A road vehicle driving risk warning method, device and storage medium based on optimal section monitoring

Abstract

The present invention provides a method, device and storage medium for early warning of road vehicle driving risk based on optimal section monitoring: divide the section according to the road characteristics of the controlled section to obtain a section set; obtain the aggregate data, non-aggregate data and vehicle driving risk characteristic data of all sections to construct a sample data set; perform feature evaluation on the candidate feature variables in the sample data set to determine the optimal section; calibrate the vehicle driving risk prediction model based on the aggregate data, non-aggregate data and vehicle driving risk characteristic data of the optimal section to obtain the optimal vehicle driving risk prediction model; deploy roadside sensing equipment at the optimal section to obtain aggregate data and non-aggregate data in real time, and use the optimal vehicle driving risk prediction model to predict driving risk in the controlled section. The present invention effectively improves highway traffic safety, provides accurate risk early warning capabilities, reduces the accident rate and ensures driving safety, and promotes the development of real-time early warning active safety management and control.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle traffic safety prediction, and in particular to a method, device and storage medium for early warning of road vehicle driving risks based on optimal section monitoring. Background Art

[0002] In order to reduce the incidence of traffic accidents, a large number of methods have been proposed at home and abroad to predict vehicle traffic safety, so that corresponding measures can be taken to reduce the accident rate and maintain a good traffic state. At the same time, these methods also provide reliable data support for relevant departments to achieve active traffic safety management. In traditional vehicle traffic safety prediction research, accident data and simulation data are mainly used as data types. However, with the development of science and technology, more and more scholars have begun to use large-scale trajectory data to establish real-time traffic safety prediction models. These trajectory data have a lot of information and can more accurately reflect vehicle behavior and traffic conditions, thereby improving the accuracy and reliability of the prediction model. At present, the vehicle accident risk prediction methods involved in the existing technology are mainly implemented in the following ways:

[0003] 1. Use accident data to establish a real-time prediction model for vehicle driving safety and analyze various influencing factors. For example, the Chinese patent application with publication number CN115238782A discloses a real-time prediction method for highway traffic safety risks based on a decision tree. This method obtains sample data by preprocessing historical traffic accident data; constructs a risk level prediction model; collects driving data in real time to obtain the safety risk level; and analyzes the degree of influence of each key risk factor on the safety risk level. This method uses historical traffic accident data. Since historical traffic accident data has the defects of strong randomness, high discreteness, long collection cycle, low timeliness, and inaccurate quality, it is not conducive to evaluating vehicle driving risks to a certain extent.

[0004] 2. The existing prediction model is established based on driver characteristics, vehicle characteristics, and environmental factors. For example, the Chinese patent with announcement number CN218825851U proposes a road traffic safety risk prediction method using vehicle status data and environmental data. This type of method uses vehicle characteristics and environmental characteristics to determine vehicle accident risks, but does not consider the impact of cross-sectional traffic flow data on vehicle driving safety.

[0005] In general, the existing vehicle driving risk warning methods have some shortcomings, such as the high discreteness of historical accident data and the deviation between simulation data and real data. These factors will affect the accuracy of traffic safety assessment. Secondly, the existing vehicle driving risk prediction model still has a certain gap in the method of using cross-sectional data to study traffic safety. Cross-sectional data is easy to obtain, the data is stable and the calculation is simple, which has the advantage of faster data processing. Therefore, there is an urgent need for a road vehicle driving risk warning method, device and storage medium based on optimal cross-sectional monitoring to solve the problems existing in the prior art. Summary of the invention

[0006] The purpose of the present invention is to provide a road vehicle driving risk warning method based on optimal section monitoring, aiming to solve the problem that the existing technology has too high requirements for the full sample and lacks the use of existing section data for accurate driving risk warning. The specific technical solution is as follows:

[0007] A road vehicle driving risk early warning method based on optimal section monitoring, comprising:

[0008] S1. Divide the sections according to the road characteristics of the controlled section to obtain a set of candidate sections;

[0009] S2, obtaining aggregate data, non-aggregate data and vehicle driving risk characteristic data of all sections to construct a sample data set;

[0010] S3, perform feature evaluation on candidate feature variables in the sample data set to determine the optimal section;

[0011] S4. Calibrate the vehicle driving risk prediction model based on the aggregate data, non-aggregate data and vehicle driving risk characteristic data of the optimal section to obtain the optimal vehicle driving risk prediction model Model * ;

[0012] S5. Deploy roadside sensing equipment at the optimal section to obtain aggregated and non-aggregated data in real time, and use the optimal vehicle driving risk prediction model Model * Carry out driving risk prediction on controlled road sections;

[0013] S6. Dynamically update the optimal vehicle driving risk prediction model Model according to the set update conditions * .

