Safety assessment method based on fault-tolerant ramp online data

Abstract

The invention relates to the field of highway ramp data processing, and particularly discloses a safety assessment method based on fault-tolerant ramp online data, which comprises the following steps: collecting real-time monitoring data on a fault-tolerant ramp; performing state transition probability analysis on the real-time monitoring data by adopting a Markov chain model, generating an attenuation curve of the fault-tolerant capability changing along with time, and establishing a dynamic risk assessment matrix based on a nonlinear mapping relation between a ramp service level index and a safety coefficient; and according to the dynamic risk assessment matrix, determining safety assessment values under different traffic flows.

Description

Technical Field

[0001] The present invention relates to the technical field of highway ramp data, and specifically relates to a safety assessment method based on fault-tolerant ramp online data. Background Art

[0002] As a key infrastructure of the intelligent transportation system, the safety assessment of highway fault-tolerant ramps directly affects road traffic efficiency and accident prevention and control capabilities. Traditional assessment methods mainly rely on static traffic parameters and periodic manual inspections, and there are three major technical bottlenecks:

[0003] (1) At the data collection level, a single sensor system (such as video surveillance or coil detection) is difficult to achieve real-time fusion of multi-physical field data, resulting in an environmental perception blind area rate as high as 32%;

[0004] (2) At the risk assessment model level, existing methods mostly adopt a linear regression model with fixed weights and cannot effectively characterize the non-linear effects of dynamic factors such as rain and fog weather and sudden changes in traffic flow on fault tolerance;

[0005] (3) At the emergency response level, the hierarchical early warning mechanism based on preset thresholds has a decision-making lag of 17 - 23 seconds and is difficult to meet the millisecond-level safety protection requirements in the vehicle-road collaborative environment.

[0006] International research shows that the data heterogeneity problem of existing ramp monitoring systems leads to a multi-source information fusion error rate exceeding 19%. Especially under low visibility conditions, the probability of collaborative failure of lidar and vision sensors increases to 41%. The 2024 annual report of the Federal Highway Administration (FHWA) of the United States pointed out that the false alarm rate of ramp accidents using traditional risk assessment models is as high as 28%, and the prediction accuracy for sudden traffic flows such as phantom traffic jams is less than 65%. Summary of the Invention

[0007] The present application proposes a safety assessment method based on fault-tolerant ramp online data. When a vehicle is on the fault-tolerant lane of a ramp, by fusing monitoring data such as traffic flow and road conditions with a Markov chain, the dynamic evolution of the safety status assessment of the fault-tolerant ramp is realized, and through observed variables, cross-domain multi-factor fusion is achieved to dynamically evaluate the safety situation of the fault-tolerant ramp.

[0008] In the first aspect, the present application provides a safety assessment method based on fault-tolerant ramp online data, including: A safety assessment method based on fault-tolerant ramp online data, including:

[0009] Collect real-time monitoring data on the fault-tolerant ramp;

[0010] The Markov chain model is used to analyze the state transition probability of real-time monitoring data, generate a decay curve of the fault tolerance ability over time, and establish a dynamic risk assessment matrix based on the non-linear mapping relationship between the ramp service level index and the safety factor;

[0011] According to the dynamic risk assessment matrix, determine the safety assessment values under different traffic flows.

[0012] In the embodiment, the present application uploads real-time monitoring data such as traffic flow data and vehicle trajectories on the fault-tolerant ramp to the Markov chain model, realizes continuous assessment of the fault tolerance ability of the fault-tolerant ramp, thereby constructing a risk assessment matrix, and evaluates the safety situation on the fault-tolerant ramp under different traffic flows, which can achieve risk assessment in complex scenarios, as well as the problem that the safety assessment of traditional ramps is only a static assessment by manual experts and the assessment results are inaccurate.

[0013] Combined with the first aspect, the collection of real-time monitoring data on the fault-tolerant ramp includes:

[0014] Pre-configure multi-source monitoring devices on the fault-tolerant ramp; among them, the multi-source monitoring devices include: lidar devices, road surface sensing devices, video monitoring devices, and meteorological sensing devices;

[0015] According to the multi-source monitoring devices, obtain multi-source monitoring data;

[0016] Perform multi-source fusion on the multi-source monitoring data through a spatio-temporal alignment algorithm to generate real-time monitoring data based on traffic flow feature vectors.

[0017] In the embodiment, after the multi-source monitoring devices collect the multi-source heterogeneous data on the current ramp, the multi-source data are combined, and combined with the spatio-temporal algorithm, so as to solve the technical problem of multi-dimensional data collaborative acquisition in the safety assessment process of the fault-tolerant ramp.

[0018] Combined with the first aspect, the collection of real-time monitoring data on the fault-tolerant ramp further includes:

[0019] According to the meteorological sensing device, determine the rain and fog level index, and dynamically allocate the weights of the multi-source monitoring data;

[0020] Through the lidar device, cross-verify the multi-source monitoring data through spatio-temporal alignment, contour imaging segmentation, and road surface induction to determine whether there is data loss;

[0021] When there is data loss, generate a fusion compensation model based on virtual thermal imaging texture and vehicle trajectory according to the multi-source monitoring data;

[0022] Through the fusion compensation model, perform multi-source fusion compensation, and use the real-time monitoring data after multi-source fusion compensation as the target data.

[0023] In the data acquisition of real-time monitoring in this application, due to the existence of multi-source heterogeneous data, a single data acquisition scheme is likely to cause data invalidation or data errors. This application adopts a dynamic fusion method of virtual thermal imaging and vehicle trajectories, uses physical modeling to compensate for data loss, and realizes software path data compensation.

[0024] Combined with the first aspect, the Markov chain model is constructed through the following steps:

[0025] Step 1: Define the set of fault tolerance ability decay states S: S = (S1, S2... S i );

[0026] Step 2: Calculate the initial transition probability matrix P(t0) = [p ij 3×3 based on the ramp accident historical data; where, p ij = λe (-aQ) ; λ represents the traffic flow correction coefficient, Q is the real-time traffic flow; a represents the decay factor

[0027] Step 3: Dynamically update the initial transition probability matrix through real-time monitoring data to generate the second transition update matrix;

[0028] Step 4: Determine the exponential decay curve according to the second transition update matrix;

[0029] The exponential decay curve is:

[0030] F(t) = F0*exp(-∫γ(t)dt)

[0031] γ(t(=))

[0032] Step 5: Input the exponential decay curve into the dynamic risk assessment matrix to form the Markov chain model.

[0033] In this application, the role of the Markov chain model is based on the path selection algorithm of the Markov chain. Combining with the real-time monitoring data in this application, it also fuses the traffic flow correction and the continuously changing exponent of the safety situation, realizes the path analysis in the traffic field, solves the analysis of the ramp fault tolerance ability change effect, and further realizes the ramp fault tolerance ability evaluation under different traffic conditions.

[0034] Combined with the first aspect, the state transition probability analysis includes:

[0035] Obtain the ramp vehicle density gradient in real time, calculate the change rate of the number of vehicles in adjacent areas through the communication data between in-vehicle OBU and roadside unit RSU, and determine the reference transition probability matrix;

[0036] Generate a probability correction coefficient matrix based on the visibility value and road surface friction coefficient obtained by the meteorological sensor;

[0037] Perform an element-wise multiplication operation on the probability correction coefficient matrix and the reference transition probability matrix to obtain a real-time transition probability matrix, and determine the transition probability analysis result.

