BIM-based traffic engineering electronic sand table simulation and analysis system

Abstract

The invention relates to a BIM-based traffic engineering electronic sand table simulation and analysis system. The method comprises the following steps: fusing a satellite image and laser radar data through an initial state modeling module, constructing a three-dimensional road model through point cloud processing and image fusion, establishing a lane-level reference in combination with environmental data, and determining the initial state of a traffic flow through trajectory smoothing, behavior recognition and particle filtering. And the simulation risk identification module predicts a braking distance by using a neural network based on the initial state of the traffic flow, simulates the multi-scene traffic flow after adjusting the traffic capacity, and identifies the high-risk area distribution. And the risk quantification early warning module calculates a risk value through Monte Carlo simulation according to high-risk area distribution, extracts a key factor optimization model and triggers graded early warning. By adopting the system, the modeling precision and the risk prediction accuracy can be improved, quantitative support is provided for traffic management, and the management scientificity and the intelligent level are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of digital simulation, and particularly relates to a traffic engineering electronic sand table simulation and analysis system based on BIM. BACKGROUND

[0002] With the acceleration of urbanization and the increasing complexity of transportation infrastructure construction, traditional traffic engineering planning and design and operation monitoring means have been difficult to meet the needs of fine management and intelligent control of the traffic system. At present, although the building information model (BIM) technology has shown strong three-dimensional visualization and collaborative design capabilities in the field of architecture, there are still problems such as low data integration, insufficient dynamic interaction capability, etc. in the application of traffic engineering, which makes it difficult to realize real-time simulation and risk prediction of traffic flow operation state. At the same time, the existing traffic simulation systems generally have defects such as single data source, insufficient simulation accuracy in multiple scenes, lack of intelligent decision-making capability, etc., cannot effectively fuse multi-source data such as satellite images and laser radar point clouds, and are also difficult to build a high-precision digital model covering road geometric features, environmental state and traffic flow parameters. SUMMARY

[0003] Therefore, it is necessary to provide a traffic engineering electronic sand table simulation and analysis system based on BIM which can improve the restoration accuracy of road geometric features and environmental information and more accurately reflect the risk evolution in complex traffic scenarios.

[0004] In a first aspect, the application provides a traffic engineering electronic sand table simulation and analysis system based on BIM, comprising:

[0005] An initial state modeling module is configured to obtain satellite images and laser radar scanning data, obtain road geometric parameters through point cloud processing, image fusion and feature extraction, and construct a lane-level modeling benchmark in combination with environmental data; and is also configured to dynamically update the lane-level modeling benchmark by using a particle filtering algorithm, and determine the initial state of traffic flow by fusing real-time vehicle trajectories and lane-changing behavior information.

[0006] A simulation risk identification module is configured to predict the braking distance under low road friction based on the initial state of traffic flow by using a neural network model, and obtain an adjusted traffic capacity index; and is also configured to build a multi-scenario traffic flow simulation model by combining environmental data and the traffic capacity index to simulate multi-scenario traffic flow and determine the distribution of high-risk areas.

[0007] A risk quantification and early warning module is configured to calculate the risk quantification value of the distribution of high-risk areas by using a Monte Carlo simulation algorithm, update the weight matrix of the neural network model by extracting key interaction influencing factors, and trigger a traffic warning signal according to the risk level.

[0008] In one embodiment, satellite images and laser radar scanning data are acquired, road geometric parameters are obtained through point cloud processing, image fusion and feature extraction, and lane-level modeling benchmarks are constructed in combination with environmental data, including:

[0009] Satellite images and laser radar scanning data are acquired, and point cloud data is extracted from the laser radar scanning data through a laser radar point cloud analysis algorithm; the point cloud data includes three-dimensional spatial coordinates of road surface sampling points.

[0010] A point cloud processing algorithm combining statistical filtering and radius filtering is used to denoise the extracted point cloud data to obtain smooth point cloud data.