[0014] Preferably, in the above technical solution, step S2 is specifically:

[0015] Obtain the vehicle trajectory data of all samples in the controlled road section and obtain the vehicle characteristic data;

[0016] Based on the vehicle characteristic data, aggregate the traffic flow parameters within all cross-sections to obtain the aggregated data for each cross-section;

[0017] The disaggregated data includes vehicle motion characteristic parameters, or includes vehicle motion characteristic parameters and environmental data; obtain vehicle motion characteristic parameters based on vehicle characteristic data, and obtain environmental data according to meteorological sensors and actual road information;

[0018] Based on the vehicle characteristic data, calculate the vehicle driving risk characteristic data of all vehicles within the controlled section every T1 time.

[0019] In the above technical solution, preferably, the vehicle characteristic data includes: the position of the vehicle center point, vehicle type, vehicle speed, vehicle acceleration, vehicle deceleration, and the position of the center point of the leading vehicle, the speed of the leading vehicle, the position of the center point of the following vehicle, and the speed of the following vehicle obtained after matching the positions of the front and rear vehicles;

[0020] The aggregated data of the cross-section includes: traffic volume within the cross-section, average value, standard deviation, maximum value, minimum value, range, and quantile value of vehicle speed within the cross-section, average value, standard deviation, and quantile value of vehicle acceleration within the cross-section, average value, standard deviation, maximum value, minimum value, range, and quantile value of vehicle deceleration within the cross-section;

[0021] The vehicle motion characteristic parameters include: vehicle type, maximum speed of the vehicle driving within the cross-section, maximum acceleration of the vehicle driving within the cross-section, and maximum deceleration of the vehicle driving within the cross-section;

[0022] The environmental data includes: weather conditions, road slipperiness, and road gradient;

[0023] The vehicle driving risk characteristic data is represented by the extended time to collision ETTC. When 0 second < EETC < EETC*, it means the vehicle does not have enough time to avoid traffic conflicts, and at this time it is judged as a dangerous state. When ETTC* ≤ EETC or ETTC ≤ 0 second, it means the vehicle has enough time to avoid traffic conflicts and is judged as a safe state; where EETC* is the discrimination threshold for the dangerous state, and the units of both EETC* and EETC are seconds.

[0024] In the above technical solution, preferably, the extended time to collision ETTC is:

[0025]

[0026] Among them, L c represents the vehicle length of the leading vehicle c, O c represents the center point of the leading vehicle c at time T, V c represents the speed of the leading vehicle c at time T, L drepresents the length of the following vehicle d, O d represents the center point of the rear vehicle d at time T, V d represents the speed of the rear vehicle d at time T, and T represents the current time.

[0027] In the above technical solution, the candidate characteristic variables in S3 are aggregated data and non-aggregated data of all sections, and the optimal section is selected according to the sum of the scores of the candidate characteristic variables of each section based on the characteristic evaluation result, specifically:

[0028] The section whose sum of scores is greater than or equal to the threshold A is taken as the optimal section; if there is no section whose sum of scores is greater than or equal to the threshold A, the sections whose sum of scores ranks top N are selected as the optimal sections, where N is a natural number greater than or equal to 1, A∈[0.5, 0.9).

[0029] Preferably, in the above technical solution, step S4 is specifically:

[0030] S4.1. Select j methods with classification prediction functions and combine them with i model prediction time intervals Δt to construct i×j vehicle driving risk prediction models; select the aggregate data with the top M contribution rankings and the non-aggregate data with the top K contribution rankings in the optimal section as the characteristic variable X of the vehicle driving risk prediction model, and take the dangerous state of a certain vehicle at the time T+Δt in the entire controlled road section as the predicted value Y, where 0 represents a safe state and 1 represents a dangerous state; where M and K are both natural numbers greater than or equal to 1, i and j are both natural numbers greater than or equal to 2, and T represents the current time;

[0031] S4.2. Automatically adjust the parameters of i×j vehicle driving risk prediction models, traverse the parameter combinations of each vehicle driving risk prediction model, find the optimal hyperparameter configuration of each vehicle driving risk prediction model, and use accuracy and F-Score to evaluate the effect of each vehicle driving risk prediction model; where F-Score is the weighted harmonic average of precision and recall;

[0032] S4.3. Select the optimal vehicle driving risk prediction model based on the accuracy and F-Score of each vehicle driving risk prediction model * .

