Driver driving-in willingness evaluation and differentiated guidance method and system based on vehicle-road cooperation during temporary opening of emergency lane

Abstract

The invention relates to the technical field of intelligent traffic, and discloses a vehicle-road cooperation-based driver driving-in intention evaluation and differential guidance method and system during temporary opening of an emergency lane, and the method comprises the steps: collecting the original data of a pre-monitored road section through a vehicle-road cooperation system, calculating the standardized indexes of a single vehicle and traffic flow dimension, and carrying out the calculation of the vehicle-road cooperation system; and dynamically updating the index weight through machine learning based on historical and newly added actual driving samples, setting an admission ratio according to the number of lanes, screening high-intention vehicles, issuing a driving guide suggestion, and issuing a departure guide suggestion to low-intention vehicles. Accurate access control and dynamic guidance after the emergency lane is opened are realized, the road passing efficiency and the driving safety are improved, and the method is suitable for various emergency lane opening scenes of highways.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a method and system for assessing and providing differentiated guidance to drivers who wish to enter emergency lanes based on vehicle-road cooperation when emergency lanes are temporarily opened. Background Technology

[0002] As a crucial component of modern transportation networks, highways directly impact regional economic development and the public's travel experience through their traffic efficiency. In scenarios such as traffic congestion, accident handling, or special events (e.g., high traffic volumes during holidays), traffic management departments temporarily open emergency lanes for other vehicles to alleviate pressure on main roads and improve overall capacity. While emergency lanes are intended for emergency rescue and breakdown response, their temporary opening, although improving road resource utilization in the short term, also brings numerous problems such as mixed traffic, traffic conflicts, and safety hazards.

[0003] Currently, both domestically and internationally, emergency lane opening information is primarily disseminated to drivers through variable message signs, broadcast announcements, and navigation software notifications to guide orderly vehicle use. For example, on some highway sections in Germany, when traffic flow exceeds design capacity, the hard shoulder (emergency lane) is opened as a temporary driving lane through a dynamic traffic signal system, combined with speed limit reminders and lane guidance for control. Some provinces and cities in my country are also piloting the opening of emergency lanes during major holidays, using roadside electronic displays and traffic broadcasts to guide vehicle diversion.

[0004] However, most existing applications remain at the "open-notification" level, lacking quantitative assessment and differentiated guidance of individual vehicle intentions for using emergency lanes. This leads to low efficiency in emergency lane use, chaotic traffic flow, and even secondary accidents. Therefore, how to achieve precise control and safe guidance while opening emergency lanes has become a pressing practical problem to be solved in the field of intelligent transportation.

[0005] Although the opening of emergency lanes has been initially implemented, the following prominent technical problems still exist in actual implementation:

[0006] The first technical problem is the lack of a standardized evaluation index system. Existing technologies mostly rely on human experience or simple rules (such as vehicle type and speed) to determine whether a vehicle is allowed to enter, without establishing scientific and quantifiable evaluation indicators. The behavior of vehicles entering the emergency lane is affected by many factors, such as vehicle condition, traffic flow conditions, and driver habits. Existing systems cannot fully collect and calculate these indicators, resulting in a lack of basis for access decisions.

[0007] The second technical problem is sample selection and weight fixation. Existing methods often use data from all vehicles for model training, without focusing on effective samples of "actually entering the emergency lane". This results in the model learning a large number of noisy features, and the weight system is difficult to accurately reflect the key factors of real driving behavior. In addition, once the weights are trained, they are fixed and cannot be dynamically updated with traffic scenarios, resulting in poor adaptability.

[0008] The third technical problem is the lack of dynamic proportional constraints in access control. After the emergency lane is opened, if there is no reasonable vehicle access ratio control, it is very easy to cause congestion or even blockage in the emergency lane, which will reduce the overall traffic efficiency. The existing technology does not dynamically set the access ratio according to the number of lanes, traffic flow status, etc., resulting in unreasonable resource allocation.

