Vehicle operation safety level evaluation method based on digital twinning and fuzzy algorithm

Abstract

The application discloses a kind of vehicle operation safety level evaluation methods based on digital twin and fuzzy algorithm, comprising: by constructing scene twinborn model and vehicle twinborn model in the range of road network, scene complexity index is generated based on the number of intersection, lane occupancy and speed fluctuation clustering and normalization, scene layer and vehicle layer hierarchical fuzzy reasoning structure is designed, and in online evaluation stage, fuzzy rule weight is adaptively adjusted according to scene complexity, so as to obtain the target vehicle operation safety level result dynamically updated with scene change;Effectively solve the problem that existing digital twin traffic safety evaluation technology and traffic operation state evaluation technology based on fuzzy comprehensive evaluation cannot consider the difference of complex road scene, vehicle individual operation characteristics and safety level interpretability.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent traffic safety monitoring and vehicle operation status assessment, and in particular to a method for assessing vehicle operation safety levels based on digital twins and fuzzy algorithms. Background Technology

[0002] With the rapid development of vehicle-to-everything (V2X), vehicle-road cooperative systems, and autonomous driving technologies, online monitoring and safety risk assessment of vehicle operation status are gradually evolving from judging static indicators of a single vehicle to integrated modeling of "human-vehicle-road-environment." Traditional solutions are mostly based on trigger-based rules from onboard sensors, such as issuing warnings for dangerous events based on thresholds for vehicle speed, braking intensity, and deviation, which are difficult to reflect the comprehensive impact of road network structure, traffic flow organization, and local congestion on vehicle safety. Although some studies have introduced machine learning or deep learning models to predict driving risks or accident probabilities, they often construct feature spaces centered on vehicles or drivers, and do not consider factors such as intersection density and lane... The "scenario complexity" such as occupancy rate and speed fluctuation lacks explicit characterization, and the model parameters are basically fixed once deployed, making it difficult to dynamically adjust the safety assessment logic according to changes in road scenarios. At the same time, existing fuzzy comprehensive evaluation methods are mostly oriented towards the road network operation status or congestion level, and the evaluation level and indicator weights are relatively static, making it difficult to take into account the fine division of the safety level of individual vehicle operation and the interpretability of the results in complex traffic scenarios. As a result, in highly complex scenarios such as complex intersections, ramp merging, and mixed traffic, the existing vehicle safety assessment results are prone to problems such as delayed response to risk levels or excessively large level jumps, affecting the reliability of the assessment results in vehicle-mounted decision-making and regulatory practices.

[0003] CN120579725A discloses an intelligent monitoring system for key road transport vehicles. This system comprises a closed-loop structure consisting of a data acquisition and fusion module, a cognitive digital twin construction module, a risk prediction and assessment module, and an adaptive intervention decision-making module. It constructs a cognitive digital twin state vector based on heterogeneous graph neural networks and LSTM, and combines prediction uncertainty quantification and reinforcement learning intervention strategies to perform time-series prediction and proactive intervention of operational risks for key transport vehicles, improving upon the shortcomings of traditional static rule models in terms of risk predictability and self-correction capabilities. However, this scheme primarily models the driver's state and the overall risk evolution of the vehicle, lacking a scene feature extraction and clustering process centered on the number of intersections, lane occupancy rate, and speed fluctuations. It also fails to construct a scene complexity index system suitable for hierarchical fuzzy reasoning, and does not establish a unified hierarchical safety level assessment framework that integrates scene-level risk levels with micro-relationships such as vehicle distance, relative speed, and lateral offset. Furthermore, the rule weights are not adaptively adjusted according to scene complexity.

[0004] CN116543552A discloses a simulation evaluation method and system for urban arterial road congestion mitigation strategies based on vehicle-road cooperation. By constructing an OD tracing platform, a multi-objective vehicle allocation model, a VISSIM road network simulation model, and a multi-level fuzzy comprehensive evaluation system, it comprehensively evaluates the road network operation status of vehicle-road cooperative congestion mitigation strategies, improving the quantitative analysis capability of urban arterial road congestion management solutions. However, this type of method focuses on the macro-level road network traffic efficiency and congestion mitigation effect, and evaluates the operation status at the scheme-road network level. It does not construct a vehicle twin model for individual vehicles within a digital twin framework, nor does it couple the dynamic scenario complexity with the individual vehicle operation status to form an online updated vehicle operation safety level result. This makes it difficult to meet the need for a refined and stable evaluation of the target vehicle safety level in complex road scenarios.

[0005] Given the shortcomings of existing digital twin traffic safety assessment technologies and traffic operation status assessment technologies based on fuzzy comprehensive evaluation, which struggle to simultaneously address the differences in complex road scenarios, individual vehicle operating characteristics, and the interpretability of safety levels, this invention constructs scenario twin models and vehicle twin models within the road network. Based on the number of intersections, lane occupancy rates, and speed fluctuations, a scenario complexity index is generated through clustering and normalization. A hierarchical fuzzy inference structure is designed for the scenario and vehicle layers. During the online assessment phase, the weights of fuzzy rules are adaptively adjusted according to scenario complexity, thereby obtaining a dynamically updated target vehicle operation safety level result that changes with the scenario. Therefore, the problem this invention aims to solve is how to construct a vehicle operation safety level assessment method that combines scenario sensitivity and hierarchical fuzzy interpretation capabilities in complex road traffic scenarios. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0007] In view of the aforementioned existing problems, the present invention is proposed.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: As a preferred embodiment of the vehicle operation safety level assessment method based on digital twins and fuzzy algorithms described in this invention, the method involves: collecting traffic flow, road structure, and vehicle operation status data within the road network, and constructing a scene twin model and a vehicle twin model based on the collected data. Based on the scenario twin model, the scenario features of the number of intersections, lane occupancy rate and speed fluctuation are extracted, and the scenario complexity index is obtained through clustering and normalization. Based on the scene complexity index, a scene-layer fuzzy inference sub-model is constructed to output the scene risk level. Then, at the vehicle layer, the scene risk level is combined with the distance between vehicles, relative speed, and lateral offset to obtain the target vehicle candidate safety level. During the online evaluation phase, the weights of the hierarchical fuzzy rules are adaptively adjusted based on the scenario complexity index, and the target vehicle's operational safety level is output based on the real-time status of the vehicle twin model.