[0033] In the above technical solution, step S4.3 selects the optimal vehicle driving risk prediction model Model * Specifically:

[0034] Compare the accuracy rates Accuracy and F-Score of i×j vehicle driving risk prediction models, and select the quasi-optimal vehicle driving risk prediction model Model according to the accuracy rates Accuracy and F-Score. Specifically, when the accuracy rates Accuracy and F-Score of a certain vehicle driving risk prediction model are both the highest, confirm that the vehicle driving risk prediction model is the quasi-optimal vehicle driving risk prediction model Model; when there are more than two models whose accuracy rates Accuracy and F-Score values ​​are consistent and both are the highest, select the model with the smallest model prediction time interval Δt as the quasi-optimal vehicle driving risk prediction model Model; when the models corresponding to the highest values ​​of the accuracy rates Accuracy and F-Score are different, select the model with the highest accuracy as the quasi-optimal vehicle driving risk prediction model Model;

[0035] Further determine whether the accuracy of the quasi-optimal vehicle driving risk prediction model Model is greater than the threshold Q1. If the judgment condition is met, the model is the optimal vehicle driving risk prediction model Model. * If the judgment condition is not met, return to step S4.2 to readjust the optimal hyperparameter configuration of the vehicle driving risk prediction model.

[0036] Preferably in the above technical solution, the aggregation time interval of the aggregated data in S2 is also i, and the model prediction time interval is equal to the aggregation time interval.

[0037] The present invention also provides an early warning device using the above-mentioned road vehicle driving risk early warning method based on optimal section monitoring, comprising:

[0038] A data acquisition module, which is used to collect aggregate data and non-aggregate data at the optimal section;

[0039] Risk prediction module, using the optimal vehicle driving risk prediction model Model * Predict whether the vehicle is at risk of collision;

[0040] Information transmission module, used for data information transmission and interaction between modules;

[0041] Storage module, used to store data and support data query and retrieval;

[0042] The risk warning module sends a warning message when the vehicle is at risk of collision.

[0043] The present invention also provides a storage medium, in which the road vehicle driving risk early warning method based on optimal section monitoring is stored.

[0044] The application of the technical solution of the present invention has the following beneficial effects:

[0045] The technical solution of the present invention utilizes the correlation characteristics between micro-trajectory data and cross-sectional aggregate data, and uses aerial photography equipment to obtain full sample data in advance. Compared with historical accident data and simulation data, it avoids data bias to a certain extent, improves the authenticity and accuracy of the data, and provides a reliable data basis for building a vehicle driving risk prediction model.

[0046] The technical solution of the present invention adopts cross-sectional aggregate data to construct a vehicle driving risk prediction model. Compared with the driver characteristics (such as age, gender, driving style, etc.) and the vehicle's full driving characteristics (such as speed, acceleration, headway and other parameters) adopted by the traditional model, the acquisition method of cross-sectional aggregate data is simpler and the calculation is simpler, thereby making the data preprocessing link in the model construction process faster.

[0047] The technical solution of the present invention comprehensively considers the influence of section aggregate data of different section intervals, and determines the optimal section position by analyzing its contribution to the vehicle driving risk prediction model, thereby avoiding the blindness of the layout method, making the layout of roadside sensing equipment more scientific and reasonable, and effectively improving resource utilization.

[0048] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The drawings constituting a part of this application 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 an improper limitation of the present invention. In the drawings:

[0050] Figure 1 It is a flow chart of the road vehicle driving risk early warning method based on optimal section monitoring of the present invention; DETAILED DESCRIPTION

[0051] In order to facilitate the understanding of the present invention, the present invention will be described more fully below, and preferred embodiments of the present invention are provided. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0053] Example:

[0054] See also Figure 1 This embodiment provides a road vehicle driving risk warning method based on optimal section monitoring, which specifically includes the following steps:

[0055] S1. Divide the sections according to the road characteristics of the controlled section to obtain a set of candidate sections;

[0056] Specifically, in this embodiment, based on historical accident data, dangerous sections with frequent accidents are selected as controlled sections, and b=L / w is solved according to the principle of rounding up to the integer, to obtain the total number of sections divided into the controlled section; wherein L is the total length L of the controlled section (unit: meter), w is the width of each section (i.e. the length of the section interval along the length direction of the controlled section, unit: meter), and b is the total number of sections in the section set.

[0057] In this embodiment, the total length L of the controlled section is generally selected to be within 1000 meters, and the value of w is determined as follows:

[0058]

[0059] From the above, we can know that when the total length L of the controlled section is less than 100 meters, the width of each section is set to 5 meters; when the total length L of the controlled section is greater than or equal to 100 meters and less than 200 meters, the width of each section is set to 10 meters; and so on. The purpose of rounding up b = L / w is to achieve: when L / w cannot be divided evenly, the tail section as the remainder will be treated as a separate section (the width of this section is less than w).

[0060] Furthermore, after the sections are divided, each section is numbered along the driving direction of the controlled section. After the vehicle enters the controlled section, it passes through sections 1, 2, 3, ..., b-1, and b in sequence.