[0009] The fourth technical problem is the lack of differentiation and real-time nature of guidance strategies. Existing guidance is mostly a one-size-fits-all broadcast prompt, which does not provide personalized guidance suggestions for different vehicles (such as entry timing, speed control, safe distance, etc.) and cannot respond to changes in traffic conditions in real time, resulting in limited guidance effectiveness.

[0010] To address the aforementioned technical issues, those skilled in the art urgently need a vehicle-road cooperative system that can integrate indicator calculation, dynamic weight learning, precise access screening, and real-time differentiated guidance. Summary of the Invention

[0011] The purpose of this invention is to address the aforementioned problems by designing a method and system for assessing and providing differentiated guidance to drivers who wish to enter emergency lanes based on vehicle-road cooperation when emergency lanes are temporarily open. It is important to note that the driver's willingness to choose an emergency lane refers to a comprehensive evaluation value of their behavioral tendency to enter the emergency lane, calculated quantitatively based on vehicle operation indicators and traffic flow indicators. This is not a direct representation of the driver's subjective psychological activity, but rather an objective reflection of the actual behavioral trend of vehicles entering the emergency lane. In the field of traffic engineering / vehicle-road cooperation, all subjective driving decision-making tendencies ultimately manifest as collectable objective vehicle / driving behavior data. This invention constructs a standardized indicator calculation system for both single-vehicle and traffic flow dimensions, focuses on "actually entering vehicle samples" for machine learning and weight self-learning, sets dynamic access ratios based on the number of lanes, and achieves precise instruction issuance based on vehicle-road cooperation. This improves the efficiency of emergency lane use while ensuring traffic safety. This technical solution has strong practicality, adaptability, and scalability, aligns with the development trend of intelligent transportation systems, and is of great significance for promoting the intelligent and refined management of highways.

[0012] The technical solution of the present invention to achieve the above objectives is a method for assessing and differentiating drivers' willingness to enter emergency lanes based on vehicle-road cooperation when emergency lanes are temporarily opened, comprising the following steps:

[0013] S1: Collect raw vehicle data and raw traffic flow data of the pre-monitored road section ahead of the open section of the emergency lane through the vehicle-road cooperative system;

[0014] S2: Through the indicator calculation module, the raw data is calculated based on the preset formula to extract single-vehicle dimension indicators and traffic flow dimension indicators;

[0015] The single-vehicle dimension indicators include: distance to destination ratio, difference between average vehicle speed and average traffic flow speed, lane change frequency, lane deviation rate, mean speed difference with the vehicle in front, and average acceleration.

[0016] The traffic flow dimension indicators include: average density, average headway, estimated delay time, average speed on the main road, speed difference between driving lane and emergency lane, proportion of large vehicles, and congestion index.

[0017] S3: Based on the original data of the pre-monitored road sections collected in history, the calculated index data and the corresponding actual driver behavior samples of entering the emergency lane, the initial index weights are obtained through machine learning algorithm training.

[0018] S4: After the emergency lane is opened, the original data of the pre-monitored road section of the newly added vehicles that actually enter the emergency lane is continuously collected. The corresponding indicator data is calculated by the indicator calculation module, associated with the entry behavior label, triggering the weight self-learning mechanism, and updating the weight of the indicator system.

[0019] S5: Calculate the emergency lane selection intention value of each vehicle in the current pre-monitored road segment based on the updated indicator weights, and sort them from high to low according to the intention value;

[0020] S6: Determine the proportion of vehicles permitted to enter the emergency lane based on the number of lanes in the open road section: in a two-lane scenario, the permitted vehicles shall not exceed 1 / 3 of the current total traffic flow; in a three-lane scenario, the permitted vehicles shall not exceed 1 / 4 of the current total traffic flow.

[0021] S7: Select vehicles whose willingness ranking is within the aforementioned admission ratio and generate entry guidance suggestions; for vehicles whose willingness ranking exceeds the admission ratio, generate stay guidance suggestions to stay away from the emergency lane.