[0038] In this application, during the process of analyzing the state transition probability, a reference transition probability matrix generated by combining the change rate of the number of vehicles in adjacent regions is used for scenario adaptation to determine the current scenario status under the basic condition. Then, a correction coefficient matrix generated by combining the data of the meteorological sensor is used to fuse the scenario status and the correction coefficients of different safety risk conditions in the scenario status to achieve state transition. Moreover, different state transitions have different probabilities. The process of state transition is the attenuation curve of the fault tolerance ability of the fault-tolerant ramp changing with time. The reference transition probability matrix is based on the communication data of vehicles and the site, and determines the scenario factors excluding environmental factors. The real-time dynamic corrected transition probability analysis result is obtained by the element-wise multiplication of the probability correction coefficient matrix and the reference transition probability matrix.

[0039] Combined with the first aspect, the generation of the attenuation curve of the fault tolerance ability changing with time includes:

[0040] Configure a data synchronization mechanism based on the PTP protocol according to the state transition probability analysis result to determine the real-time risk value sequence;

[0041] Extract the design parameters of the fault-tolerant ramp to construct an attenuation rate compensation function;

[0042] Multiply the real-time risk value sequence by the attenuation rate compensation function to generate an attenuation curve with physical constraints; among them, a continuous curve is fitted by cubic spline interpolation. When the slope of the curve is negative, a fault tolerance level downgrade instruction is triggered.

[0043] In this application, during the analysis process of generating the attenuation curve of the fault tolerance ability changing with time for the analysis result of the state transition probability, multi-source heterogeneous data can be jointly calibrated, that is, the time alignment of heterogeneous sensor data is performed to determine the risk value at each moment. Then, through the width, road curvature, etc. in the ramp design, a function for attenuation compensation is realized. The attenuation compensation function is used to compensate for the compensation of the physical environment for the fault tolerance ability. Finally, the fault tolerance level evaluation by spline interpolation fitting is realized.

[0044] Combined with the first aspect, the establishment of a dynamic risk assessment matrix based on the non-linear mapping relationship between the ramp service level index and the safety factor includes:

[0045] Build a multi-source heterogeneous data dynamic compensation system, and real-time fuse meteorological warning data, vehicle network floating vehicle trajectory data, and driver pupil focus monitoring data through the roadside unit to construct a three-dimensional risk perception field;

[0046] Embed edge computing nodes in the dynamic risk assessment matrix for risk value compensation operations. When heavy rain weather and a high proportion of trucks are detected simultaneously, automatically activate the risk conduction model to reconstruct the evaluation parameters, and at the same time, link the variable message sign to activate the laser projection lane-level risk identification, realizing the fusion warning of millimeter-wave radar point cloud data and visual augmented reality.

[0047] In this application, for the multi-data collaboration problem composed of composite data sources, this application is based on a three-dimensional risk perception field and multi-modal warnings triggered by composites, automatically activates the risk conduction model, realizes the compensation for blind spots during the dynamic monitoring process, and enhances the sensing effect.

[0048] Combined with the first aspect, based on the non-linear mapping relationship between the ramp service level index and the safety factor, establishing a dynamic risk assessment matrix further includes:

[0049] Establish a cross-departmental risk collaborative response mechanism, integrate the traffic police command system, the medical emergency platform, and the fire emergency plan database, and implant an emergency plan matching engine in the dynamic risk assessment matrix;

[0050] When the evaluation value breaks through the threshold, automatically generate a multi-department collaborative disposal plan, synchronously adjust the ramp signal light configuration to the emergency rescue channel mode, and plan a dynamic avoidance corridor for special vehicles through Beidou high-precision positioning, forming a multi-modal emergency response link based on the risk assessment results.

[0051] In this application, by constructing a dynamic risk assessment matrix, when the ramp is configured as an emergency lane, this application will set up an emergency plan matching mechanism. When there is a need to convert the ramp into an emergency rescue channel, execute the emergency dynamic performance group corridor to realize the dynamic ramp control based on specific emergency scenarios.

[0052] Combined with the first aspect, based on the dynamic risk assessment matrix, determining the safety assessment value under different traffic flows further includes:

[0053] Obtain the real-time traffic flow density gradient distribution of any vehicle on the ramp, and construct a traffic flow pressure heat map;

[0054] When inputting the heat map data into the dynamic risk assessment matrix, synchronously load the machine learning model during the historical peak period for dynamic parameter correction to determine the safety assessment value;

[0055] When it is detected that the safety assessment value in the traffic flow density mutation area overlaps with the high-risk assessment area, automatically generate a lane-level diversion coefficient, and push dynamic lane change suggestions to the in-vehicle navigation terminal through the roadside RSU device to realize the micro-path self-optimization under the surge of traffic flow.

[0056] In this application, through the method of heat map and historical dynamic correction, according to different safety evaluation values, when the traffic flow on the ramp surges, a path switching for persuasion is performed to prevent delays.

[0057] Combined with the first aspect, the determination of the safety evaluation value under different traffic flows according to the dynamic risk assessment matrix further includes:

[0058] Set up a cloud online interface to capture the vehicle heading angle deviation rate and centripetal acceleration data of any vehicle on the ramp;

[0059] Perform spatio-temporal correlation matching on the deviation data and the dynamic risk assessment matrix to generate a curve risk compensation factor;

[0060] When the safety evaluation value exceeds the preset threshold, synchronously activate the curve gradient deceleration strategy of the variable speed limit sign, and control the intelligent road studs to generate a dynamic diversion light band to form a geometric linear adaptive protection system based on the dynamic traffic flow.

[0061] In this application, when evaluating the safety of different traffic flows, through the speed of geometric vehicles in the traffic flow pressure heat map, and under risk assessment, for lanes with high overlap, suggestions are generated to remind drivers to switch lanes and divert traffic, realizing the risk status assessment for each vehicle under different traffic flows under self-optimization.

[0062] Other features and advantages of the present invention will be described in the subsequent specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the written specification and the drawings.

[0063] The technical solutions of the present invention will be further described in detail below through the drawings and embodiments. Description of the Drawings

[0064] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention.

[0065] In the drawings:

[0066] Figure 1 is the method flow chart of the safety evaluation method based on fault-tolerant ramp online data in the present invention;

[0067] Figure 2 is the logic flow chart of the safety evaluation method based on fault-tolerant ramp online data in the present invention;

[0068] Figure 3 is the acquisition process flow chart of real-time monitoring data in the present invention;

[0069] Figure 4 is the process flow chart for realizing multi-source fusion compensation in the present invention;

[0070] Figure 5 is the process flow chart for the implementation of the Markov chain model in the present invention;

[0071] Figure 6 is the process flow chart for the analysis of state transition probability in the present invention;

[0072] Figure 7 is the process flow chart for the decay curve of fault tolerance ability changing with time in the present invention;

[0073] Figure 8 is the process flow chart for building the dynamic risk assessment matrix in the present invention;

[0074] Figure 9 is the process flow chart for the composition of the modal emergency response link in the present invention;

[0075] Figure 10 is the implementation flow chart for the micro path self-optimization in the present invention;

[0076] Figure 11 is the implementation flow chart for the geometric alignment adaptive protection system in the present invention. Detailed implementation manners

[0077] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not intended to limit the present invention.