[0061] According to the above division rules, taking the total length of a controlled road section of 176 meters as an example, the section width w is selected as 10 meters for division, and the section is divided starting from the driving starting point of the controlled road section. The 0-10 meter position of the controlled road section is defined as section 1, 10-20 meters is defined as section 2, ..., 160-170 meters is defined as section 17, and 170-176 meters is defined as section 18, forming a candidate section set.

[0062] Although the present embodiment describes how to divide the sections according to the road characteristics of the controlled section, those skilled in the art can also adjust the section division rules according to actual conditions. For example, the width w of each section can be set to be the same, and by controlling the values ​​of the total length L and the width w, it is ensured that b=L / w is an integer.

[0063] S2. Obtain aggregate data, non-aggregate data and vehicle driving risk characteristic data of all sections to construct a sample data set; the sample data set contains a total of m samples;

[0064] Furthermore, step S2 specifically includes:

[0065] 1) Obtain the vehicle trajectory data of all samples in the controlled road section and obtain the vehicle feature data;

[0066] Use drones to take aerial photos of the controlled road sections, obtain high-definition video images, use video recognition systems to analyze the videos, perform trajectory smoothing and noise elimination, obtain full sample vehicle trajectory data within the controlled road sections, and obtain vehicle feature data; specifically, vehicle feature data includes: vehicle center point position O, vehicle type V type , the vehicle speed V, the vehicle acceleration ACC, the vehicle deceleration DEC, and the center point position O of the front vehicle c obtained by matching the positions of the front and rear vehicles c , the speed V of the front vehicle c c , the center point position O of the rear vehicle d d and the speed V of the following vehicle d d .

[0067] 2) Based on the vehicle characteristic data, the traffic flow parameters in all sections are aggregated to obtain the aggregate data of each section;

[0068] Set the aggregation time interval and model prediction time interval to aggregate traffic flow parameters and build the subsequent vehicle driving risk prediction model, where aggregation time interval = model prediction time interval = Δt. The value of Δt includes: 5 seconds, 10 seconds, 20 seconds, 30 seconds, 1 minute, 2 minutes, etc., with a total of i time intervals; when i = 1, it means that the aggregation time interval and model prediction time interval are both 5 seconds, when i = 2, it means that the aggregation time interval and model prediction time interval are both 10 seconds, when i = 3, it means that the aggregation time interval and model prediction time interval are both 20 seconds; and so on.

[0069] Those skilled in the art can understand that the value of Δt here is not limited to that listed in this embodiment, and those skilled in the art can adjust the time interval according to actual conditions. In this embodiment, i is a natural number greater than or equal to 2.

[0070] Furthermore, the aggregation method is to first calculate the average value of the parameter of all vehicles in the section every 1 second, and then aggregate it with the aggregation time interval Δt seconds as the unit. The aggregation method includes taking the average value, standard deviation, extreme value, etc. Take a as any positive integer from 1 to b, and obtain the aggregate data of the a-th section, where the aggregate data of the a-th section includes: the traffic volume in the a-th section, the average value, standard deviation, maximum value, minimum value, range and quantile value of the vehicle speed in the a-th section, the average value, standard deviation and quantile value of the vehicle acceleration in the a-th section, and the average value, standard deviation, maximum value, minimum value, range and quantile value of the vehicle deceleration in the a-th section, as follows:

[0071] Sectional traffic volume FVO_a: the number of vehicles passing through the ath section interval within Δt seconds.

[0072] Average cross-sectional velocity V mean_a : Calculate the average speed of all vehicles in the ath section every 1 second, and then calculate the average speed every 1 second within Δt seconds.

[0073] Cross-sectional velocity standard deviation V std_a : Calculate the average speed of all vehicles in the ath section every 1 second, and then calculate the standard deviation of the average speed every 1 second within Δt seconds.

[0074] Maximum cross-sectional velocity V max _a: Calculate the average speed of all vehicles in the ath section every 1 second, and then find the maximum value of the average speed every 1 second within Δt seconds.

[0075] Minimum cross-sectional velocity V min _a: Calculate the average speed of all vehicles in the ath section every 1 second, and then find the minimum value of the average speed every 1 second within Δt seconds.

[0076] Section velocity difference V range _a: Calculate the average speed of all vehicles in the ath section every 1 second, and then find the range of the average speed within Δt seconds.

[0077] Section velocity quantile value V quantile _a: Calculate the average speed of all vehicles in the ath section every 1 second, and then find the percentile value within Δt seconds for the average speed every 1 second.

[0078] Average cross-sectional acceleration ACC mean _a: Calculate the average acceleration of all vehicles in the ath section every 1 second, and then calculate the average value of the average acceleration every 1 second within Δt seconds.