[0009] The beneficial effects of this invention are as follows: By constructing scene twin models and vehicle twin models within the road network, this invention clusters and normalizes scene features such as the number of intersections, lane occupancy rate, and speed fluctuations to form a scene complexity index. Based on this, a hierarchical safety assessment structure is constructed that combines scene-level fuzzy inference and vehicle-level fuzzy inference. During the online assessment stage, the weights of the hierarchical fuzzy rules are adaptively adjusted according to the scene complexity index, and the target vehicle's operational safety level result is given in combination with the real-time status of the vehicle twin model. This solves the problems of existing safety assessment models being insensitive to different road scene complexities and struggling to balance the risk amplification effect of highly complex scenes with the time stability of results. It achieves the effects of improving the accuracy of vehicle operational safety level assessment, enhancing the adaptability of assessment results to scene differences, and reducing abrupt changes in safety levels under complex road sections and traffic congestion conditions. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the vehicle operation safety level assessment method based on digital twins and fuzzy algorithms as shown in this invention. Detailed Implementation

[0011] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0012] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0013] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0014] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a method for assessing vehicle operational safety levels based on digital twins and fuzzy algorithms, which specifically includes the following steps: S1. Collect traffic flow, road structure, and vehicle operating status data within the road network area, and construct scene twin models and vehicle twin models based on the collected data. Note that the following points should be noted in this step: S1.1 Obtain road structure data, including road centerline, lane boundaries, speed limit information and signal control parameters, within the road network area, and generate road network topology in a unified coordinate system; It should be noted that within the road network, road structure data, including road centerlines, lane boundaries, speed limits, and signal control parameters, can be exported from existing road geographic information systems. Road centerline and lane boundary data can be obtained from high-precision electronic map data, recording the geometry and number of lanes of each road in vector form. Speed ​​limit information records road type, speed limit value, and special speed limit sections through road attribute fields. Signal control parameters are exported from the traffic light control system, including the timing scheme and phase sequence of traffic lights at each intersection. After importing the above data, coordinate transformation is performed on the road centerlines and lane boundaries under a unified coordinate system, converting different source coordinate systems into unified planar coordinates or geographic coordinates. Road nodes and connections are automatically identified based on the road intersection topology, constructing the road network topology.

[0015] For example, a node coding method based on road intersections can be used to uniformly manage road numbers and node numbers, generating a road network topology data structure that includes a node table, a connection table, and an attribute table.

[0016] S1.2 Collect traffic flow data and target vehicle operation status data of each road segment at preset time intervals in the road network topology, and record a unified timestamp and road network identifier for the traffic flow data and vehicle operation status data; It should be noted that the preset time interval is set according to business needs. For example, 1 second is selected as the fixed sampling period in the urban expressway scenario to balance real-time performance and data volume control. Traffic flow data includes the number of vehicles passing through each road segment in the current time interval, average speed, and headway index. Target vehicle operation status data includes the road network identifier of the target vehicle, displacement along the road centerline, longitudinal speed, lateral offset, acceleration, and vehicle orientation angle. During collection, a timestamp and road network identifier are uniformly added to each traffic flow record and target vehicle operation status record so that the two types of data can be associated with the same time axis and the same road identifier in subsequent processing.

[0017] For example, the timestamp uses millisecond-level Unix time, and the road network identifier uses a combination of road number and direction identifier.

[0018] S1.3 Align road structure data, traffic flow data, and vehicle operation status data under a unified coordinate system and a unified time reference to generate a fusion dataset containing lane-level traffic status and vehicle pose. In a preferred embodiment, the displacement and lateral offset in the vehicle operation status data are transformed into coordinates based on the road centerline and lane boundaries in the road structure data, so that they correspond to specific lane positions in a unified coordinate system. According to the timestamps attached to each record, the traffic flow data and vehicle operation status data are aligned under a unified time reference. A time window division method is adopted to divide the continuous time axis into multiple fixed-length time windows, for example, each time window is 10 seconds long. For the original sampling points within each time window, the corresponding traffic flow records and vehicle operation status records can be filtered out according to the start and end times of the time window, and arranged in the order of timestamps within the time window.

[0019] Furthermore, traffic state data for each lane within the time window is statistically analyzed at the lane level, such as average flow rate, average speed, and occupancy rate. The corresponding target vehicle poses (position, speed, acceleration, and lateral offset) are then combined to generate a fusion dataset containing lane-level traffic state and vehicle poses.

[0020] S1.4 Construct a scene twin model containing lane-level grid indexes and conflict area identifiers based on the fused dataset, and generate a corresponding vehicle twin model for each target vehicle's running trajectory in the scene twin model.

[0021] In a preferred embodiment, the centerline of each lane is discretely sampled based on road structure data. Sampling can be performed on the geometric curve of the lane centerline at fixed spatial intervals, such as generating a sampling point every 1 meter. Under a unified coordinate system, all sampling points are numbered, and lane-level grid units are divided between each pair of adjacent sampling points. Each grid unit records the starting coordinates, ending coordinates, lane number, and length information. A lane-level grid index is generated based on the grid unit number, providing discrete spatial units for subsequent trajectory mapping and conflict area identification.