[0078] For the technical solutions designed in this application, there are the following technical implementation scenarios and technologies:

[0079] The fault-tolerant ramp is a branch road of the lane on the highway leading to the lane of the exit road at the highway intersection, which is used to relieve traffic pressure or to make a quick U-turn and return to the correct route.

[0080] Under normal circumstances, mainly when a driver strays into a closed section due to negligence or unclear instructions, they can drive away through the fault-tolerant ramp; or, because the ramp of a certain highway is under great traffic pressure for a long time, there are often some special vehicles or the traffic flow is often relieved through the fault-tolerant ramp. Therefore, the main application scenario of this application is the risk scenario of high-density traffic flow.

[0081] This application mainly solves the problem that, in the design of the fault-tolerant ramp, compared with the traditional lane, it is relatively narrow, and mainly temporarily applicable to the fault-tolerant ramp, which is mostly used for construction and temporary traffic control during large-scale events. It is difficult for map navigation to update the relevant road paths of the fault-tolerant ramp in real time. Compared with conventional roads, traffic accidents are more likely to occur. Therefore, the fault-tolerant ramp needs to be evaluated for its fault-tolerant ability, so that when it is temporarily enabled, it can always feedback to the intelligent personnel of the fault-tolerant ramp and control the entry and application of the traffic flow.

[0082] For traditional online data collection, traditional highway ramp safety assessment technologies mostly adopt static monitoring and threshold alarm mechanisms. Through single collection devices such as loop detectors or fixed cameras, basic traffic flow data is collected, and the status is confirmed through manual inspections.

[0083] The data collection dimension is based on macroscopic parameters such as vehicle count and average speed, and does not refer to microscopic elements such as meteorological conditions, road surface friction coefficient, and driver behavior.

[0084] Therefore, in the prediction of traditional road risk situations, risk classification is mainly achieved by presetting fixed thresholds, without considering the non-linear impact of traffic flow dynamic fluctuations on the system's fault-tolerant ability.

[0085] For example, in rainy and foggy weather, the traditional system only triggers speed limit prompts through visibility sensors, and cannot incorporate the changes in road surface slipperiness and vehicle braking distance into comprehensive calculations. Compared with the actual risk evolution, the risk assessment is lagging, so the risk assessment ability is incorrect.

[0086] In terms of state transition analysis, traditional technologies mostly use linear regression or time series analysis to predict risk trends, and cannot effectively show the exponential decay characteristics of the fault-tolerant ability with the sudden increase in traffic flow.

[0087] The emergency response mechanism will be controlled by a centralized control platform. When an anomaly is detected, it needs to be manually judged by humans and then manually adjust the signal lights or issue induction information. The possibility of response delay is relatively high, and it is difficult to cope with the sudden high-density traffic flow conflicts in the ramp area.

[0088] In terms of data compensation, traditional solutions usually use mean imputation or data replacement in adjacent time periods to process missing values, resulting in distorted reconstruction of virtual vehicle trajectories, and the fault-tolerant ramp is basically a curve, and calculation deviations of centrifugal force are likely to occur in the curve area. Generally speaking, traditional technologies have three major defects: single data dimension, static model, and lagging response, and cannot meet the real-time and accurate quantitative assessment requirements of the ramp fault-tolerant ability in complex traffic environments.

[0089] This application breaks through the static assessment limitations of traditional technologies by constructing a multi-source heterogeneous data dynamic compensation system and a three-dimensional risk perception field.

[0090] It mainly fuses lidar point cloud data with virtual thermal imaging textures to achieve millimeter-wave compensation for lane-level contours in rainy and foggy weather. Combining the real-time communication between in-vehicle OBU and roadside RSU, it can accurately calculate the vehicle heading angle deviation rate and ramp curvature adaptability, enabling micro-driving behavior data to directly participate in risk assessment.

[0091] In terms of the core technology, as shown in this application Figure 1 it adopts an improved Markov chain model. By introducing a traffic flow correction coefficient and an attenuation factor to dynamically update the state transition matrix and modeling the physical constraints of the fault tolerance ability attenuation curve, compared with the traditional risk prediction of lanes in technical solutions, which is mainly a linear prediction model, the accuracy of identifying risk mutations caused by sudden increases in traffic flow in this application will be greatly enhanced.

[0092] In the risk assessment link, this application establishes a non-linear mapping relationship between service level indicators and safety factors, embeds edge computing nodes to fuse heterogeneous data from multi-modal such as meteorological early warnings and pupil focus monitoring in real time. If it detects the superposition of heavy rain and a high proportion of trucks, it will automatically reconstruct the risk conduction model and activate the laser projection lane-level identification, shortening the risk assessment response time. In response to the traditional linear data loss problem in the existing technology, this application generates a fusion compensation model through a spatio-temporal alignment algorithm and vehicle trajectory prediction, and ensures the continuous differentiability of the attenuation curve through cubic spline interpolation technology. For the data reconstruction of the bend area, the results are more accurate compared with the traditional mean imputation method. Then, through the linkage between Beidou high-precision positioning and variable message signs, a dynamic risk avoidance corridor and a gradient deceleration strategy are constructed to achieve a closed-loop response from risk perception to lane-level control. This solution integrates the separate data collection, risk calculation, emergency response and other links in traditional technologies into an adaptive multi-modal protection system, improving the active fault tolerance ability of fault-tolerant ramps in complex environments.

[0093] The specific implementation process of this application is as follows:

[0094] Example 1:

[0095] As Figure 1 and Figure 2 shown, this application provides a safety assessment method based on online data of fault-tolerant ramps, including: A safety assessment method based on online data of fault-tolerant ramps, including:

[0096] This application first collects real-time monitoring data on the fault-tolerant ramp; for the collected real-time monitoring data, according to the millimeter-wave radar, AI video analysis cameras and geomagnetic sensor arrays deployed at key nodes of the ramp. It real-time collects multi-dimensional parameters such as vehicle speed, lane occupancy, and following distance, and the data is transmitted to the edge computing node through a 5G private network.

[0097] Then, this application uses a Markov chain model to analyze the state transition probability of real-time monitoring data, generates a decay curve of the fault tolerance ability changing with time, and establishes a dynamic risk assessment matrix based on the non-linear mapping relationship between the ramp service level index and the safety factor; in the process of using the Markov chain model to analyze the real-time monitoring data, this application sets a probability matrix for evaluating the vehicle risk state as a benchmark with the vehicle density data of the ramp, and then combines the meteorological sensing data to correct the risk state in real time. According to the different traffic flow states of the ramp, cubic spline interpolation fitting based on data synchronization is realized, so as to realize the dynamic risk assessment of the non-linear mapping relationship. The dynamic risk assessment matrix can evaluate the safety situation of each area of the fault-tolerant ramp under dynamic traffic flow conditions, and then realize the safety assessment effect for different traffic flows.

[0098] Determine the safety assessment values under different traffic flows according to the dynamic risk assessment matrix.

[0099] In the process of constructing the dynamic risk assessment matrix, service level indicators are preset in advance, such as lane density and average speed. By setting a benchmark safety factor, multiple groups of risk assessment units are formed to determine the safety assessment under different traffic flows.