[0022] S8: The guidance suggestions are sent to the vehicle terminal of the corresponding vehicle through the vehicle-road cooperative communication module to complete the dynamic guidance.

[0023] The calculation formula of the indicator calculation module is as follows:

[0024] (1) The calculation formula for single-vehicle dimension indicators includes:

[0025] Distance to destination ratio R d for:

[0026]

[0027] Among them, D r D represents the remaining distance between the vehicle and its destination. t This represents the total distance traveled by the vehicle.

[0028] The difference Δv between the average vehicle speed and the average traffic flow speed is:

[0029] Δv=v veh -v flow

[0030] Among them, v veh v represents the average speed of the target vehicle on the pre-monitored road segment. flow To monitor the average speed of all vehicles within the pre-monitored road section;

[0031] Lane changing frequency f l for:

[0032]

[0033] Where, N l T represents the number of lane changes a vehicle makes in a pre-monitored road segment, and T represents the time a vehicle takes to pass through the pre-monitored road segment.

[0034] Lane Departure Rate R ld for:

[0035]

[0036] Where, d i d0 is the real-time lane centerline offset distance of the vehicle, d0 is the lane half width, and n is the number of samplings during the pre-monitoring period;

[0037] Mean speed difference with the vehicle in front for:

[0038]

[0039] Among them, v veh,i v represents the vehicle's real-time speed. f,i The real-time speed of the vehicle in front;

[0040] average acceleration for:

[0041]

[0042] Among them, v i Let v be the sampling rate for the i-th time. i-1 The sampling rate is denoted as Δt, where Δt is the sampling time interval.

[0043] (2) The calculation formulas for traffic flow dimension indicators include:

[0044] The average density ρ is:

[0045]

[0046] Where N is the total number of vehicles in the pre-monitored road segment, and L is the length of the pre-monitored road segment;

[0047] Average headway for:

[0048]

[0049] Among them, T total This represents the total time taken for all vehicles to pass through the pre-monitored road segment;

[0050] Estimated delay time T d for:

[0051]

[0052] Among them, v free v is the free-flow velocity of the road segment. flow This represents the actual average speed of the road segment.

[0053] Average speed v on the main road main for:

[0054]

[0055] Among them, v i To monitor the average speed of each vehicle within the pre-monitored road section;

[0056] Speed ​​difference Δv between driving lane and emergency lane lane for:

[0057] Δv lane =v eng -v main

[0058] Among them, v eng The average speed v during the pre-monitoring period of the emergency lane. main Average speed on the main road;

[0059] Large vehicle ratio R large for:

[0060]

[0061] Where, N large To predict the number of large vehicles in the road section, N total This represents the total number of vehicles within the road segment.

[0062] The congestion index C is:

[0063]

[0064] The machine learning algorithm is one or a combination of gradient boosting tree, random forest or neural network, and the initial weights are obtained by training the mapping relationship between the calculated index features in the historical dataset and the "actually driving into the emergency lane" behavior label.

[0065] It should be noted that this invention uses the random forest algorithm to complete the initial weight training and self-learning update, focusing on "vehicle samples that actually enter the emergency lane". The initial weight training process (based on historical entry samples) is as follows:

[0066] Step 1: Sample set construction;

[0067] Historical sample data source: Raw data of pre-monitored road sections under the scenario of emergency lane opening on highways in the past 3 years. After obtaining 13 indicators through the indicator calculation module, vehicles that actually entered the emergency lane were selected as valid samples, totaling 10,000 groups.