[0011] Using the three-dimensional spatial coordinates of the obtained smooth point cloud data as a framework, a SIFT feature matching-based image fusion algorithm is used to register and fuse the satellite images and the smooth point cloud data, road feature points are extracted to establish a coordinate mapping relationship, and a high-resolution three-dimensional road model is generated; the road feature points include lane line inflection points and curb end points.

[0012] Based on the generated three-dimensional road model, a geometric feature extraction algorithm is used to take equidistant cross sections at a predetermined interval along the road center line, measure the lane line spacing within the cross section to determine the lane width, fit the road center line curve to calculate the curvature, and obtain the road geometric parameters.

[0013] After data standardization, the environmental data is integrated into the three-dimensional road model together with the extracted road geometric parameters to obtain a lane-level modeling benchmark containing road geometric features and real-time environmental state; the environmental data includes meteorological information and traffic flow information.

[0014] In one embodiment, a particle filter algorithm is used to dynamically update the lane-level modeling benchmark, and real-time vehicle trajectory and lane changing behavior information are fused to determine the initial state of the traffic flow, including:

[0015] Real-time vehicle trajectory data is collected on the target road section, and a Kalman filter combined with a sliding window trajectory smoothing algorithm is used to denoise and correct the data to generate a continuous and smooth trajectory sequence; the real-time vehicle trajectory data includes vehicle timestamp, spatial coordinates and instantaneous speed.

[0016] Lane change intention features including lateral displacement rate of change and trajectory curvature sudden change value are extracted based on the smooth trajectory sequence, the features are input into a pre-trained behavior classification model to output vehicle behavior labels, and lane changing behavior information is obtained after screening; the behavior classification model is constructed based on a random forest algorithm.

[0017] The vehicle behavior labels are spatially associated with the road geometric parameters in the lane-level modeling benchmark, the number of vehicles per unit time in each lane is calculated through a density estimation algorithm, and the traffic flow density distribution is obtained by combining the lane size.

[0018] The lane-level modeling benchmark is adjusted based on the smooth trajectory sequence, lane-changing behavior information, and traffic flow density distribution, and the initial state of the traffic flow is obtained through weight normalization and particle state estimation; the initial state of the traffic flow includes lane-level traffic flow density, road section average speed, lane-level traffic flow, lane-level lane-changing frequency, vehicle following distance mean value, and road surface friction coefficient.

[0019] In one of the embodiments, the traffic flow density distribution is calculated by the following formula:

[0020]

[0021] p = {p1, p2,..., pN} L}

[0022] wherein p represents the traffic flow density distribution of the target road section, p l represents the traffic flow density of the lth lane, N l represents the average number of vehicles in the lth lane within a unit time T, L v represents the standard vehicle length, D represents the length of the target road section, and W represents the width of a single lane.

[0023] In one of the embodiments, the simulation risk identification module further comprises:

[0024] The risk area identification unit is configured to:

[0025] The real-time road surface friction coefficient and the road section average speed are extracted from the initial state of the traffic flow, input into the pre-trained neural network model, and the braking distance prediction value is calculated.

[0026] The road section benchmark traffic capacity is adjusted based on the braking distance prediction value to obtain the adjusted traffic capacity index.

[0027] The traffic capacity index is combined with the road geometric parameters and environmental data in the lane-level modeling benchmark to generate a model input parameter set containing normal, rainfall, and snow and ice scenarios.

[0028] The model input parameters are imported into a traffic simulation software to build a multi-scenario traffic flow simulation model, simulate the traffic flow running state under different scenarios, and obtain a running feature data set including vehicle delay, congestion duration, and conflict point distribution data.

[0029] Based on the preset risk judgment threshold, the high-risk data association area in the running feature data set is identified by a spatial clustering algorithm, and the high-risk area distribution represented by polygon boundary coordinates is determined.

[0030] The traffic flow optimization unit is configured to:

[0031] The risk level data including the average brake distance extension rate and the vehicle conflict probability are extracted from the high-risk area distribution, and the dynamic traffic control instructions including the speed limit, lane control, and warning prompt are generated according to preset rules.