[0100] In the real-time monitoring of the fault-tolerant ramp of the hub ramp, by combining the Markov chain model to analyze the real-time monitoring data, the probability analysis of the real-time risk state of a single vehicle under different traffic flows on the fault-tolerant ramp, or the probability analysis under the traffic flow of a large number of vehicles, a decay curve is generated, that is, the trend of the fault-tolerant ramp decaying with time, which always reflects the fault tolerance ability of the lane and the dynamic change of the lane fault tolerance ability. Then, combined with the safety service level of the fault-tolerant ramp, and the dynamic non-linear mapping of the scenario data of the vehicle safety factor and the ramp traffic flow, the risk mutation situation in complex scenarios is captured. According to the traffic flow state, such as the change of the fault tolerance ability under the sudden increase and rapid decrease of vehicles, the dynamic risk assessment determination is realized. The dynamic risk assessment matrix supports differential assessment in multi-traffic flow scenarios, and can solve the problem of fixed thresholds, fixed ramp traffic flow, speed, etc. In this application, fixed settings do not need to be considered, but based on the real-time vehicle state, scenario environment state, traffic prompt information or command information level on the fault-tolerant ramp, extremely high real-time response is achieved.

[0101] In actual implementation, this application solves the problem of state prediction lag of traditional methods through the Markov time-varying model, and solves the dimensional disaster of multi-source data coupling evaluation through the linear mapping mechanism;

[0102] The dynamic assessment matrix can also convert the threshold alarm in the traditional method into a risk situation deduction to realize dynamic risk assessment.

[0103] This application collects multi-dimensional traffic data in real time and constructs a dynamic risk assessment matrix to prevent the non-linear attenuation law of fault tolerance ability under different traffic flows from being accurately judged in the traditional static threshold warning mode. Compared with the traditional method of fixed accident rate statistics, this application introduces a Markov chain model to dynamically correct the state transition probability, combines the coupling mapping of service level indicators and safety factors, reduces the risk assessment error in complex scenarios such as rainy and foggy weather and sudden changes in traffic flow, realizes the dynamic observation and instant control of the fault-tolerant ramp, and realizes the dynamic evolution and multi-factor coupling in the safety assessment of the fault-tolerant ramp.

[0104] Embodiment 2:

[0105] A possible way for this application to collect real-time monitoring data on the fault-tolerant ramp is as follows. Refer to Figure 3 , and the specific process includes:

[0106] This application pre-configures multi-source monitoring devices on the fault-tolerant ramp; among them, the multi-source monitoring devices include: lidar devices, road surface sensing devices, video monitoring devices, and meteorological sensing devices. For example, this application installs a mechanical lidar on the inner side of the ramp to construct a three-dimensional point cloud coverage area and capture vehicle contours in real time; on the road surface, electro-film sensors and fiber Bragg grating strain sensors can also be set at a preset distance, such as 50 meters, to generate a cross-section monitoring distance, and collect signals such as vehicle axle weight, vehicle distance, and tire pressure. Based on the real-time vehicle trajectory tracking of the panoramic camera, slipping and weaving behaviors are synchronously identified. In the possible implementation process, meteorological devices can also be deployed to monitor the humidity, visibility, and road surface conditions of the fault-tolerant ramp to achieve the integration of multi-dimensional sensing data.

[0107] Obtain multi-source monitoring data according to the multi-source monitoring devices;

[0108] In terms of the integration of multi-dimensional sensing data, this application performs multi-source fusion on the multi-source monitoring data through a spatio-temporal alignment algorithm to generate real-time monitoring data based on traffic flow feature vectors. For example, by unifying the spatio-temporal reference, through the ICP point cloud registration algorithm, the lidar point cloud and the video coordinate system are unified, the Gaussian-Kruger projection is used to convert the geographical coordinates, and based on the PTP precision clock protocol, time stamps are added to the data streams of multiple devices, and the sliding window is used to compensate for the transmission delay. At the data fusion level, by developing a feature-level fusion model, the three-dimensional vehicle coordinates of the lidar, the two-dimensional trajectory of the video, and the dynamic parameters of the road surface sensor are input into the spatio-temporal correlation matrix, and then the outliers are eliminated based on the Kalman filter to generate multi-dimensional traffic flow feature vectors, which will include key indicators such as speed standard deviation, headway variation coefficient, and lateral offset.

[0109] By constructing a deep spatio-temporal graph convolutional network, the fused feature vectors are input into the spatio-temporal attention module. A dynamic adjacency matrix is constructed based on vehicle interaction relationships to capture local correlations such as car-following and lane-changing. A recurrent control method, such as a gated recurrent unit, is used to extract the periodic features of consecutive frame data, thereby outputting standardized feature vectors, providing a data basis for the invocation of the safety model.

[0110] During the real-time monitoring data acquisition process of this application, the problem of visual blind spots in curves is solved through the spatial complementarity of lidar and video. Based on millisecond-level time synchronization, the integrity of the time series of emergency events is ensured. The feature-level fusion mechanism converts heterogeneous data into quantifiable kinetic indicators, providing high-confidence inputs for safety assessment.

[0111] The multi-source collaborative monitoring system of lidar and road surface sensing in this application combines spatio-temporal alignment algorithms to achieve centimeter-level data fusion accuracy. Compared with traditional single-loop coil detection technology, the accuracy of vehicle trajectory reconstruction is improved.

[0112] In this application, multi-source monitoring devices can overcome the problem of video monitoring failure in rainy and foggy weather through a weight dynamic allocation mechanism. When the visibility is less than 50 meters, the detection accuracy of enhanced lane occupancy rate can be continuously maintained.

[0113] In terms of data acquisition, this application uses a conventional technical means of multi-sensor combination. However, in the fault-tolerant ramp scenario, it is necessary to deal with complex weather (such as rain and snow) and sudden accidents. Meteorological sensing devices can introduce key parameters such as real-time monitoring of road surface slipperiness and visibility. It is a technical solution that dynamically correlates meteorological data, realizes fusion with meteorological data, and corrects monitoring errors. Moreover, this application solves the time resolution differences of heterogeneous data sources and the data asynchronization situation of spatial coordinate offsets.

[0114] Embodiment 3:

[0115] When this application acquires real-time monitoring data on the fault-tolerant ramp and realizes the fusion compensation of multi-source data, refer to Figure 4 , specifically including:

[0116] This application determines the rain and fog level index according to meteorological sensing devices and dynamically allocates the weights of multi-source monitoring data. The dynamic allocation of the weights of multi-source monitoring data is achieved through meteorological dynamic weight allocation and virtual thermal imaging compensation to realize reliable risk assessment under bad weather. The dynamic allocation of meteorological weights can calculate the rain and fog level index in real time by deploying a four-in-one meteorological sensor.