[0068] Sample characteristics: 13 calculated indicator data (6 items in the single vehicle dimension + 7 items in the traffic flow dimension);

[0069] Sample label: uniformly marked as "1" (representing actual entry into the emergency lane);

[0070] Data preprocessing: outlier removal (using the 3σ criterion), missing value imputation (using KNN interpolation, k=5), and Min-Max normalization of all features;

[0071] Step 2: Model training;

[0072] Divide the training set and the test set: randomly divide them in a 7:3 ratio, with 7000 sets for the training set and 3000 sets for the test set;

[0073] Random forest parameter settings: 100 decision trees, maximum depth of each decision tree is 8, feature sampling number of each tree is 5 (randomly selected from 13 indicators), and node splitting criterion is Gini coefficient;

[0074] Model training: Input the training set into the random forest model and iteratively optimize it to achieve a feature recognition accuracy of ≥90% for "driving behavior";

[0075] Initial weight extraction: Based on the importance scores (Gini coefficient reduction) of each feature in the model, the initial weights of 13 indicators are obtained after normalization, and the total weights are 1.

[0076] The triggering conditions for the weight self-learning mechanism are: the number of new samples actually entering the emergency lane reaches a preset threshold (≥500 groups) or the emergency lane is open for a preset period (≥2 hours). The self-learning process updates the indicator weights through incremental training and retains effective historical feature weight information.

[0077] It should be noted that the update process of the weight self-learning mechanism (based on newly added samples) is as follows:

[0078] Step 1: Determine the triggering condition;

[0079] Real-time monitoring of the number of newly entered samples: When the number of newly valid samples actually entering the emergency lane reaches 500, or when the emergency lane is open for 2 hours, self-learning is triggered;

[0080] Requirements for new samples: Collect the original data of the pre-monitored road sections of the new vehicles, obtain 13 indicators through the indicator calculation module, and label them with the "entering" tag (1). The data preprocessing method is the same as that of the historical samples.

[0081] Step 2: Incremental training;

[0082] Sample fusion: The newly entered samples are fused with the historical samples according to their weights. The weight of the historical samples is calculated with a decay coefficient of 0.95 / cycle (e.g., the weight of the historical samples in the first self-learning is 1×0.95, the weight in the second time is 0.95×0.95, and so on). The weight of the newly entered samples is fixed at 1.

[0083] Model update: Based on the fused inbound sample set, the incremental random forest algorithm is used to update the model parameters, retain the weight information of effective features in the historical model, and only adjust the weights of features that are significantly different in the new samples (feature importance changes ≥10%).

[0084] Weight output: The updated weights are normalized to ensure that the sum is 1, replacing the original weight system, and are used for subsequent calculation of vehicle willingness values.

[0085] Step 3: Model Validation;

[0086] After each weight update, the latest 100 groups of entering samples are used for verification. If the model's feature recognition accuracy for entering behavior is ≥88%, the weight update takes effect; if it is lower than 88%, the samples are re-fused (increasing the weight ratio of historical samples) and retrained until the accuracy requirement is met.

[0087] The intention value is calculated as follows: the weighted sum W of the normalized values ​​of each indicator and their corresponding weights, and its mathematical expression is:

[0088]

[0089] Where, x i w is the normalized value of the i-th indicator. i Let n be the weight of the i-th indicator, and n be the total number of indicators (13).

[0090] It should be noted that this invention uses linear normalization (Min-Max normalization) to process the 13 calculated indicators, converting the original indicator values ​​into normalized values ​​in the [0,1] interval, as shown in the following formula:

[0091] Positive indicators (the higher the indicator value, the higher the probability of actually entering the area):

[0092]

[0093] Negative indicators (the smaller the indicator value, the higher the actual probability of entering):

[0094]

[0095] Where: x i The original value after the indicator calculation; x max The maximum value of this indicator in the historical inbound sample; x min The minimum value of this indicator in the historical inbound sample; x i ′ represents the normalized value. Indicator attribute definition (positive / negative) is as follows: Figure 3 As shown;

[0096] A specific calculation example (based on the statistical extreme values ​​of historical driving samples) is as follows:

[0097] Assuming that the extreme values ​​of some calculated indicators in the historical driving sample are as follows: Figure 4 As shown:

[0098] Example 1: Positive Indicator (Proportion of Distance to Destination)

[0099] The calculated value x of this indicator for a certain vehicle i =0.6, calculated as follows:

[0100]

[0101] Example 2: Positive indicator (the difference between the average vehicle speed and the average traffic flow speed)