[0079] Cross-sectional acceleration standard deviation ACC std_a: Calculate the average acceleration of all vehicles in the ath section every 1 second, and then calculate the standard deviation of the average acceleration every 1 second within Δt seconds.

[0080] Maximum cross-sectional acceleration ACC max _a: Calculate the average acceleration of all vehicles in the ath section every 1 second, and then find the maximum value of the average acceleration every 1 second within Δt seconds.

[0081] Minimum cross-sectional acceleration ACC min _a: Calculate the average acceleration of all vehicles in the ath section every 1 second, and then find the minimum value of the average acceleration every 1 second within Δt seconds.

[0082] Section Acceleration Range ACC range _a: Calculate the average acceleration of all vehicles in the ath section every 1 second, and then find the range of the average acceleration every 1 second within Δt seconds.

[0083] Section acceleration percentile ACC quantile _a: Calculate the average acceleration of all vehicles in the ath section every 1 second, and then calculate the percentile value of the average acceleration every 1 second within Δt seconds.

[0084] Average value of cross-sectional deceleration DEC mean _a: Calculate the average deceleration value of all vehicles in the ath section every 1 second, and then calculate the average value of the deceleration value every 1 second within Δt seconds.

[0085] Standard deviation of section deceleration DEC std _a: Calculate the average deceleration of all vehicles in the ath section every 1 second, and then calculate the standard deviation of the average deceleration every 1 second within Δt seconds.

[0086] Maximum value of cross-section deceleration DEC max _a: Calculate the average deceleration of all vehicles in the ath section every 1 second, and then find the maximum value of the average deceleration every 1 second within Δt seconds.

[0087] Minimum value of cross-section deceleration DEC min _a: Calculate the average deceleration of all vehicles in the ath section every 1 second, and then find the minimum value of the average deceleration every 1 second within Δt seconds.

[0088] Section deceleration extreme difference DEC range _a: Calculate the average deceleration value of all vehicles in the ath section every 1 second, and then find the range within Δt seconds for the average deceleration value every 1 second.

[0089] Section deceleration fraction value DEC quantile_a: Calculate the average deceleration value of all vehicles in the ath section within one second, and then find the percentile value within Δt seconds for the average deceleration value within one second.

[0090] 3) The non-aggregate data includes vehicle motion characteristic parameters, or includes vehicle motion characteristic parameters and environmental data; wherein the vehicle motion characteristic parameters are obtained based on the vehicle characteristic data, and the environmental data are obtained based on meteorological sensors and actual road information;

[0091] Specifically, the vehicle motion characteristic parameters include: vehicle type V type 、Max.V, the maximum speed of the vehicle in the ath section a , the maximum acceleration of the vehicle in the ath section Max.ACC a and the maximum deceleration Max.DEC of the vehicle in the ath section a .

[0092] The environmental data includes: weather conditions Weather, road wetness Wetness and road slope Slope.

[0093] It should be noted that environmental data is not a necessary condition for the early warning method of this embodiment. In some embodiments, non-aggregate data is allowed to exclude environmental data.

[0094] 4) Based on the vehicle characteristic data, the vehicle driving risk characteristic data of all vehicles in the controlled road section is calculated every T1 time; in this embodiment, T1 is 0.1 seconds, and those skilled in the art can adjust the value of T1 according to actual conditions;

[0095] Preferably, the vehicle driving risk characteristic data is expressed by extended distance to collision time ETTC as:

[0096]

[0097] Among them, L c represents the length of the preceding vehicle c, O c represents the center point of the front vehicle c at time T, V c represents the speed of the preceding vehicle c at time T, L d represents the length of the following vehicle d, O d represents the center point of the rear vehicle d at time T, V d represents the speed of the rear vehicle d at time T, where T represents the current time;

[0098] When 0 seconds < EETC < EETC*, it indicates that the vehicle does not have enough time to avoid a traffic conflict, and it is judged as a dangerous state at this time. When ETTC* ≤ EETC or ETTC ≤ 0 seconds, it means that the vehicle has enough time to avoid a traffic conflict and is judged as a safe state; where EETC* is the discrimination threshold for the dangerous state, and the units of both EETC* and EETC are seconds.

[0099] S3. Conduct feature evaluation on the candidate feature variables in the sample dataset to determine the optimal section;

[0100] Preferably, select the aggregated data and disaggregated data of all sections from T - Δt to T as candidate feature variables. The candidate feature variables are specifically:

[0101] Section aggregated data:

[0102]

[0103] The disaggregated data includes vehicle motion characteristic parameters, or includes vehicle motion characteristic parameters and environmental data; where the vehicle motion characteristic parameters are {V type , Max.V a , Max.ACC a , Max.DEC a}, and the environmental data are {Weather, Wetness, Slope}.