[0032] The parameters of the traffic flow simulation model are adjusted according to the dynamic traffic control instructions, the model parameters after adjustment are combined with the road geometric constraints to redistribute the traffic flow of the surrounding road network associated with the high-risk area, and the total delay time, the number of high-risk areas, and the vehicle conflict probability before and after optimization are compared to verify the optimized high-risk area distribution; the road geometric constraints include the number of lanes, the single-lane width, the road curvature radius, and the road section design slope.

[0033] In one of the embodiments, the brake distance prediction value is calculated by the following formula:

[0034]

[0035] wherein, represents the brake distance prediction value, x j represents the input layer variable, j = 1, 2, corresponding to the real-time road surface friction coefficient μ and the average speed of the road section v, N i represents the number of input layer neurons, N i = 2, N h represents the number of hidden layer neurons, ω j,i represents the weight matrix from the input layer to the hidden layer, ω i,out represents the weight from the hidden layer to the output layer, b i , b out represents the bias vector of the hidden layer and the output layer, σ hid represents the hidden layer activation function, α represents the hyperparameter, σ out represents the output layer activation function, adopts the linear activation function to ensure that the output is continuous σ out (z) = z.

[0036] In one of the embodiments, the Monte Carlo simulation algorithm is used to calculate the risk quantization value of the high-risk area distribution, the key interaction influencing factors are extracted to update the weight matrix of the neural network model, and the traffic warning signal is triggered according to the risk level, including:

[0037] The space-time distribution features are extracted from the traffic flow data of the high-risk area distribution; the space-time distribution features include the traffic flow and the speed fluctuation.

[0038] The Monte Carlo simulation algorithm is used to calculate the risk quantization value of the space-time distribution features to obtain the risk level quantization result.

[0039] The key interaction factors influencing the risk are extracted from the mutual information analysis result of the risk level quantization result; the key interaction factors include the intersection congestion degree and the signal light switching frequency.

[0040] The neural network model parameters are updated by back propagation according to the key interaction factors, and an optimized weight matrix is obtained.

[0041] If the weight matrix satisfies a preset convergence condition, a traffic warning signal of a corresponding level is triggered according to the risk level.

[0042] In one of the embodiments, the risk quantization value of a single simulation of the Monte Carlo simulation algorithm is calculated by the following formula:

[0043]

[0044] wherein R represents the risk quantization value, the value range of which is [0, 1], q represents the current traffic flow, q max represents the historical maximum traffic flow of the target road section, represents the speed fluctuation standard deviation, represents the historical observed maximum speed fluctuation standard deviation, and ω1 and ω2 represent weight coefficients, ω1+ω2=1.

[0045] In a second aspect, the present application further provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor realizes the steps of the foregoing system when executing the computer program.

[0046] In a third aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the steps of the foregoing system.