[0102] The calculated value x of this indicator for a certain vehicle i =5km / h, calculated as follows:

[0103]

[0104] Example 3: Negative Indicator (Average Density)

[0105] The calculated value x of this indicator for a certain road section i =15 vehicles / km, calculated as follows:

[0106]

[0107] Example 4: Positive indicator (speed difference between driving lane and emergency lane)

[0108] The calculated value x of this indicator for a certain road section i =6km / h, calculated as follows:

[0109]

[0110] After normalizing all 13 indicators in the above manner, they are substituted into the intention value calculation formula:

[0111]

[0112] This will give you the emergency lane selection preference value for each vehicle (the higher the preference value, the higher the probability of actually entering the lane).

[0113] The entry guidance suggestions include suggestions on when to enter the emergency lane, the entry section, the driving speed range, and the safe distance from adjacent vehicles; the maintenance guidance suggestions include suggestions on the recommended driving lane, the minimum safe distance from the emergency lane (≥5m), and the vehicle speed stability control.

[0114] A driver intention assessment and differentiated guidance system based on vehicle-road cooperation when emergency lanes are temporarily opened includes the following modules:

[0115] Data acquisition module: used to collect raw vehicle data (location, speed, vehicle type, driving trajectory, etc.) and raw traffic flow data (segment length, free flow speed, etc.) of the pre-monitored road segment through vehicle-mounted sensors, roadside units (RSUs) and traffic monitoring platforms;

[0116] Indicator Calculation Module: Used to call preset calculation formulas to process raw data and output vehicle-dimensional indicators and traffic flow-dimensional indicators;

[0117] Sample filtering module: Used to filter out the actual vehicle samples that entered the emergency lane from historical vehicle data and newly added vehicle data after the lane was opened, extract the corresponding calculated index data and label them with "entered".

[0118] Weight learning module: used to train initial indicator weights based on historically selected inbound samples, and to update the weights through self-learning using newly added inbound samples after the system is opened.

[0119] Willingness assessment module: used to calculate and sort the emergency lane selection willingness values ​​of each vehicle based on the updated weights;

[0120] Access screening module: used to determine the access ratio based on the number of lanes in the open road section and screen vehicles that meet the conditions;

[0121] Guidance strategy generation module: used to generate entry guidance suggestions for authorized vehicles and departure guidance suggestions for non-authorized vehicles;

[0122] Communication delivery module: Used to deliver guidance suggestions to the vehicle terminal via vehicle-to-everything (V2X) communication technology.

[0123] The data acquisition module has a collection frequency of 10Hz, and the time window for the raw data is the pre-monitored road section data within 5 minutes before the emergency lane is opened. The data accuracy meets the following requirements: speed error ≤ ±0.5km / h, distance error ≤ ±1m, and location positioning error ≤ ±1m.

[0124] The weight learning module has a built-in incremental training algorithm. During the self-learning process, newly added samples and historical samples are weighted and fused. The weight of historical samples decays over time by a coefficient of 0.95 per cycle.

[0125] The calculation module has a computation delay of ≤100ms and a calculation result error of ≤±2%. It supports dynamic configuration of formula parameters (such as free flow speed, lane half width, etc., which can be adjusted according to road segment attributes).

[0126] Compared with the prior art, the present invention has the following non-obvious technical features:

[0127] First, this invention constructs a two-layer indicator calculation system that includes single-vehicle dimensions and traffic flow dimensions, and provides specific calculation formulas for each indicator, realizing a standardized process from raw data to evaluation indicators.

[0128] Secondly, this invention uses only "vehicles that actually enter the emergency lane" as training samples, eliminating the interference of vehicles that do not enter, thus improving the model's learning accuracy of entry behavior features.