[0104] Use a tree - based model to conduct feature evaluation on the candidate feature variables, and select the optimal section according to the sum of the scores of the candidate feature variables for each section;

[0105] Specifically, take the section whose sum of scores (i.e., the sum of contribution degrees) is greater than or equal to the threshold A as the optimal section; if there is no section whose sum of scores is greater than or equal to the threshold A, then select the top N sections ranked by the sum of scores as the optimal section, where N is a natural number greater than or equal to 1, and A ∈ [0.5, 0.9). In this embodiment, N = 2 and A = 0.5, that is, select the top 2 sections ranked by the sum of scores as the optimal section.

[0106] It should be noted that in some embodiments, other methods may be used to conduct feature evaluation on the candidate feature variables, and those skilled in the art can make flexible selections.

[0107] S4. Calibrate the vehicle driving risk prediction model based on the aggregated data, disaggregated data, and vehicle driving risk characteristic data of the optimal section to obtain the optimal vehicle driving risk prediction model Model * ;

[0108] Preferably, the step S4 is specifically:

[0109] S4.1. Select j methods with classification prediction functions and combine them with i different model prediction time intervals Δt to construct i×j vehicle driving risk prediction models; select the aggregate data with the top M contribution rankings and the non-aggregate data with the top K contribution rankings in the optimal section as the characteristic variable X of the vehicle driving risk prediction model, and take the dangerous state of a certain vehicle at the time T+Δt in the entire controlled section as the predicted value Y, where 0 represents a safe state and 1 represents a dangerous state; where M and K are both natural numbers greater than or equal to 1, and i and j are both natural numbers greater than or equal to 2;

[0110] Preferably, the method with classification prediction function in step S4.1 can be machine learning or deep learning, such as random forest, support vector machine, gradient boosting tree, decision tree, K nearest neighbor, support vector machine algorithm, etc.

[0111] Preferably, the feature variable X in step S4.1 needs to be standardized before being input into the vehicle driving risk prediction model. Specifically, the StandardScaler function of the sklearn library in Python can be called to implement the standardization of the feature variable X; wherein the feature variable X is expressed as:

[0112]

[0113] Among them, m is the total number of sample data, n is the total number of feature variables, and x mn Represents the nth characteristic variable of the mth sample data.

[0114] S4.2. Automatically adjust the parameters of i×j vehicle driving risk prediction models using grid search technology, traverse the parameter combinations of each vehicle driving risk prediction model, find the optimal hyperparameter configuration of each vehicle driving risk prediction model, and use accuracy and F-Score to evaluate the effect of each vehicle driving risk prediction model;

[0115] Preferably, the accuracy and F-Score in step S4.2 are respectively expressed as:

[0116]

[0117] Preferably, as known to those skilled in the art, F-score, also known as F-measure, is a weighted harmonic average of precision and recall.

[0118] S4.3. Select the optimal vehicle driving risk prediction model based on the accuracy and F-Score of each vehicle driving risk prediction model * .

[0119] Specifically, in step S4.3, the accuracy and F-Score of the i×j vehicle driving risk prediction models are compared, and the quasi-optimal vehicle driving risk prediction model Model is first selected according to the accuracy and F-Score, specifically:

[0120] When the accuracy and F-Score of a vehicle driving risk prediction model are both the highest, the vehicle driving risk prediction model is confirmed to be the quasi-optimal vehicle driving risk prediction model Model;

[0121] When there are more than two models with the same Accuracy and F-Score values ​​and both are the highest, the model with the smallest model prediction time interval Δt is selected as the quasi-optimal vehicle driving risk prediction model;

[0122] When the accuracy and F-Score values ​​have the highest values, the corresponding models are different, and the model with the highest accuracy is used as the quasi-optimal vehicle driving risk prediction model;

[0123] After obtaining the quasi-optimal vehicle driving risk prediction model Model, it is further determined whether the accuracy of the quasi-optimal vehicle driving risk prediction model Model is greater than the threshold Q1. If the judgment condition is met, the model is the optimal vehicle driving risk prediction model Model. * If the judgment condition is not met, return to step S4.2 to readjust the optimal hyperparameter configuration of the vehicle driving risk prediction model.

[0124] Preferably, the threshold Q1 can be set according to actual conditions. In this embodiment, the threshold Q1 is set to 80%.

[0125] S5. Deploy roadside sensing equipment at the optimal section to obtain aggregated and non-aggregated data in real time, and use the optimal vehicle driving risk prediction model Model * Carry out driving risk prediction on controlled road sections;

[0126] Preferably, the roadside sensing equipment includes a laser radar, a radar-vision integrated device, a camera and a meteorological sensor, which are used to obtain the aggregated data and the non-aggregated data of the optimal section; the aggregated data and the non-aggregated data collected by the roadside sensing equipment are input into the optimal vehicle driving risk prediction model Model * In the prediction of the vehicle's driving risk at the future time T+Δt, the optimal vehicle driving risk prediction model Model * When it is determined that a vehicle is at risk of collision, an early warning message is sent to the management platform through the early warning system.