[0047] The BIM-based traffic engineering electronic sand table simulation and analysis system, the computer device and the storage medium, the initial state modeling module first acquires satellite images and laser radar scanning data, performs point cloud analysis, statistical filtering and radius filtering denoising, fuses the images and the point cloud data by using a SIFT feature matching algorithm, constructs a high-resolution three-dimensional road model, and extracts geometric parameters such as lane width and curvature; then, in combination with environmental data such as weather and traffic flow, a lane-level modeling benchmark is established. On this basis, real-time vehicle trajectory data is smoothed by Kalman filtering, lane-changing behavior is identified by using a random forest algorithm, traffic flow density distribution is obtained by using a density estimation algorithm, and finally, the particle filtering algorithm is used to dynamically update the modeling benchmark to determine the initial state of the traffic flow including parameters such as traffic flow density, vehicle speed and lane-changing frequency. The simulation risk identification module inputs the road surface friction coefficient and the vehicle speed in the initial state of the traffic flow into a pre-trained neural network, predicts the braking distance and adjusts the road section capacity index; then, in combination with the environmental data and the capacity, a model input parameter set of normal, rainfall, snow and other scenes is generated, which is imported into a traffic simulation software to simulate the operation of the traffic flow, and data such as vehicle delay and conflict point distribution are obtained, and the high-risk area distribution is identified by using a spatial clustering algorithm. The risk quantification early warning module extracts the space-time characteristics such as traffic flow and speed fluctuation in the high-risk area, calculates the risk quantification value by using Monte Carlo simulation, analyzes and screens key influence factors such as intersection congestion degree and signal light switching frequency by using mutual information, and updates the neural network weight matrix by using the back propagation algorithm; when the weight matrix meets the convergence condition, the hierarchical early warning signal is triggered according to the risk level. The system constructs a lane-level modeling benchmark by fusing BIM and multi-source data, which improves the restoration accuracy of road geometric features and environmental information compared with two-dimensional drawings or coarse-grained models. Secondly, intelligent algorithms such as particle filtering and neural networks are used to realize dynamic deduction of the traffic flow state and risk quantification prediction, which can more accurately reflect the risk evolution in complex traffic scenarios compared with single empirical formula or static model. The system provides quantitative support for traffic planning and design, real-time management and control, and emergency decision-making, and effectively improves the scientificity and intelligent level of traffic engineering management. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the drawings needed to be used in the embodiment or related art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creating any inventive labor.

[0049] Figure 1 The structural block diagram of the BIM-based traffic engineering electronic sand table simulation and analysis system provided by the embodiments of the present application is shown in the figure.

[0050] Figure 2The flowchart provided by the embodiment of the application dynamically updates the lane-level modeling benchmark by using a particle filtering algorithm, and fuses real-time vehicle trajectory and lane-changing behavior information to determine the initial state of the traffic flow. DETAILED DESCRIPTION

[0051] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0052] In one of the embodiments, as shown in Figure 1 The BIM-based traffic engineering electronic sand table simulation and analysis system provided by the present application can include:

[0053] The initial state modeling module 101 is configured to acquire satellite images and laser radar scanning data, obtain road geometric parameters through point cloud processing, image fusion and feature extraction, and construct a lane-level modeling benchmark in combination with environmental data; and is further configured to dynamically update the lane-level modeling benchmark by using a particle filtering algorithm, and fuse real-time vehicle trajectory and lane-changing behavior information to determine the initial state of the traffic flow.

[0054] Specifically, the satellite images and laser radar scanning data are acquired, the point cloud is parsed from the laser radar data, the noise points are removed through statistical filtering and radius filtering, the satellite images and the point cloud are fused based on a SIFT feature matching algorithm, the lane lines, kerbs and other features are extracted to construct a three-dimensional road model, and the geometric parameters such as the lane width and the road curvature are measured and acquired. Secondly, the environmental data such as the weather and the traffic flow are standardized and integrated with the road geometric parameters to construct the lane-level modeling benchmark. Finally, the vehicle trajectory data are collected, smoothed through Kalman filtering, the lane-changing behavior is identified by using a random forest model, the vehicle density on the lane is counted, the particle filtering algorithm is used to dynamically adjust the modeling benchmark, and the initial state of the traffic flow including the vehicle flow density, the vehicle speed and the lane-changing frequency is determined.

[0055] The simulation risk identification module 102 is configured to predict the braking distance under the condition of low road friction by using a neural network model based on the initial state of the traffic flow, acquire the adjusted traffic capacity index, and further configured to build a multi-scenario traffic flow simulation model in combination with the environmental data and the traffic capacity index to simulate multi-scenario traffic flow and determine the distribution of high-risk areas.