[0129] Third, this invention designs a dynamic weight self-learning mechanism based on the number of samples and the open duration, and achieves continuous optimization of the weight system through incremental training;

[0130] Fourth, this invention dynamically sets the permitted vehicle ratio based on the number of lanes in an open road section (two lanes ≤ 1 / 3, three lanes ≤ 1 / 4), thereby achieving a reasonable allocation of resources;

[0131] Fifth, it provides a differentiated guidance suggestion system, including specific parameters such as when to enter, road section, speed range, and safe distance;

[0132] Finally, this invention realizes a closed-loop system covering the entire process from data collection, indicator calculation, sample screening, weight learning to guidance and distribution, and has strong system integration and real-time performance.

[0133] Compared with the prior art, this application has the following beneficial effects:

[0134] 1. This application achieves the standardization and calculability of evaluation indicators, solving the problem that indicators are difficult to collect directly in traditional methods;

[0135] 2. This application improves the accuracy and scenario adaptability of driving behavior prediction by focusing on actual driving samples and dynamic weight learning;

[0136] 3. This application, based on the access ratio control of the number of lanes, effectively reduces the risk of traffic conflicts while improving traffic efficiency;

[0137] 4. This application provides personalized and parameterized guidance suggestions, which enhances driver compliance and driving safety;

[0138] 5. The system disclosed in this application has the advantages of fast overall response, high accuracy and strong scalability, and is applicable to various highway emergency lane opening scenarios. Attached Figure Description

[0139] Figure 1 This is a flowchart of a method for assessing and differentiating drivers' willingness to enter emergency lanes based on vehicle-road cooperation when emergency lanes are temporarily opened, as described in this invention.

[0140] Figure 2 This is a structural block diagram of a driver's willingness to enter and differentiated guidance system based on vehicle-road cooperation when an emergency lane is temporarily opened, as described in this invention.

[0141] Figure 3 This is the indicator attribute definition (positive / negative) table described in this invention;

[0142] Figure 4 This is an extreme value table of some calculated indicators from the historical driving samples described in this invention. Detailed Implementation

[0143] The present invention will now be described in detail with reference to the accompanying drawings;

[0144] Example 1:

[0145] The application process of a driver intention assessment and differentiated guidance method based on vehicle-road cooperation in the scenario of temporary emergency lane opening on a two-lane urban highway is as follows: Figure 1 As shown:

[0146] 1. Basic parameter configuration: Pre-monitored road segment length L = 2km, free flow velocity v free =100km / h, lane half width d0 = 3.75m, sampling time interval Δt = 0.1s;

[0147] 2. Data collection: Raw data (location, speed, trajectory, etc.) of 100 vehicles within 5 minutes are collected through roadside RSUs and on-board terminals at a collection frequency of 10Hz;

[0148] 3. Indicator Calculation: The indicator calculation module calls preset formulas to calculate 6 individual vehicle indicators for each vehicle and 7 traffic flow indicators for each road segment. For example:

[0149] The remaining distance D of a certain vehicle r =8km, total distance D t =20km, then the distance to the destination R d =8 / 20 = 0.4;

[0150] The total number of vehicles in the road section is N = 100, and the average density is ρ = 100 / 2 = 50 vehicles / km;

[0151] The actual average speed v of the road section flow =60km / h, congestion index C = (100-60) / 100×100 = 40;

[0152] 4. Sample screening: Subsequently, V2X positioning data was used to determine that 35 vehicles actually entered the emergency lane, and their indicator data were extracted as valid samples.

[0153] 5. Weight Calculation: The initial weights are trained based on 10,000 historical inbound samples. After the system is opened, 600 new inbound samples are added to trigger self-learning and update the weight system.

[0154] 6. Willingness assessment and screening: Calculate and rank the willingness scores of 100 vehicles waiting to pass, with a two-lane access ratio of 1 / 3, and select the top 33 vehicles with the highest willingness scores as the access targets;

[0155] 7. Guidance and Execution: Issue entry instructions to 33 authorized vehicles, recommending they enter the emergency lane 300m from the end of the pre-monitored road section (at this time Δv lane =5km / h), control the speed at 80-90km / h; issue a keeping instruction to the remaining 67 vehicles, recommending that they drive in the inner lane and maintain a distance of 6m from the emergency lane.