[0022] In a preferred embodiment, based on vehicle operating status data and lane-level grid index, the poses of the target vehicle and other vehicles at each sampling time are mapped to the corresponding lane-level grid cells. For each grid cell, the number of vehicle trajectories passing through the cell within a certain time window, the number of times they enter and leave from different directions, the speed distribution, and the headway distribution are counted to describe the distribution of vehicle trajectories and their relationship with the direction of entry and exit within the spatial cell. By analyzing the interweaving of trajectories from different lane directions or different entrance sources in the same grid cell, a set of candidate conflict areas can be obtained.

[0023] In a preferred embodiment, a conflict intensity threshold is set in the candidate conflict region set based on the vehicle trajectory intersection angle, relative speed difference, and historical accident weight. Specifically, the trajectory intersection angle can be divided into several intervals, such as less than 30 degrees, 30-60 degrees, and greater than 60 degrees, corresponding to different potential conflict severity levels. The relative speed difference is based on different value ranges for high speed, low speed, and rear-end collision risks. The historical accident weight is assigned based on the number and type of historical accidents near the grid cell.

[0024] For example, the conflict intensity threshold is set as the upper boundary of the medium risk range. When a lane-level grid cell has a conflict intensity score that exceeds the threshold after comprehensively considering the above three factors, the grid cell is marked as a conflict area. For example, in an urban intersection scenario, the threshold can be set to 0.6, which corresponds to an area with a higher probability of conflict.

[0025] In a preferred embodiment, lane-level grid indexes and conflict zone identifiers are associated with road structure data and traffic flow data. Specifically, for each grid cell, the system records its road number, lane number, and direction information, and appends the traffic flow and lane occupancy rate of the grid cell in different time windows, as well as whether it belongs to a conflict zone, to the same data structure. Through a hierarchical organization, a road structure layer, a lane-level traffic state layer, and a conflict zone identifier layer are established in the scene twin model, thereby describing the road geometry and the traffic state that changes over time under the same spatial indexing system.

[0026] In a preferred embodiment, the pose time series of each target vehicle's running trajectory is extracted from the fused dataset in the scene twin model. For each trajectory point, the corresponding lane-level grid index is matched according to its position coordinates and timestamp, so that the pose time series and the lane-level grid index sequence correspond one-to-one. In this way, within any time window, the target vehicle's running trajectory can be located to a specific grid cell in the scene twin model, which is convenient for association with the trajectories of surrounding vehicles and conflict area information.

[0027] In a preferred embodiment, based on the pose time series and lane-level grid index sequence, the operating status of adjacent vehicles and the nearest conflict area identifier are associated at each timestamp. Specifically, for the grid cell where the target vehicle is located at the current timestamp, the operating status data of other vehicles in several surrounding grid cells are searched, and parameters such as the headway, lateral distance, and relative speed between the target vehicle and adjacent vehicles are calculated. Based on the nearest conflict area identifier, the local traffic state corresponding to that timestamp is determined. By repeating the above operations on all timestamps, a local traffic state sequence corresponding to each target vehicle's operating trajectory is generated.

[0028] In a preferred embodiment, a vehicle twin model state vector is constructed based on a local traffic state sequence, incorporating pose parameters, lane-level grid indexes, relative distances, relative speeds, and conflict zone identifiers into the same state vector. Specifically, the state vector for each timestamp includes the target vehicle's position, speed, orientation, and lateral offset within the current grid, the distance and relative speed between the target vehicle and the nearest preceding and lateral vehicles, and whether the current grid and surrounding grids belong to a conflict zone. While constructing the state vector, a unique vehicle twin model identifier is assigned to each target vehicle's trajectory to facilitate rapid retrieval of the corresponding temporal state information within the scene twin model.

[0029] In a preferred embodiment, the vehicle twin model is updated synchronously when the scene twin model is updated. Specifically, when lane-level grid indexes in the scene twin model are added or deleted due to road topology adjustments or data updates, the grid index field in the vehicle twin model's state vector is synchronously adjusted according to the correspondence between grid indexes and timestamps to avoid invalid indexes. When the conflict area identifier changes in a new accident statistics period, the conflict state of the corresponding grid in the vehicle twin model's state vector is updated according to the latest identifier, so that the state vector of each timestamp in the vehicle twin model always reflects the latest scene twin model information.

[0030] For example, the scene twin model is updated in batches at midnight every day, and the batch synchronization process of the vehicle twin model is triggered during the update.

[0031] Preferably, this step involves collecting road structure data, traffic flow data, and vehicle operation status data under a unified coordinate system and time reference. It then constructs a scene twin model containing lane-level grid indexes and conflict area markers. Based on this, a vehicle twin model corresponding to the target vehicle's trajectory is built. This solves the technical problem in traditional traffic safety assessments where vehicle status and road structure / local traffic conflict relationships are recorded separately, making unified analysis at the lane-level granularity difficult. By combining pose parameters, lane-level grid indexes, relative distance, relative speed, and conflict area markers into a temporal state vector, the vehicle's operation process is fully mapped in the virtual scene. The interaction between the vehicle and its surrounding environment is more detailed, thereby improving the accuracy of risk identification and spatial positioning capabilities in subsequent scene complexity and safety level assessments.

[0032] S2. Extract scene features such as the number of intersections, lane occupancy rate, and speed fluctuations from the scene twin model, and obtain the scene complexity index through clustering and normalization. Note that the following should be noted in this step: S2.1 Based on the lane-level grid index in the scene twin model, count the number of grids belonging to the intersection, lane occupancy rate and speed standard deviation in each time window, and generate a scene feature vector set containing the number of intersections, lane occupancy rate and speed fluctuation; In a preferred embodiment, using pre-divided time windows as units, all grid cells within each time window are traversed according to the lane-level grid index in the scene twin model. Further, based on the relationship between intersection nodes and adjacent grid cells in the road network topology, the number of grids belonging to the intersection range within each time window is counted and recorded as the intersection quantity feature. Then, based on the lane occupancy records in the lane-level traffic state layer, the occupancy rate of each lane-level grid within the time window is counted, and the occupancy rates of all grids are averaged or weighted averaged to obtain the lane occupancy rate feature of the current time window. Then, based on the speed records of vehicles in each grid, the standard deviation of all vehicle speed samples within the time window is calculated and used as the speed fluctuation feature. Through the above process, a scene feature vector composed of the number of intersections, lane occupancy rate, and speed standard deviation is constructed for each time window.

