use the internal characteristics of the vehicle group, the section geometric characteristics, the section traffic characteristics, the accident risk of the downstream section, and the accident risk sequence as the input of the parallel model, and output the accident risk value of the target vehicle group arriving at the target section.

10. A joint prediction and analysis system for road section - vehicle group accident risks, characterized in that, The system is used to implement the method described in any one of claims 1 to 9 above.

Images