[0156] Example 2:

[0157] The application process of a driver intention assessment and differentiated guidance method based on vehicle-road cooperation in the scenario of temporary emergency lane opening on a three-lane intercity highway is as follows: Figure 1 As shown:

[0158] 1. Basic parameter configuration: Pre-monitored road segment length L = 1.5km, free flow velocity v free =120km / h, lane half width d0=3.75m, sampling time interval Δt=0.1s;

[0159] 2. Data Collection and Indicator Calculation: Raw data from 120 vehicles within 5 minutes is collected, and 13 indicators are obtained through the indicator calculation module. For example, if a vehicle's passage time T = 90s, lane change count N1 = 2 times, and lane change frequency f... t =2 / 90≈0.022;

[0160] 3. Sample selection and weight update: The initial weights are trained based on 8,000 historical incoming samples. After opening, self-learning is triggered every 2 hours to update the weights by incorporating newly added incoming samples.

[0161] 4. Ranking and screening of preferences: The admission ratio for the three lanes is 1 / 4, and the top 30 vehicles with the highest preference scores are selected as the admission targets;

[0162] 5. Guidance and Implementation: Approved vehicles should enter the emergency lane 500m from the end of the pre-monitored road section, and control their speed at 90-100km / h; non-approved vehicles should stay in the middle and inner lanes and maintain a distance of more than 5m from the emergency lane.

[0163] Everything else is the same as in Example 1.

[0164] Example 3:

[0165] A driver intention assessment and differentiated guidance system based on vehicle-road cooperation when emergency lanes are temporarily opened, such as... Figure 2 As shown, it includes the following modules:

[0166] Data acquisition module: Used to collect raw vehicle data (location, speed, vehicle type, driving trajectory, etc.) and raw traffic flow data (segment length, free flow speed, etc.) of the pre-monitored road segment through vehicle-mounted sensors, roadside units (RSUs), and traffic monitoring platforms. The vehicle-mounted sensors include a GPS positioning module (accuracy ≤1m), an inertial navigation sensor, and a vehicle speed sensor; the roadside sensors include millimeter-wave radar (detection distance ≥500m) and a high-definition camera (frame rate 25fps).

[0167] The indicator calculation module is used to call preset calculation formulas to process the raw data and output vehicle-dimensional indicators and traffic flow-dimensional indicators. It is deployed in the CPU core layer of the edge computing node, uses an Intel Core i7 processor, supports parallel computing, and ensures low latency.

[0168] Sample screening module: Used to screen out the actual vehicle samples that entered the emergency lane from historical vehicle data and newly added vehicle data after opening, extract the corresponding calculated index data and label them with "entered". The screening accuracy is ≥99.8% through the fusion of roadside video monitoring and V2X positioning data.

[0169] Weight learning module: used to train initial index weights based on historically selected inbound samples, and to update the weights through self-learning using newly added inbound samples after opening. It adopts a CPU (Intel Core i7) + GPU (NVIDIA Tesla T4) architecture with a computation latency of ≤500ms.

[0170] Willingness assessment module: used to calculate and sort the emergency lane selection willingness values ​​of each vehicle based on the updated weights;

[0171] Access screening module: used to determine the access ratio based on the number of lanes in the open road section and screen vehicles that meet the conditions;

[0172] Guidance strategy generation module: used to generate entry guidance suggestions for authorized vehicles and departure guidance suggestions for non-authorized vehicles;

[0173] Communication delivery module: Used to deliver guidance suggestions to the vehicle terminal through vehicle-to-everything (V2X) communication technology. It adopts 5G-V2X communication technology, with a communication latency of ≤100ms and a command delivery success rate of ≥99.5%.

[0174] The above technical solutions only embody the preferred technical solutions of the present invention. Any modifications that may be made by those skilled in the art to certain parts therein embody the principles of the present invention and fall within the protection scope of the present invention.