[0127] S6. Dynamically update the optimal vehicle driving risk prediction model Model according to the set update conditions * .

[0128] Preferably, in order to ensure the optimal vehicle driving risk prediction model Model * The prediction accuracy of the optimal vehicle driving risk prediction model Model * Perform rolling updates, that is, retrain and optimize the vehicle driving risk prediction model based on the historical data of the previous 7 days, and re-determine the optimal vehicle driving risk prediction model Model * At the same time, this embodiment also sets the optimal vehicle driving risk prediction model Model * When the average value of the prediction accuracy is lower than the threshold Q2, step S4.2 is executed to readjust the optimal hyperparameter configuration of the vehicle driving risk prediction model and complete the model update to ensure the reliability and practicality of the prediction results.

[0129] Preferably, the threshold Q2 can be set by those skilled in the art according to actual conditions, wherein Q2 is less than Q1. In this embodiment, Q2 is set to 70%.

[0130] This embodiment also provides a road vehicle driving risk warning device based on optimal section monitoring, the warning device adopts the above-mentioned warning method, and the device includes:

[0131] A data acquisition module, which is responsible for collecting aggregate data and non-aggregate data at the optimal section;

[0132] Risk prediction module, using the optimal vehicle driving risk prediction model Model * Predict whether the vehicle is at risk of collision;

[0133] Information transmission module, used for data information transmission and interaction between modules;

[0134] Storage module, used to store data and support data query and retrieval;

[0135] The risk warning module sends a warning message when the vehicle is at risk of collision.

[0136] Preferably, the data acquisition module includes the above-mentioned roadside perception equipment, the risk prediction module includes a cloud computing platform, a traffic management system, etc., and the risk warning module can push warning information through text messages, electronic display screens, application push, etc.

[0137] Preferably, the storage module includes a storage medium, which is used not only to store real-time collected data, preprocessed data, and historical data, but also to store the above-mentioned early warning method. When the early warning method in the storage medium is executed by the processor, the early warning method in this embodiment is implemented.

[0138] 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. A road vehicle driving risk early warning method based on optimal section monitoring, characterized in that: include: S1. Divide the sections according to the road characteristics of the controlled section to obtain a set of candidate sections; S2, obtaining aggregate data, non-aggregate data and vehicle driving risk characteristic data of all sections to construct a sample data set; S3, perform feature evaluation on candidate feature variables in the sample data set to determine the optimal section; S4. Calibrate the vehicle driving risk prediction model based on the aggregate data, non-aggregate data and vehicle driving risk characteristic data of the optimal section to obtain the optimal vehicle driving risk prediction model ; S5. Deploy roadside sensing equipment at the optimal section to obtain aggregated and non-aggregated data in real time, and use the optimal vehicle driving risk prediction model Carry out driving risk prediction on controlled road sections; S6. Dynamically update the optimal vehicle driving risk prediction model according to the set update conditions ; The step S4 specifically comprises: S4.

1. Selection j A method with classification prediction function and i Model prediction time interval t Combine and construct i × j A vehicle driving risk prediction model is constructed; the aggregate data with the top M contribution and the non-aggregate data with the top K contribution in the optimal section are selected as the characteristic variable X of the vehicle driving risk prediction model, and the vehicle driving risk prediction model is constructed based on the vehicle driving risk prediction model. The dangerous state at time t is the predicted value Y, 0 represents a safe state, and 1 represents a dangerous state; where M and K are both natural numbers greater than or equal to 1, i and j All are natural numbers greater than or equal to 2, and T represents the current time; S4.

2. Yes i × j The vehicle driving risk prediction model is automatically adjusted, the parameter combination of each vehicle driving risk prediction model is traversed, the optimal hyperparameter configuration of each vehicle driving risk prediction model is found, and the accuracy is used and To evaluate the effect of each vehicle driving risk prediction model; among them, It is the weighted harmonic average of precision and recall; S4.

3. Accuracy of the driving risk prediction model for each vehicle and Select the optimal vehicle driving risk prediction model .

2. The method for early warning of road vehicle driving risks based on optimal section monitoring according to claim 1 is characterized in that: Step S2 specifically includes: Obtain the vehicle trajectory data of all samples in the controlled road section and obtain the vehicle characteristic data; Based on the vehicle characteristic data, the traffic flow parameters in all sections are aggregated to obtain the aggregate data of each section; The non-aggregate data includes vehicle motion characteristic parameters, or includes vehicle motion characteristic parameters and environmental data; Obtain vehicle motion characteristic parameters based on vehicle characteristic data, and obtain environmental data based on meteorological sensors and actual road information; Based on the vehicle characteristic data, the vehicle driving risk characteristic data of all vehicles in the controlled road section is calculated every T1 time.