[0033] For example, in an urban arterial road scenario, the number of grids to which an intersection belongs within a single time window can be 10 to 50, the lane occupancy rate can vary between 0.2 and 0.8, and the speed standard deviation can vary depending on whether it is a peak or off-peak period.

[0034] S2.2. The number of intersections, lane occupancy rate and speed fluctuation in the scene feature vector set are normalized without dimension. Weighting factors are set according to the impact of the number of intersections and lane occupancy rate on safety. The normalized scene features are then weighted and combined to obtain a weighted scene feature vector set. In a preferred embodiment, the three components of the scene feature vector set are subjected to dimensionless normalization. A linear normalization method based on the historical observation range can be used to map the number of intersections, lane occupancy rate, and speed fluctuation to the interval [0,1], respectively. After normalization, weight factors are set according to the degree of impact of the number of intersections and lane occupancy rate on safety. For example, the weight of the number of intersections is set to 0.5, the weight of lane occupancy rate is set to 0.3, and the weight of speed fluctuation is set to 0.2. For each time window sample, the three normalized features are weighted and combined according to the above weights to form a weighted scene feature vector.

[0035] S2.3. A density-based clustering algorithm is used to cluster the weighted scene feature vector set. The scene complexity level is divided according to the number of intersections, lane occupancy rate and speed fluctuation range of each cluster center, and a corresponding scene complexity level label is assigned to each time window. Specifically, the distance distribution between weighted scene feature vector samples is calculated in the feature space composed of the number of intersections, lane occupancy rate and speed fluctuation. For any two time window samples, Euclidean distance is selected as the distance metric. Then, based on the distance sorting curve from each sample to its k nearest neighbor sample, the density radius threshold and minimum sample number parameter are selected. Finally, a density-based clustering algorithm is used to divide the samples into clusters and noise point sets. Furthermore, the average value of the weighted scene feature vector within each cluster is calculated for the three components of number of intersections, lane occupancy rate, and speed fluctuation, generating the corresponding cluster center feature vector. Each cluster center feature vector reflects the typical characteristics of this type of time window in terms of number of intersections and traffic conditions, which facilitates subsequent scene complexity scoring. Weights are assigned based on the impact of the number of intersections, lane occupancy rate, and speed fluctuations on safety. The cluster center feature vectors are then weighted and summed to form a scene complexity score. In this embodiment, the following scoring formula is used: in, Score the scenario complexity for a given cluster; This is the normalized intersection count component in the feature vector of the cluster center; This is the normalized lane occupancy component in the feature vector of the cluster center; This is the normalized velocity fluctuation component in the feature vector of the cluster center; , , These are the weighting coefficients for the number of intersections, lane occupancy rate, and speed fluctuation, respectively. For example, take To highlight the safety impact of densely populated intersections and high-occupancy road sections; and based on The numerical range is used to classify each cluster into multiple scene complexity levels from low to high, such as low complexity, medium complexity and high complexity levels. In a preferred embodiment, for each time window sample in the weighted scene feature vector set, the cluster to which it belongs is determined according to the clustering result, and the corresponding scene complexity level is used as the scene complexity level label of the time window; specifically, a unique level identifier is assigned to each cluster, and the level identifier is written into the corresponding time window record, so that subsequent steps can directly obtain the scene complexity level label through the time window index. In a preferred embodiment, for time window samples in the noise point set, the scene complexity level corresponding to the nearest cluster is selected based on the distance to the feature vector of each cluster center, and the selected scene complexity level is written into the scene complexity level label of the corresponding time window. In this way, even if individual time window samples are classified as noise points in density clustering, they can be assigned a reasonable scene complexity level based on their proximity to typical scene features, thus avoiding unlabeled time windows in the subsequent evaluation process.

[0036] S2.4. Generate a scene complexity index based on the correspondence between scene complexity level labels and weighted scene feature vectors.

[0037] It should be noted that the complexity level corresponding to each time window can be represented by a numerical mapping, for example, low complexity is mapped to 1, medium complexity to 2, and high complexity to 3, or it can be based on the complexity score of the clusters. It is directly used as a scene complexity indicator; for all time windows on the timeline, a scene complexity indicator sequence is constructed, denoted as... ,in Let be the scene complexity index for the t-th time window, used for subsequent fuzzy inference and hierarchical weight adjustment.

[0038] Preferably, this step extracts the number of intersections, lane occupancy rate, and speed fluctuation features at the time window granularity. After normalization, weighting, and density clustering, a sequence of scene complexity indicators reflecting the differences in different road scenarios is obtained. This solves the technical problem in traditional traffic safety analysis of the difficulty in quantifying the impact of road structure complexity and traffic operation status on risk. By introducing weighted features and scene complexity scores, scenarios with different intersection densities and traffic flow levels are numerically distinguished. In subsequent fuzzy inference, a higher basic risk level can be given for complex scenarios, thereby improving the sensitivity of complex road sections and high-flow areas in vehicle safety level assessment.