Claims

1. A method for assessing and providing differentiated guidance to drivers willing to enter emergency lanes based on vehicle-road cooperation when emergency lanes are temporarily opened, characterized in that... The method includes the following steps: S1: Collect raw vehicle data and raw traffic flow data of the pre-monitored road section ahead of the open section of the emergency lane through the vehicle-road cooperative system; S2: Calculate the raw data based on the preset formula and extract single-vehicle dimension indicators and traffic flow dimension indicators; S3: Initial index weights are obtained by training a machine learning algorithm based on historical samples of vehicles that actually entered the emergency lane. S4: Trigger the weight self-learning mechanism based on the newly added actual driving samples after the opening to update the indicator weights; S5: Calculate and sort the emergency lane selection intention values ​​of each vehicle based on the updated weights; S6: Determine the proportion of vehicles allowed to enter based on the number of lanes; S7: Generate driving guidance suggestions for vehicles whose willingness scores rank within the admission ratio, and generate driving guidance suggestions for vehicles with lower rankings. S8: Send guidance suggestions to the vehicle terminal through the vehicle-road cooperative communication module.

2. The method according to claim 1, characterized in that, The single-vehicle dimension indicators include: distance to destination ratio, difference between average vehicle speed and average traffic flow speed, lane change frequency, lane deviation rate, mean speed difference with the vehicle in front, and average acceleration. The traffic flow dimension indicators include: average density, average headway, estimated delay time, average speed on the main road, speed difference between the driving lane and the emergency lane, proportion of large vehicles, and congestion index.

3. The method according to claim 1, characterized in that, The machine learning algorithm is one or a combination of gradient boosting tree, random forest, or neural network.

4. The method according to claim 1, characterized in that, The triggering conditions for the weight self-learning mechanism are: the number of newly entered samples ≥ 500 or the open time ≥ 2 hours.

5. The method according to claim 1, characterized in that, The intention value is the weighted sum of the normalized values ​​of each indicator and their corresponding weights.

6. The method according to claim 1, characterized in that, The admission ratio is as follows: no more than 1 / 3 of the total traffic flow in two-lane scenarios and no more than 1 / 4 in three-lane scenarios.

7. The method according to claim 1, characterized in that, The driving guidance suggestions include: driving timing, driving section, speed range and safe distance suggestions; The guidance recommendations include: recommended lanes, a minimum safe distance of ≥5m from the emergency lane, and speed control recommendations.

8. A driver intention assessment and differentiated guidance system based on vehicle-road cooperation when an emergency lane is temporarily opened, wherein the system can implement the method of any one of claims 1-7, characterized in that, include: The data acquisition module is used to collect raw vehicle data and raw traffic flow data for the pre-monitored road sections; The indicator calculation module is used to call preset calculation formulas to process the raw data and output vehicle-dimensional indicators and traffic flow-dimensional indicators. The sample screening module is used to filter out the actual vehicle samples that entered the emergency lane from historical vehicle data and newly added vehicle data after the lane was opened, extract the corresponding calculated index data and label them with "entered". The weight learning module is used to train the initial index weights based on the historically selected inbound samples, and to update the weights through self-learning using newly added inbound samples after the system is opened. The willingness assessment module is used to calculate and sort the emergency lane selection willingness values ​​of each vehicle based on the updated weights. The access screening module is used to determine the access ratio based on the number of lanes in the open road section and to screen vehicles that meet the conditions. The guidance strategy generation module is used to generate entry guidance suggestions for authorized vehicles and departure guidance suggestions for non-authorized vehicles. The communication distribution module is used to distribute guidance suggestions to the vehicle terminal through vehicle-road cooperative communication technology.

9. The system according to claim 8, characterized in that, The data acquisition module has a acquisition frequency of 10Hz, a time window of 5 minutes before opening, a speed error of ≤±0.5km / h, and a distance error of ≤±1m.

10. The system according to claim 8, characterized in that, The weight learning module has a built-in incremental training algorithm, and the historical sample weight decay coefficient over time is 0.95 / cycle.

Images