3. The method for early warning of road vehicle driving risk based on optimal section monitoring according to claim 2 is characterized in that: The vehicle characteristic data includes: the center point position of the vehicle, the vehicle type, the speed of the vehicle, the acceleration of the vehicle, the deceleration of the vehicle, and the center point position of the front vehicle, the speed of the front vehicle, the center point position of the rear vehicle and the speed of the rear vehicle obtained after matching the positions of the front and rear vehicles; The aggregate data of the section include: the traffic volume in the section, the average value, standard deviation, maximum value, minimum value, range and percentile value of the vehicle speed in the section, the average value, standard deviation and percentile value of the vehicle acceleration in the section, and the average value, standard deviation, maximum value, minimum value, range and percentile value of the vehicle deceleration in the section; The vehicle motion characteristic parameters include: vehicle type, maximum speed of the vehicle traveling in the section, maximum acceleration of the vehicle traveling in the section, and maximum deceleration of the vehicle traveling in the section; The environmental data include: weather conditions, road slipperiness and road slope; The vehicle driving risk characteristic data is extended distance collision time ETTC Indicates that when 0 seconds < EETC < EETC When the vehicle does not have enough time to avoid traffic conflicts, it is considered to be in a dangerous state. ETTC ≤ EETC or ETTC ≤0 seconds, indicating that the vehicle has enough time to avoid traffic conflicts and is considered to be in a safe state; EETC is the threshold value for judging dangerous status, EETC and EETC The unit is seconds.

4. The method for early warning of road vehicle driving risk based on optimal section monitoring according to claim 3 is characterized in that: The extended distance collision time ETTC for: ； in, Indicates the vehicle ahead c The driver, Indicates the front car at time T c The center point of Indicates the front car at time T c speed, Indicates the following vehicle d The driver, Indicates the following car at time T d The center point of Indicates the following car at time T d speed, T represents the current time.

5. The method for early warning of road vehicle driving risks based on optimal section monitoring according to claim 1 is characterized in that: The candidate characteristic variables in S3 are aggregated data and non-aggregated data of all sections. Based on the characteristic evaluation results, the optimal section is selected according to the sum of the scores of the candidate characteristic variables of each section. Specifically: The section whose sum of scores is greater than or equal to the threshold A is taken as the optimal section; if there is no section whose sum of scores is greater than or equal to the threshold A, the sections whose sum of scores ranks top N are selected as the optimal sections, where N is a natural number greater than or equal to 1, A∈[0.5, 0.9).

6. The road vehicle driving risk early warning method based on optimal section monitoring according to claim 1 is characterized in that: Step S4.3 Select the optimal vehicle driving risk prediction model Specifically: Compare i × j The accuracy of the vehicle driving risk prediction model and , according to the accuracy and First select the quasi-optimal vehicle driving risk prediction model , specifically: when the accuracy of a vehicle driving risk prediction model and When both are the highest, the vehicle driving risk prediction model is confirmed to be the quasi-optimal vehicle driving risk prediction model ; When there are more than two models, the accuracy and When the values ​​are consistent and both are the highest, the model prediction time interval is selected t The smallest model as a quasi-optimal vehicle driving risk prediction model When the accuracy and When the value is the highest, the corresponding model is different. The model with the highest accuracy is taken as the quasi-optimal vehicle driving risk prediction model. ; Further determine the quasi-optimal vehicle driving risk prediction model Is the accuracy greater than the threshold? Q 1. If the judgment conditions are met, the model is the optimal vehicle driving risk prediction model If the judgment condition is not met, return to step S4.2 to readjust the optimal hyperparameter configuration of the vehicle driving risk prediction model.

7. The road vehicle driving risk early warning method based on optimal section monitoring according to claim 1 is characterized in that: The aggregation time interval of the aggregated data in S2 is also i The model prediction time interval is equal to the aggregation time interval.

8. An early warning device using the road vehicle driving risk early warning method based on optimal section monitoring as described in any one of claims 1 to 7, characterized in that: include: A data acquisition module, which is used to collect aggregate data and non-aggregate data at the optimal section; Risk prediction module, using the optimal vehicle driving risk prediction model Predict whether the vehicle is at risk of collision; Information transmission module, used for data information transmission and interaction between modules; Storage module, used to store data and support data query and retrieval; The risk warning module sends a warning message when the vehicle is at risk of collision.

9. A storage medium, characterized in that: The storage medium stores the road vehicle driving risk warning method based on optimal section monitoring as described in any one of claims 1 to 7.

Images