[0039] S3. Construct a scene-layer fuzzy inference sub-model based on the scene complexity index to output the scene risk level, and combine the scene risk level with the distance between vehicles, relative speed, and lateral offset at the vehicle layer to obtain the candidate safety level of the target vehicle. Note that the following should be noted in this step: S3.1. Set the input membership function of the scene complexity index according to the numerical distribution of the scene complexity index on the time axis, and define the output membership function of the scene risk level and the low, medium and high levels at the scene layer; It should be noted that the scenario complexity index in the statistical historical data... The range and probability distribution of the values ​​are divided into three intervals: low, medium, and high. For example, when the value is [1,3], [1,1.7] is defined as the low complexity interval, [1.7,2.3] as the medium complexity interval, and [2.3,3] as the high complexity interval. At the scene layer, output membership functions are defined for the scene risk level, and low risk, medium risk, and high risk are respectively assigned to different output fuzzy sets. The membership function can take the form of a triangle, with a high membership degree between the low complexity interval and the low risk set, and a high membership degree between the high complexity interval and the high risk set. The middle interval gradually changes between the three risk levels so that the scene near the boundary of the complexity index can be assigned an appropriate risk level according to the actual situation.

[0040] S3.2. Based on the correspondence between the scenario complexity index and historical accident samples, set up a fuzzy rule table for the scenario layer, and input the scenario complexity index into the scenario layer fuzzy inference sub-model to obtain the scenario risk level corresponding to each time window. Specifically, based on the accident markers and corresponding scenario complexity indices for each time window in the historical accident samples, the frequency and severity of accidents within different scenario complexity index intervals are statistically analyzed. For example, the scenario complexity index range can be divided into several sub-intervals, and the number of accidents and the total number of time windows can be counted for each sub-interval, and the accident frequency can be calculated. in, Let be the accident frequency of the k-th complexity sub-interval; This represents the number of incidents that occurred within this sub-interval. This represents the total number of time window samples within the sub-interval. For the severity of the accident, weight coefficients can be assigned according to the accident level. For example, a minor accident has a weight of 1, a general accident has a weight of 2, and a serious accident has a weight of 3. The weighted average severity of each sub-interval is then calculated to obtain the correlation between the scenario complexity index and the accident risk, which is used to classify the scenario risk level. Based on the above correlation, the membership interval of the scenario complexity index is correlated with the accident frequency and the accident severity level to define low-risk, medium-risk, and high-risk scenario risk levels. A risk index can be constructed by combining accident frequency and accident severity. in, The risk index for the k-th complexity sub-interval; Accident frequency; For weighted average severity; based on The numerical distribution of the complexity index range is used to divide the risk level into three categories, such as... The lower range corresponds to a lower risk level. The median value corresponds to a medium-risk level. Higher intervals correspond to higher risk levels, and corresponding membership degrees are set for these intervals in the membership function; Furthermore, in the scene-layer fuzzy inference sub-model, the scene complexity index corresponding to each time window is input into the scene-layer fuzzy rule table, and the membership degree of each scene risk level is obtained according to the preset membership function; for a certain time window, its scene complexity index The membership degrees of the three fuzzy sets corresponding to low risk, medium risk, and high risk are obtained through membership function mapping, and are denoted as follows: , , According to preset rules, such as when the complexity is high and the accident frequency is high, the scenario risk level tends to be high risk, and the membership degree is inferred and calculated; during the defuzzification process, a weighted average method can be used to calculate the scenario risk score: in, The scenario risk score for the t-th time window; , These are representative scores for low risk, medium risk, and high risk, respectively; for example, 1, 2, and 3 can be used. , , These are the complexity metrics Membership degree in the corresponding risk level fuzzy set; In a preferred embodiment, based on scenario risk scoring The comparison between the scenario risk score and the preset risk threshold range categorizes the scenario risk score into low-risk, medium-risk, or high-risk levels; three risk thresholds can be set. , , ,in Corresponding to low risk, Corresponding to medium risk, Corresponding to high risk; For example, take ,when When the time window falls within the low-risk classification threshold range, it is marked as a low-risk scenario risk level; when When the risk level falls within the medium-risk classification threshold range, it is marked as a medium-risk scenario; when When a scenario falls within the high-risk grading threshold range, it is marked as a high-risk scenario. At the dividing point, the scenario risk level of the side with the higher accident frequency is selected based on the comparison of accident frequencies between adjacent intervals to improve the sensitivity to critical scenario risks.

[0041] S3.3 Set the vehicle layer membership function with vehicle distance, relative speed, lateral offset and scene risk level as input variables, and adjust the weight coefficients of each input variable in the vehicle layer fuzzy rule table according to the scene risk level; Specifically, the distance between vehicles can be divided into three fuzzy intervals: short distance, medium distance, and long distance; the relative speed can be divided into three fuzzy intervals: chasing, parallel, and discrete; and the lateral offset can be divided into three fuzzy intervals: near the lane edge, stable in the middle, and near the lane line. The scenario risk level is taken from the aforementioned low risk, medium risk, and high risk. In the vehicle-layer fuzzy rule table, fuzzy conclusions for output candidate safety levels are preset for different input combinations. For example, when the distance between vehicles is short, the relative speed is chasing, and the scenario risk level is high, the candidate safety level is low. At the same time, the weight coefficients of each input variable in the vehicle-layer fuzzy rule table are adjusted according to the scenario risk level. For example, in high-risk scenarios, the influence weight of the distance between vehicles and the relative speed in the fuzzy rules is increased, while in low-risk scenarios, the influence level of the lateral offset in the rules is appropriately reduced.

[0042] S3.4 Input the distance between vehicles, relative speed, lateral offset and scene risk level into the vehicle layer fuzzy inference model according to the time window, and output the candidate safety level of the target vehicle in each time window according to the vehicle layer fuzzy rule table.

[0043] Specifically, for the target vehicle status within each time window, the membership degree of each input variable on different fuzzy sets is calculated based on the vehicle layer membership function. Then, reasoning is performed based on the vehicle layer fuzzy rule table to obtain the fuzzy output set of candidate safety levels. The fuzziness is then resolved by weighted averaging or centroid method to obtain the candidate safety level of the target vehicle within that time window. The candidate safety level can be represented by a graded numerical value, such as level 1 to 4, with a higher value indicating a higher risk.