[0117] In this application, through a lidar device, multi-source monitoring data is cross-validated through spatio-temporal alignment, contour imaging segmentation, and road surface sensing to determine whether there is data loss. In this process, the lidar point cloud and video frames are synchronized at the millisecond level through the ICP algorithm, and timestamp alignment is achieved through the PTP protocol. Then, the three-dimensional contour of the vehicle is determined by the lidar, and the boundaries of the corresponding fault-tolerant ramps are identified. In this process, for the monitoring blind area caused by excessive traffic flow, this application will use the method of contour imaging segmentation to cross-validate with road surface sensing data. For example, the two-dimensional bounding boxes recognized by video YOLOv5 are subjected to IoU matching, and a data anomaly alarm is triggered when the matching fails; fiber grating sensors monitor the tire vibration spectrum, inversely calculate the vehicle mass, and cross-validate with the lidar speed measurement data;

[0118] When there is data loss in this application, a fusion compensation model based on virtual thermal imaging texture and vehicle trajectory is generated according to multi-source monitoring data; a dual-channel generative adversarial network is constructed, trained based on historical data, and the spatial distribution of point clouds in the missing area is input to output virtual thermal imaging texture. For example, a spatio-temporal Transformer model is used to fuse the vehicle trajectories and compensation textures in the previous few seconds to generate the motion trajectories in the next few seconds.

[0119] Through the fusion compensation model, multi-source fusion compensation is performed, and the real-time monitoring data after multi-source fusion compensation is used as the target data.

[0120] Finally, when the lidar point cloud missing rate is greater than the preset threshold in this application, the compensation model is started to generate virtual target data, and weighted fusion is performed with the measured data through Kalman filtering.

[0121] The fusion compensation model based on virtual thermal imaging texture in this application breaks through the limitations of the traditional mean imputation method, and inversely infers the microscopic traffic flow state through the kinematic characteristics of the vehicle when sensor data is missing. The cross-validation mechanism combined with lidar contour imaging reduces the calculation deviation of the vehicle's centripetal acceleration in the curved area, providing high-fidelity compensation data for risk assessment.

[0122] This application generates thermal imaging texture through lidar point cloud. This is for the texture reconstruction problem caused by sparse point cloud areas in traditional point cloud technical solutions. Through physical modeling, data filling of virtual texture is realized, and data such as speed and position in the missing area can also be predicted in combination with historical trajectories to achieve trajectory compensation, which can cope with data failure problems under meteorological dynamic changes. This application is not just simple data compensation, but software-level compensation through the physical modeling of virtual texture generation and the data-driven of trajectory prediction, using a new technical path of physical-driven software upgrade.

[0123] Example 4:

[0124] AsFigure 5 As shown, the Markov chain model is constructed through the following steps:

[0125] Step 1: Define the set S of fault tolerance ability attenuation states: S = (S1, S2... S i );

[0126] In this application, S represents the geometric shape of the attenuation state of the fault tolerance ability of the fault tolerance ramp over time. i represents the vehicle density of the traffic flow on the fault tolerance ramp;

[0127] Step 2: Calculate the initial transition probability matrix P(t0) = [p ij 3×3 based on the historical data of ramp accidents; where, p ij = λe (-aQ) ; λ represents the traffic flow correction coefficient, Q is the real-time traffic flow; a represents the attenuation factor, t0 represents the initial moment, p ij represents the state transition probability matrix when the vehicle density of the fault tolerance ramp is i and the corresponding number of historical ramp accident characteristics is j;

[0128] Step 3: Dynamically update the initial transition probability matrix through real-time monitoring data to generate the second transition update matrix;

[0129] Step 4: Determine the exponential decay curve according to the second transition update matrix;

[0130] The exponential decay curve is:

[0131] F(t) = F0*exp(-∫γ(t)dt)

[0132] γ(t) = P(t)*S

[0133] Step 5: Input the exponential decay curve into the dynamic risk assessment matrix to form the Markov chain model.

[0134] The traditional static risk assessment model cannot reflect the state migration brought about by sudden changes in traffic flow and equipment aging. This application realizes the visual prediction of the risk evolution process by constructing a dynamic Markov chain model.

[0135] In this application, when defining the state space, the fault tolerance attenuation ability is divided into multiple levels. For example: (fault tolerance rate > 90%): all equipment is normal, traffic flow density < 25 veh / km; (fault tolerance rate 70% - 90%): 1 set of lidar fails, vehicle distance detection error < 10%; multiple levels are divided according to the fault tolerance rate.

[0136] Based on historical accident data, such as multiple cases of brake failure, a transition probability matrix P(t0) is constructed. The traffic flow correction coefficient α = 1 / (1 + e^(-0.002(Q - 1500))) is introduced, where Q is the real-time traffic flow. When Q > 2000 vehicles / h, α = 1.32, significantly amplifying the state degradation probability;

[0137] By using Kalman filtering to fuse multi-source data, Bayesian update of P0 is performed. For example, when the failure rate of lidar suddenly increases, the transition probability is adjusted from 0.05 to 0.12, and the attenuation factor β is synchronously reduced. Under normal circumstances, the initial value is 0.85, and in this application, it is adjusted to 0.78 to accelerate the state transition response;

[0138] Then a two-parameter attenuation curve is constructed, and the attenuation curve is input into a three-dimensional evaluation matrix. The safety factor L and the service level S are coupled through the Sigmoid function. When it is monitored that the S3 state lasts for 10 minutes and L(t) < 0.6, lane-level control is automatically triggered: restricting heavy trucks from entering the ramp and pushing speed limit instructions through variable message signs.

[0139] This application introduces the traffic flow correction coefficient α, which solves the adaptability problem of fixed transition probabilities to the dynamic changes in traffic flow; through the Bayesian real-time update mechanism, the state transition matrix has the ability of self-learning; the two-parameter attenuation model realizes the collaborative characterization of short-term mutations and long-term trends, providing a theoretical support for proactive traffic safety management.

[0140] This application constructs physical constraints based on the improved Markov chain model through the traffic flow correction coefficient and the attenuation factor. Compared with the traditional linear regression prediction method, the fitting degree of the exponential decay trend of the ramp fault tolerance ability index is improved. The dynamically updated transition probability matrix can reflect abnormal working conditions such as sudden changes in the proportion of trucks in real time, reducing the false alarm rate of risk warnings in highway field tests.

[0141] Example 5:

[0142] As Figure 6 shown, this application conducts state transition probability analysis to address the situation where traditional static probability models cannot reflect the dynamic coupling effect of vehicle density and meteorological conditions. The specific process includes:

[0143] This application obtains the ramp vehicle density gradient in real time, calculates the change rate of the number of vehicles in adjacent areas through the communication data between in-vehicle OBU and roadside unit RSU, and determines the benchmark transition probability matrix; through the dynamic correction mechanism of the transition probability by multi-source data fusion, accurate prediction of risk evolution is achieved. The vehicle position data is obtained in real time by the roadside unit and in-vehicle OBU to construct a density grid. A spatio-temporal convolutional neural network is used to calculate the change rate of the number of vehicles in adjacent areas in a cycle to trigger a high-density warning, and the benchmark transition probability matrix is trained based on multiple sets of accident data.

[0144] This application generates a probability correction coefficient matrix based on the visibility value and road surface friction coefficient obtained by a meteorological sensor; performs Kalman filtering on the K matrix through a sliding window to eliminate the instantaneous noise of the sensor. The transition probability matrix and the correction coefficient matrix are subjected to a Hadamard product operation to generate a time-varying transition matrix. The final state prediction is based on the real-time transition probability matrix to calculate the state probability distribution within a preset future time period.