[0044] Preferably, this step constructs a fuzzy inference sub-model based on scene complexity indicators at the scene layer to obtain a scene risk score and risk level that reflects the comprehensive risk of road structure and traffic conditions. At the vehicle layer, the scene risk level is combined with the distance between vehicles, relative speed, and lateral offset to obtain the candidate safety level of the target vehicle. This solves the technical problems of separating the evaluation of scene risk and local vehicle behavior in traditional methods and making it difficult to comprehensively consider road complexity and vehicle interaction status within a unified framework. Through hierarchical fuzzy inference, scene risk and vehicle behavior are clearly divided and correlated in safety level determination, making the determination of vehicle safety level under different complexity scenarios more reasonable, thereby making the system more sensitive when identifying high-risk combination scenarios.

[0045] S4. During the online evaluation phase, the weights of the hierarchical fuzzy rules are adaptively adjusted based on the scenario complexity index, and the target vehicle's operational safety level result is output based on the real-time status of the vehicle twin model. It should be noted that in this step: S4.1 During the online evaluation phase, the corresponding record in the scene twin model is found based on the scene complexity index of the current time window, and the scene complexity index is mapped to the scene weight coefficient. Specifically, the time window number is used in the sequence of scene complexity indicators. The complexity index of the current time window is located in the model, and the road structure and traffic status information corresponding to the time window are read in the scene twin model. The scene complexity index is mapped to the scene weight coefficient. For example, a piecewise linear mapping method is used to assign a smaller weight to the low complexity interval, a medium weight to the medium complexity interval, and a larger weight to the high complexity interval.

[0046] For example, complexity index 1 is mapped to weight 0.5, complexity index 2 is mapped to weight 1, and complexity index 3 is mapped to weight 1.5, so as to reflect the impact of scene complexity when adjusting the weights of fuzzy rules in the vehicle layer.

[0047] S4.2. Adjust the weights of the hierarchical fuzzy rules related to the distance between vehicles, relative speed and lateral offset in the vehicle layer fuzzy inference model according to the scene weight coefficient to obtain the weights of the hierarchical fuzzy rules for the current time window. Specifically, during the offline phase, basic weights are set for each input variable, such as a weight of 0.4 for distance between vehicles, a weight of 0.4 for relative speed, and a weight of 0.2 for lateral offset. During the online phase, these basic weights are scaled according to the scene weight coefficient of the current time window. For example, when the scene weight coefficient is greater than 1, the weights of rule items related to distance between vehicles and relative speed are increased, making the system pay more attention to longitudinal and lateral safety distances in complex scenarios; when the scene weight coefficient is close to 0.5, the weight scaling is smaller, making the system maintain lower sensitivity in simple scenarios. Through this hierarchical rule weight adjustment mechanism, the importance of input variables can be dynamically changed according to the complexity of the scene in vehicle-layer fuzzy inference.

[0048] S4.3 Read the inter-vehicle distance, relative speed, lateral offset and scene risk level of the target vehicle in the current time window from the vehicle twin model, and input the inter-vehicle distance, relative speed, lateral offset and scene risk level into the fuzzy inference model corresponding to the adjusted hierarchical fuzzy rule weights to obtain the candidate operation safety level score of the target vehicle. Specifically, based on the input parameters of the current time window, its membership degree on different fuzzy sets is calculated. Combined with the fuzzy rule table after hierarchical weight adjustment, the output security level score of each rule is weighted and superimposed. In this embodiment, the candidate operation security level score is described in the following form: in, The current time window is used to score the operational safety level of the target vehicle candidate; M is the number of fuzzy rules at the vehicle layer; It is the product of the comprehensive membership degree of the j-th rule under the current input and the hierarchical weight factor; Let j be the security level representative value corresponding to the j-th rule. For example, assign 1, 2, 3 and 4 to level 1 security, level 2 security, level 3 security and level 4 security respectively. Through the above weighted summation, the outputs of multiple fuzzy rules are integrated into a continuous candidate security level score.

[0049] S4.4. The target vehicle's candidate operational safety level score and the operational safety level score of the previous time window are weighted by a preset smoothing coefficient to obtain the target vehicle's operational safety level result for the current time window, where: The candidate operational safety level of the target vehicle is scored based on a preset smoothing coefficient. The current time window's operational safety level score is obtained by weighting the score with the previous time window's operational safety level score. Furthermore, regarding candidate scores Safety level score for operation in the previous time window Perform weighted combination: in, The operational safety level score is determined by smoothing the current time window. The preset smoothing coefficient has a value between 0 and 1, for example, 0.6 to 0.8 can be selected to balance the ability to respond to new risk changes and the ability to suppress short-term fluctuations. Assess the safety level of candidates operating within the current time window; The safety level score for the previous time window; when Less than or equal to the first threshold When the target vehicle's operational safety level is set to Level 1 safety, the result will be determined accordingly. Greater than the first threshold And less than or equal to the second threshold When the target vehicle's operational safety level is set to Level 2 safety, the safety level of the target vehicle will be set accordingly. Greater than the second threshold And less than or equal to the third threshold When the target vehicle's operational safety level is set to Level 3, the safety level will be determined accordingly. Greater than the third threshold At that time, the target vehicle's operational safety level was set to Level 4; among which... ; For example, take .

[0050] For example, a fourth threshold is defined. For the threshold of sudden change in safety level, such as taking ,when Furthermore, when the safety level corresponding to the candidate score is significantly higher than the safety level of the previous time window, the target vehicle operation safety level result of the current time window is adjusted to a higher risk level adjacent to the original determined level, and the adjusted level is used as the operation safety level result of the current time window to enhance the level of vigilance against rapidly deteriorating scenarios.

[0051] Preferably, in the online evaluation stage, this step maps the scene complexity index to scene weight coefficients, adaptively adjusts the key input weights in the vehicle-layer fuzzy rules, and outputs the target vehicle's operational safety level by combining real-time status data from the vehicle twin model. This solves the technical problems of traditional safety level assessments, such as the lack of comprehensive consideration of scene changes and time series smoothing, and the difficulty in providing timely responses to sudden risks. Furthermore, by performing weighted smoothing and threshold grading on the candidate operational safety level scores and the scores of the previous time window, it suppresses short-term random fluctuations while maintaining sensitivity to the continuous upward trend of risks. This results in a more stable level output effect in vehicle operational safety level assessment that is closer to the actual driving experience, which is beneficial for continuous monitoring and graded early warning of the target vehicle's operational status in complex road environments.

[0052] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for assessing vehicle operational safety levels based on digital twins and fuzzy algorithms, characterized in that, include: Collect traffic flow, road structure, and vehicle operation status data within the road network, and construct scene twin models and vehicle twin models based on the collected data; Based on the scenario twin model, the scenario features of the number of intersections, lane occupancy rate and speed fluctuation are extracted, and the scenario complexity index is obtained through clustering and normalization. Based on the scene complexity index, a scene-layer fuzzy inference sub-model is constructed to output the scene risk level. Then, at the vehicle layer, the scene risk level is combined with the distance between vehicles, relative speed, and lateral offset to obtain the target vehicle candidate safety level. During the online evaluation phase, the weights of the hierarchical fuzzy rules are adaptively adjusted based on the scenario complexity index, and the target vehicle's operational safety level is output based on the real-time status of the vehicle twin model.

2. The vehicle operation safety level assessment method based on digital twin and fuzzy algorithm as described in claim 1, characterized in that, The construction of scene twin models and vehicle twin models based on the collected data includes: Within the road network, acquire road structure data including road centerlines, lane boundaries, speed limits, and signal control parameters, and generate the road network topology in a unified coordinate system; In the road network topology, traffic flow data and target vehicle operation status data of each road segment are collected at preset time intervals, and a unified timestamp and road network identifier are recorded for the traffic flow data and the vehicle operation status data. The road structure data, traffic flow data, and vehicle operating status data are aligned under a unified coordinate system and a unified time reference to generate a fusion dataset containing lane-level traffic status and vehicle pose. Based on the fused dataset, a scene twin model containing lane-level grid indexes and conflict area identifiers is constructed, and a corresponding vehicle twin model is generated for each target vehicle trajectory in the scene twin model.

3. The vehicle operation safety level assessment method based on digital twin and fuzzy algorithm as described in claim 2, characterized in that, Generating the scene twin model includes: Based on the road structure data, the centerline of each lane is discretely sampled, lane-level grid cells are divided in a unified coordinate system, and corresponding lane-level grid indices are generated. Based on the vehicle operation status data and the lane-level grid index, the distribution of vehicle operation trajectories and the relationship between entry and exit directions within each lane-level grid cell are statistically analyzed to obtain a set of candidate conflict areas. In the candidate conflict area set, a conflict intensity threshold is set based on the vehicle trajectory intersection angle, relative speed difference, and historical accident weight. Conflict area identifiers are added to lane-level grid cells whose conflict intensity exceeds the conflict intensity threshold. By associating the lane-level grid index and the conflict area identifier with the road structure data and traffic flow data, a scene twin model containing lane-level traffic status and conflict area identifier layers is constructed.

4. The vehicle operation safety level assessment method based on digital twin and fuzzy algorithm as described in claim 2, characterized in that, Generating the vehicle twin model includes: In the scene twin model, the pose time series of each target vehicle's running trajectory is extracted based on the fusion dataset, and the pose time series is matched with the corresponding lane-level grid index sequence. Based on the pose time series and the lane-level grid index sequence, the operating status of adjacent vehicles and the nearest conflict area identifier are associated under each timestamp to generate a local traffic state sequence corresponding to each target vehicle's operating trajectory. Based on the local traffic state sequence, a vehicle twin model state vector is constructed, incorporating pose parameters, lane-level grid index, relative distance, relative speed, and conflict zone identifier into the same state vector, and assigning a corresponding vehicle twin model identifier to each target vehicle trajectory. The vehicle twin model's state vector is kept synchronized with the lane-level mesh index and conflict area identifier in the scene twin model to obtain the vehicle twin model.

5. The vehicle operation safety level assessment method based on digital twin and fuzzy algorithm as described in claim 1 or 2, characterized in that, The scene complexity index is obtained by: Based on the lane-level grid index in the scene twin model, the number of grids belonging to the intersection, lane occupancy rate, and speed standard deviation are counted in each time window to generate a scene feature vector set containing the number of intersections, lane occupancy rate, and speed fluctuations. The number of intersections, lane occupancy rate, and speed fluctuation in the scene feature vector set are normalized without dimension, and weight factors are set according to the impact of the number of intersections and lane occupancy rate on safety. The normalized scene features are then weighted and combined to obtain a weighted scene feature vector set. A density-based clustering algorithm is used to cluster the weighted scene feature vector set. The scene complexity level is divided according to the number of intersections, lane occupancy rate and speed fluctuation range of each cluster center, and a corresponding scene complexity level label is assigned to each time window. The scene complexity index is generated based on the correspondence between the scene complexity level label and the weighted scene feature vector.