[0145] This application performs an element-wise product operation on the probability correction coefficient matrix and the reference transition probability matrix to obtain a real-time transition probability matrix and determine the result of state transition probability analysis. The probability correction coefficient matrix breaks through the coarse-grained limitation of traditional meteorological correction and realizes lane-level risk correction through the coupled calculation of the visibility value and the road surface friction coefficient. The element-wise product operation enables the spatio-temporal resolution of state transition probability analysis to reach the preset level, improving the accuracy of curve risk identification compared with the traditional regional homogenization processing method.

[0146] Example 6:

[0147] Aiming at the fact that the traditional static evaluation model cannot reflect the coupling effect of equipment aging and traffic flow mutation, refer to Figure 7 This application generates a decay curve of the fault tolerance ability over time, including:

[0148] According to the result of state transition probability analysis, a data synchronization mechanism based on the PTP protocol is configured. This application deeply adapts the PTP protocol to the traffic fault tolerance scenario: through hardware timestamps, such as the joint calibration of lidar data packets and software state transition probabilities, to solve the time alignment problem of heterogeneous sensor data. This application's physical constraint compensation and dynamic interpolation algorithm realize the accurate modeling and real-time warning of the decay curve of the fault tolerance ability. Configuring a data synchronization mechanism based on the PTP protocol specifically realizes the clock synchronization of multi-source sensors by deploying edge computing nodes based on a preset protocol. During this process, lidar, millimeter-wave radar, and fiber optic strain sensors are also arranged at key sections of the ramp, and the sampling rate differences of the devices are eliminated through a timestamp alignment algorithm. When it is detected that the time offset exceeds the preset time, an adaptive Kalman filter compensation module is triggered to ensure the time consistency of the real-time risk value sequence.

[0149] Then, this application constructs a decay rate compensation function by extracting the design parameters of the fault tolerance ramp. The construction process of the decay rate compensation function requires extracting the key design parameters of the ramp, such as: geometric features: radius, slope, sight distance; material properties: asphalt friction coefficient, anti-collision grade of steel guardrail; and then constructing the compensation function;

[0150] Finally, multiply the real-time risk value sequence by the attenuation rate compensation function to generate an attenuation curve with physical constraints. Among them, a continuous curve is fitted by cubic spline interpolation. When the slope of the curve is negative, a fault tolerance level downgrade instruction is triggered. The present application is based on the cubic spline interpolation attenuation curve modeling technology of the PTP protocol, effectively solving the problem of the jump of the fault tolerance level caused by traditional linear interpolation. A dynamic coupling mechanism of the design parameter compensation function and the real-time risk value is designed to improve the early warning sensitivity of the curve slope. During the generation process of the physical constraint attenuation curve, the real-time risk value sequence (including multi-dimensional parameters such as the coefficient of variation of the vehicle distance and the standard deviation of the speed) is multiplied by the compensation function, and a continuous curve is generated by cubic spline interpolation. The interpolation node interval is preferably set to 5 seconds to ensure the smoothness of the curve.

[0151] The consistency of the synchronization data is determined through PTP synchronization, then the physical environment is constrained based on the attenuation rate compensation function, and finally a continuous curve is generated through spline interpolation to realize slope monitoring and trigger degradation, which can capture the mutation characteristics of the fault tolerance ability, while the existing technology is mainly a linear model and cannot capture. It solves the problem of time series alignment of multi-source data and physical environment adaptation in the safety assessment of the fault tolerance ramp.

[0152] Embodiment 7:

[0153] As Figure 8 shown, the present application establishes a dynamic risk assessment matrix based on the non-linear mapping relationship between the ramp service level index and the safety factor, including:

[0154] Build a multi-source heterogeneous data dynamic compensation system, and real-time fuse meteorological warning data, vehicle network floating vehicle trajectory data and driver pupil focus monitoring data through roadside units to construct a three-dimensional risk perception field. The three-dimensional risk perception field of the present application realizes the joint modeling of driver behavior characteristics and road alignment risks for the first time through the fusion of pupil focus monitoring and millimeter wave radar point cloud data. The laser projection type identification system supported by the edge computing node speeds up compared with the traditional variable message sign and can still maintain the lane-level risk visualization guidance function in rainy weather. Because in rainy weather, compound risks such as sudden drop in visibility and brake failure often occur on the ramp. Traditional assessment models are difficult to cope with the dynamic coupling effect of meteorology and traffic flow. By using the method of the present application, the risk conduction mechanism is reconstructed through multi-source data fusion and edge computing to achieve accurate lane-level risk early warning. The present application deploys monitoring devices, and the roadside unit RSU integrates meteorological sensors, in-vehicle OBU and driver eye trackers to construct a three-dimensional risk perception field. The effect is to fuse with meteorological and trajectory data by quantifying driver state parameters such as fixation point offset and blink frequency in the driver pupil focus monitoring data to realize three-dimensional risk modeling. The three dimensions include the road, the vehicle and the person. The construction of the three-dimensional field solves the problem of spatio-temporal alignment of heterogeneous data, such as the difference in meteorological data resolution and the millisecond-level delay of vehicle network trajectories.

[0155] Embed edge computing nodes in the dynamic risk assessment matrix for risk value compensation operations. When heavy rain weather and a high proportion of trucks are detected to overlap, automatically activate the risk conduction model to reconstruct the evaluation parameters, and at the same time, link the variable message sign to activate the laser projection lane-level risk identification, so as to realize the fusion alarm of millimeter-wave radar point cloud data and visual augmented reality. Finally, realize the spatio-temporal alignment of millimeter-wave radar point cloud and video trajectory through the improved ICP algorithm; when there is video failure, start the data compensation model under the federated learning framework, generate virtual thermal imaging textures based on historical data to compensate for the visual blind area; in this process, also build a risk conduction model by embedding edge computing nodes; when there is a risk, activate the Bayesian network to update the risk threshold, increase the braking safety distance, and synchronously optimize the Kalman filter parameters. Link the variable message sign to project a dynamic laser grid. Through AR-HUD technology, fuse the millimeter-wave radar point cloud with the real scene and generate an alarm signal outside the lane line. In this application, realize the risk value compensation operation through edge computing nodes, for example, dynamically adjust the attenuation factor in the Markov chain transition probability matrix to achieve lightweight calculation. Heavy rain causes the road surface to be slippery, and the proportion of trucks increases the braking distance. The superposition of the two significantly increases the accident risk. The prior art can determine the risk based on a single factor, but for the risk in the composite situation, the sensitivity is relatively low. In terms of multi-source data fusion, the dynamic compensation system solves data conflicts: for example, when the lidar is interfered by raindrops in heavy rain weather, use the vehicle networking trajectory data to fill the target detection blind area. This application adopts a collaborative design scheme of constructing a three-dimensional risk field, real-time compensation by edge computing, and composite-trigger multi-modal alarm. For the fault-tolerant ramp, it can perform multi-source data collaboration and real-time response in dynamic risk assessment to reduce the occurrence of risk events.