6. The vehicle operation safety level assessment method based on digital twin and fuzzy algorithm as described in claim 5, characterized in that, The process of assigning corresponding scene complexity level labels to each time window includes: The distance distribution between samples is calculated in the feature space formed by the number of intersections, lane occupancy rate and speed fluctuation based on the weighted scene feature vector set. The density radius threshold and minimum number of samples are selected according to the k-nearest neighbor distance curve. Then, a density-based clustering algorithm is used to divide the weighted scene feature vector set into clusters and noise point sets. For each cluster, the weighted scene feature vector is averaged on three components: number of intersections, lane occupancy rate, and speed fluctuation, to generate the corresponding cluster center feature vector. The cluster center feature vectors are weighted and superimposed to form a scene complexity score based on the number of intersections, lane occupancy rate, and the degree of impact of speed fluctuation on safety. Each cluster is then classified into multiple scene complexity levels from low to high according to the scene complexity score. For each time window sample in the weighted scene feature vector set, the cluster to which it belongs is determined according to the clustering results, and the corresponding scene complexity level is used as the scene complexity level label for that time window. For time window samples in the noise point set, the scene complexity level corresponding to the nearest cluster is selected based on the distance to the feature vector of each cluster center, and the selected scene complexity level is written into the scene complexity level label of the corresponding time window.

7. The vehicle operation safety level assessment method based on digital twin and fuzzy algorithm as described in claim 5, characterized in that, The candidate safety level of the target vehicle is obtained by: The input membership function of the scenario complexity index is set according to the numerical distribution of the scenario complexity index on the time axis, and the output membership function of the scenario risk level and the low, medium and high levels are defined at the scenario layer. Based on the correspondence between the scenario complexity index and historical accident samples, a scenario-layer fuzzy rule table is set, and the scenario complexity index is input into the scenario-layer fuzzy inference sub-model to obtain the scenario risk level corresponding to each time window. In the vehicle layer, the vehicle layer membership function is set with the distance between vehicles, the relative speed, the lateral offset, and the scene risk level as input variables, and the weight coefficients of each input variable in the vehicle layer fuzzy rule table are adjusted according to the scene risk level. The distance between vehicles, the relative speed, the lateral offset, and the scene risk level are input into the vehicle layer fuzzy inference model according to time windows, and the candidate safety level of the target vehicle in each time window is output according to the vehicle layer fuzzy rule table.

8. The vehicle operation safety level assessment method based on digital twin and fuzzy algorithm as described in claim 7, characterized in that, The risk level of the scenario is obtained, including: Based on the accident markers and corresponding scenario complexity indices for each time window in the historical accident samples, the frequency and severity of accidents within different scenario complexity index intervals are statistically analyzed to obtain the correlation between scenario complexity indices and accident risks. Based on the aforementioned correspondence, the membership interval of the scenario complexity index is correlated with the accident frequency and the accident severity level, defining low-risk, medium-risk, and high-risk scenario risk levels. In the scene-layer fuzzy inference sub-model, the scene complexity index corresponding to each time window is input into the scene-layer fuzzy rule table, the membership degree of each scene risk level is obtained according to the preset membership function, and the scene risk score of the time window is obtained by weighted average defuzzification. Based on the comparison between the scenario risk score and the preset classification threshold range, the scenario risk score is classified into low-risk, medium-risk, or high-risk scenario risk levels.

9. The vehicle operation safety level assessment method based on digital twin and fuzzy algorithm as described in claim 8, characterized in that, Also includes: Based on the historical scenario risk score distribution and combined with the accident frequency, the numerical range of the scenario risk score is segmented to determine the boundaries of the preset classification threshold ranges corresponding to low risk, medium risk and high risk. The scenario risk score for each time window is compared with the boundary of the preset classification threshold interval: When the scenario risk score falls into the low-risk classification threshold range, the time window is marked as a low-risk scenario risk level. When the scenario risk score falls within the medium-risk classification threshold range, the time window is marked as a medium-risk scenario risk level. When the scenario risk score falls into the high-risk classification threshold range, the time window is marked as a high-risk scenario risk level; at the dividing point, the scenario risk level of the side with the higher accident frequency is selected first according to the accident frequency of adjacent intervals.

10. The vehicle operation safety level assessment method based on digital twin and fuzzy algorithm as described in claim 7, characterized in that, The output target vehicle operating safety level result includes: During the online evaluation phase, the corresponding record in the scene twin model is found based on the scene complexity index of the current time window, and the scene complexity index is mapped to the scene weight coefficient. Based on the scene weight coefficients, the weights of the hierarchical fuzzy rules related to the distance between vehicles, relative speed, and lateral offset in the vehicle layer fuzzy inference model are proportionally adjusted to obtain the weights of the hierarchical fuzzy rules for the current time window. The vehicle twin model is used to read the inter-vehicle distance, relative speed, lateral offset, and scene risk level of the target vehicle in the current time window. The inter-vehicle distance, relative speed, lateral offset, and scene risk level are input into the fuzzy inference model corresponding to the adjusted hierarchical fuzzy rule weights to obtain the candidate operational safety level score of the target vehicle. The target vehicle's operational safety level is calculated by weighting its candidate operational safety level score with the operational safety level score from the previous time window using a preset smoothing coefficient. The result is then obtained for the current time window. The target vehicle's candidate operational safety level is scored based on a preset smoothing coefficient. The current time window's operational safety level score is obtained by weighting the score with the previous time window's operational safety level score. when Less than or equal to the first threshold When the target vehicle's operational safety level is set to Level 1 safety, the result will be determined accordingly. Greater than the first threshold And less than or equal to the second threshold When the target vehicle's operational safety level is set to Level 2 safety, the safety level of the target vehicle will be set accordingly. Greater than the second threshold And less than or equal to the third threshold When the target vehicle's operational safety level is set to Level 3, the safety level will be determined accordingly. Greater than the third threshold At that time, the target vehicle's operational safety level will be set to Level 4. when The difference between the safety level score and the previous time window is greater than the fourth threshold. If the target vehicle's operational safety level is determined, the result will be adjusted to a higher risk level adjacent to that level, and the adjusted level will be used as the target vehicle's operational safety level result for the current time window.

Images