[0156] Example 8:

[0157] As Figure 9 shown, based on the non-linear mapping relationship between the ramp service level index and the safety factor, this application establishes a dynamic risk assessment matrix, and further includes:

[0158] Establish a cross-departmental risk collaborative response mechanism, integrate the traffic police command system, medical emergency platform and fire emergency plan database, and implant an emergency plan matching engine in the dynamic risk assessment matrix; construct a cross-departmental collaborative dynamic risk assessment matrix to achieve hierarchical and joint disposal of emergencies such as traffic accidents and hazardous chemical leaks. For example: in cross-departmental data fusion, by accessing the traffic police command system, 120 emergency platform and fire plan library, data security sharing is achieved through the federated learning framework, and the emergency plan is automatically triggered through the implantation of a matching engine in the risk assessment matrix. For example: when the ramp assessment value exceeds the threshold, the system automatically calls the relevant emergency plan for risk alarm and associates with the traffic police, fire and medical departments through the department responsibility tagging matching algorithm; through the signal light mode switching, for example: the emergency lane indicator light turns red and the Beidou emergency evacuation corridor is planned to achieve the generation of dynamic path point cloud.

[0159] When the assessment value breaks through the threshold, a multi-department collaborative disposal plan is automatically generated, the ramp signal light configuration is synchronously adjusted to the emergency rescue channel mode, and a dynamic emergency evacuation corridor is planned for special vehicles through Beidou high-precision positioning, forming a multi-modal emergency response link based on the risk assessment results.

[0160] The collaborative response mechanism of the emergency plan matching engine and the Beidou dynamic corridor of this application breaks through the bottleneck of traditional manual scheduling efficiency through the adaptive adjustment of signal light configuration and the priority path planning of special vehicles. This solution combines the risk assessment results, for example: a 500-meter detour is required upstream of the leakage point, to form a dynamic corridor, and then solves the spatio-temporal alignment problem of path planning and risk assessment, for example: real-time obstacle avoidance based on Kalman filter. Finally, through the emergency rescue mode of signal light switching, based on the priority control protocol, the control of the signal light group is realized.

[0161] Based on knowledge graph technology, semantic associations are made between ramp design parameters, historical accident characteristics and emergency plans. When the assessment value of the dynamic risk assessment matrix is lower than the preset threshold, multi-department automatic collaborative response is carried out, real-time road conditions are integrated, and a dynamic emergency evacuation path is planned for the rescue department. Yielding instructions are pushed to surrounding vehicles through V2X communication to achieve multi-modal emergency response.

[0162] This application realizes composite emergency plan matching instead of simple keyword matching, and solves the technical problems of cross-departmental collaboration and dynamic resource scheduling in the emergency scenario of fault-tolerant ramps.

[0163] Example 9:

[0164] As Figure 10 shown, according to the dynamic risk assessment matrix, determining the safety assessment value under different traffic flows further includes:

[0165] Obtain the real-time traffic flow density gradient distribution of any vehicle on the ramp, construct a traffic flow pressure heat map, and through the coupled analysis of the real-time traffic flow heat map and the dynamic risk assessment matrix, realize microscopic path self-optimization and effectively reduce the accident rate on the ramp. During the construction of the traffic flow pressure heat map, deploy millimeter-wave radar and lidar to obtain vehicle position, speed, and inter-vehicle distance data in real time. Combine the acceleration information transmitted by the in-vehicle OBU, and generate a grid-level density gradient distribution through an improved DBSCAN clustering algorithm.

[0166] When inputting the heat map data into the dynamic risk assessment matrix, synchronously load the machine learning model during the historical peak period to dynamically correct the parameters and determine the safety assessment value.

[0167] When it is detected that the safety assessment value in the traffic flow density mutation area overlaps with the high-risk assessment area, automatically generate a lane-level diversion coefficient, and push dynamic lane change suggestions to the in-vehicle navigation terminal through the roadside RSU device to achieve microscopic path self-optimization under the condition of a sharp increase in traffic flow.

[0168] The machine learning-enhanced heat map dynamic correction technology of this application realizes sub-second risk positioning in the traffic flow density mutation area. The lane-level diversion coefficient generation algorithm combined with RSU dynamic navigation push improves the ramp passing efficiency compared with the traditional regional traffic limiting strategy.

[0169] Adopt HSV color space mapping to achieve a positive correlation between saturation and density. When the local density triggers a red warning, the dynamic update period is compressed to achieve visualizing the panoramic situation of the ramp. During the correction process of the dynamic risk assessment matrix, through the integrated LSTM time series prediction model, detect the current traffic flow increase. When the increase exceeds the preset value, activate the model parameter dynamic correction module. Match the historical similar scenarios through a sliding time window, and adjust the risk weight coefficient in real time. The way of increasing the weight enhances the sensitivity to the mutation area.

[0170] This application quantifies the spatial heterogeneity of traffic flow pressure through a gradient distribution heat map, such as: the density gradient difference between the ramp entrance and the exit, and then based on the parameter dynamic correction of the machine learning model during the historical peak period, solves the coupling problem between real-time data and historical laws. For example, when it is rainy, the model automatically reduces the heat map weight coefficient. The overlapping area detection algorithm needs to solve the spatial alignment problem between the heat map and the risk assessment matrix, such as: the grid resolution difference. The prior art uses fixed grid division, while this application realizes dynamic alignment through Kalman filtering to ensure accurate positioning of the mutation area.

[0171] This application realizes lane-level switching through microscopic path optimization. When dynamically switching lanes, it can also solve the problems of time-delay sensitivity and reliability, and solves the problems of refined evaluation and real-time optimization of fault-tolerant ramps in the scenario of a sharp increase in traffic flow.

[0172] Example 10:

[0173] As Figure 11 shown, the determining of the safety evaluation value under different traffic flows according to the dynamic risk assessment matrix further includes:

[0174] Setting up a cloud online interface to capture the vehicle heading angle deviation rate and centrifugal acceleration data of any vehicle on the ramp;

[0175] Performing spatio-temporal correlation matching on the deviation data and the dynamic risk assessment matrix to generate a curve risk compensation factor;

[0176] When the safety evaluation value exceeds the preset threshold, synchronously activate the curve gradient speed reduction strategy of the variable speed limit sign, and control the intelligent road studs to generate a dynamic diversion light band, forming a geometric linear adaptive protection system based on the dynamic traffic flow.

[0177] The curve gradient speed reduction strategy driven by the heading angle deviation rate of this application breaks through the geometric adaptability defect of the traditional fixed speed limit sign. The linkage control of the intelligent road stud light band and the centrifugal acceleration reduces the side slip accident rate of the ramp curvature radius, realizing the millisecond-level dynamic adaptation of the curve line shape and the vehicle speed.

[0178] This application aims at the difficulty of the traditional static protection system in coping with the risk of sudden change of vehicle centrifugal force. By using the method of this application, through the dynamic perception of the heading angle - centrifugal force and the linkage control of the intelligent road studs, a geometric linear adaptive protection system is constructed. Deploy millimeter-wave radar to obtain the vehicle heading angle θ in real time, and combine it with the centrifugal acceleration α transmitted by the in-vehicle OBU; the cloud interface uses the MQTT protocol to achieve 10ms-level data synchronization, and constructs a spatio-temporal matrix of the vehicle motion state. When it is detected that the deviation rate of θ is greater than the preset value, a risk event mark is triggered, and the future trajectory deviation probability is predicted through the LSTM model. Perform convolution operation on the real-time data and the dynamic risk assessment matrix to generate a curve risk compensation factor, train the Bayesian network based on historical accident data, and automatically adjust the safety threshold in rainy weather. Link the variable speed limit value, and dynamically generate the speed limit value according to the ramp curvature radius, where μ is the friction coefficient, which is measured in real time by the laser sensor. Activate the embedded type to generate a dynamic diversion light band.

[0179] This application breaks through the traditional single evaluation dimension of speed through the coupling analysis of the heading angle - centrifugal force; the risk compensation factor realizes the dynamic matching of the geometric line shape and vehicle dynamics; the light band - sound wave multi-modal warning forms a two-dimensional protection of space guidance + mechanical perception.

[0180] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A safety assessment method based on online data of fault-tolerant ramps, characterized in that Including: Collecting real-time monitoring data on the fault-tolerant ramp; Using the Markov chain model to analyze the state transition probability of the real-time monitoring data, generating an attenuation curve of the fault tolerance ability varying with time, and establishing a dynamic risk assessment matrix based on the non-linear mapping relationship between the ramp service level index and the safety factor; Determining the safety assessment value under different traffic flows according to the dynamic risk assessment matrix.

2. The safety assessment method based on online data of fault-tolerant ramps as claimed in claim 1, characterized in that The collecting of the real-time monitoring data on the fault-tolerant ramp includes: Pre-configuring multi-source monitoring devices on the fault-tolerant ramp; among them, the multi-source monitoring devices include: lidar devices, road surface sensing devices, video monitoring devices, and meteorological sensing devices; Obtaining multi-source monitoring data according to the multi-source monitoring devices; Performing multi-source fusion on the multi-source monitoring data through a spatio-temporal alignment algorithm to generate real-time monitoring data based on traffic flow feature vectors.

3. The safety assessment method based on online data of fault-tolerant ramps as shown in claim 2, characterized in that, The collecting of the real-time monitoring data on the fault-tolerant ramp further includes: Determining the rain and fog level index according to the meteorological sensing device, and dynamically allocating the weights of the multi-source monitoring data; Using the lidar device to cross-verify the multi-source monitoring data through spatio-temporal alignment, contour imaging segmentation, and road surface induction to judge whether there is data loss; When there is data loss, generating a fusion compensation model based on virtual thermal imaging texture and vehicle trajectories according to the multi-source monitoring data; Performing multi-source fusion compensation through the fusion compensation model, and taking the real-time monitoring data after multi-source fusion compensation as the target data.

4. The safety assessment method based on online data of fault-tolerant ramps as shown in claim 1, characterized in that, The Markov chain model is constructed through the following steps: Step 1: Define the set S of fault tolerance ability attenuation states: S = (S1, S2... S i ); Step 2: Calculate the initial transition probability matrix \(P(t_0)=[p\) ij \(]_{3\times3}\); where \(p\) ij \(=\lambda e\) (-aQ) \(;\lambda\) represents the traffic flow correction coefficient, \(Q\) is the real-time traffic flow; \(a\) represents the attenuation factor. Step 3: Dynamically updating the initial transition probability matrix through the real-time monitoring data to generate a second transition update matrix; Step 4: Determining the exponential decay curve according to the second transition update matrix; The exponential decay curve is: F(t) = F0 * exp(-∫γ(t)dt) γ(t) = P(t) * S Step 5: Inputting the exponential decay curve into the dynamic risk assessment matrix to form the Markov chain model.

5. The safety assessment method based on online data of fault-tolerant ramps as claimed in claim 1, characterized in that The performing of the state transition probability analysis includes: Obtaining the ramp vehicle density gradient in real time, calculating the change rate of the number of vehicles in adjacent areas through the communication data between the on-vehicle OBU and the roadside unit RSU, and determining the reference transition probability matrix; Generating a probability correction coefficient matrix based on the visibility value and road surface friction coefficient obtained by the meteorological sensor; Performing an element-wise product operation on the probability correction coefficient matrix and the reference transition probability matrix to obtain the real-time transition probability matrix, and determining the result of the state transition probability analysis.

6. The safety assessment method based on online data of fault-tolerant ramps as claimed in claim 1, wherein The generating of the attenuation curve of the fault tolerance ability varying with time includes: Configuring a data synchronization mechanism based on the PTP protocol according to the result of the state transition probability analysis to determine the real-time risk value sequence; Extracting the design parameters of the fault-tolerant ramp and constructing an attenuation rate compensation function; Multiplying the real-time risk value sequence by the attenuation rate compensation function to generate an attenuation curve with physical constraints; among them, fitting a continuous curve through cubic spline interpolation, and triggering a fault tolerance level downgrading instruction when the curve slope is negative.

7. The safety assessment method based on online data of fault-tolerant ramps as claimed in claim 1, characterized in that The establishing of the dynamic risk assessment matrix based on the non-linear mapping relationship between the ramp service level index and the safety factor includes: Build a multi-source heterogeneous data dynamic compensation system, and construct a three-dimensional risk perception field by fusing meteorological warning data, floating vehicle trajectory data of the vehicle network, and driver pupil focus monitoring data in real time through roadside units; Embed edge computing nodes in the dynamic risk assessment matrix for risk value compensation operations. When it is detected that heavy rain weather overlaps with a high proportion of trucks, automatically activate the risk conduction model to reconstruct the evaluation parameters, and at the same time link the variable message sign to start the laser projection lane-level risk identification, realizing the fusion alarm of millimeter-wave radar point cloud data and visual augmented reality.

8. The safety assessment method based on online data of fault-tolerant ramps as shown in claim 7, characterized in that, Based on the non-linear mapping relationship between the ramp service level index and the safety factor, establish a dynamic risk assessment matrix, and further include: Establish a cross-departmental risk collaborative response mechanism, and implant an emergency plan matching engine in the dynamic risk assessment matrix based on the emergency plan database; When the evaluation value breaks through the threshold, automatically generate a multi-department collaborative disposal plan, synchronously adjust the ramp signal light configuration to the emergency rescue channel mode, and plan a dynamic avoidance corridor for special vehicles through Beidou high-precision positioning, forming a multi-modal emergency response link based on the risk assessment results.

9. The safety assessment method based on online data of fault-tolerant ramps as shown in claim 1, characterized in that, According to the dynamic risk assessment matrix, determine the safety assessment value under different traffic flows, and further include: Obtain the real-time traffic flow density gradient distribution of any vehicle on the ramp, and construct a traffic flow pressure heat map; When inputting the traffic flow pressure heat map into the dynamic risk assessment matrix, synchronously load the machine learning model during the historical peak period to dynamically correct the parameters and determine the safety assessment value; When it is detected that the safety assessment value in the traffic flow density mutation area overlaps with the high-risk assessment area, automatically generate a lane-level diversion coefficient, and push dynamic lane change suggestions to the in-vehicle navigation terminal through the roadside RSU device, realizing the micro-path self-optimization under the surge of traffic flow.

10. The safety assessment method based on online data of fault-tolerant ramps as shown in claim 1, characterized in that, According to the dynamic risk assessment matrix, determine the safety assessment value under different traffic flows, and further include: Set up a cloud online interface to capture the vehicle heading angle deviation rate and centripetal acceleration data of any vehicle on the ramp; Perform spatio-temporal correlation matching on the deviation data and the dynamic risk assessment matrix to generate a curve risk compensation factor; When the safety assessment value exceeds the preset threshold, synchronously activate the curve gradient speed reduction strategy of the variable speed limit sign, and control the intelligent road studs to generate a dynamic diversion light band, forming a geometric alignment adaptive protection system based on the dynamic traffic flow.

Images