[0056] The road friction coefficient and vehicle speed are extracted from the initial state of traffic flow, input into the pre-trained BP neural network model, the braking distance prediction value is calculated, and the road capacity index is adjusted based on this. Then, the road capacity index is combined with road geometric parameters and environmental data to generate model input parameters for normal, rainfall, ice and snow, and other scenarios. Next, the traffic simulation software is used to import the parameters to simulate traffic flow operation under different scenarios and obtain vehicle delay, congestion duration, and other data. Finally, the risk judgment threshold is set, the high-risk data area is identified by the DBSCAN spatial clustering algorithm, and the distribution of high-risk areas is determined.

[0057] The risk quantification early warning module 103 is used to calculate the risk quantification value of the high-risk area distribution using the Monte Carlo simulation algorithm, extract the key interaction influencing factors to update the weight matrix of the neural network model, and trigger the traffic warning signal according to the risk level.

[0058] For the traffic flow data of the high-risk area distribution, the space-time characteristics such as traffic volume and speed fluctuation are extracted, the Monte Carlo simulation algorithm is used for 1000 times of iteration calculation, the risk quantification value is obtained and divided into 5 risk levels. Through mutual information analysis, key interaction factors such as intersection congestion degree and signal light switching frequency are selected, and the neural network model parameters are updated using the back propagation algorithm. When the weight matrix gradient norm is less than 0.01, the convergence condition is met, and the hierarchical warning is triggered according to the risk level: no warning for level 1-2, yellow warning for level 3, speed limit of 60 km / h for level 4, and closure of part of the lane and red warning for level 5.

[0059] The BIM-based traffic engineering electronic sand table simulation and analysis system, the initial state modeling module first acquires satellite images and laser radar scanning data, denoises through point cloud analysis, statistical filtering and radius filtering, fuses the images and point cloud data by using a SIFT feature matching algorithm, constructs a high-resolution three-dimensional road model, and extracts geometric parameters such as lane width and curvature; then, in combination with environmental data such as weather and traffic flow, a lane-level modeling benchmark is established. On this basis, real-time vehicle trajectory data is smoothed through Kalman filtering, lane-changing behavior is identified using a random forest algorithm, traffic flow density distribution is obtained using a density estimation algorithm, and finally, particle filtering algorithm is used to dynamically update the modeling benchmark to determine the initial state of the traffic flow including parameters such as traffic flow density, vehicle speed and lane-changing frequency. The simulation risk identification module inputs the road surface friction coefficient and vehicle speed in the traffic flow initial state into the pre-trained neural network, predicts the braking distance and adjusts the road section capacity index; then, in combination with environmental data and capacity, a model input parameter set for normal, rainfall, ice and snow scenes is generated, which is imported into a traffic simulation software to simulate traffic flow operation and obtain data such as vehicle delay and conflict point distribution, and the spatial clustering algorithm is used to identify the distribution of high-risk areas. The risk quantification early warning module extracts the space-time characteristics such as traffic flow and speed fluctuation in the high-risk area, calculates the risk quantification value through Monte Carlo simulation, filters key influence factors such as intersection congestion degree and signal light switching frequency using mutual information analysis, and updates the neural network weight matrix using the back propagation algorithm; when the weight matrix meets the convergence condition, the hierarchical early warning signal is triggered according to the risk level. The system builds a lane-level modeling benchmark through BIM and multi-source data fusion, which improves the restoration accuracy of road geometric features and environmental information compared with two-dimensional drawings or coarse-grained models. Secondly, intelligent algorithms such as particle filtering and neural networks are used to realize dynamic deduction of traffic flow state and risk quantification prediction, which can more accurately reflect the risk evolution in complex traffic scenarios compared with single empirical formula or static model. It provides quantitative support for traffic planning and design, real-time control and emergency decision-making, and effectively improves the scientificity and intelligence level of traffic engineering management.

[0060] In one of the embodiments, acquiring satellite images and laser radar scanning data, processing the point cloud, fusing the images and extracting the road geometric parameters, and combining the environmental data to construct the lane-level modeling benchmark can include the following steps:

[0061] In step S101, satellite images and laser radar scanning data are acquired, and point cloud data is extracted from the laser radar scanning data through a laser radar point cloud analysis algorithm; the point cloud data contains three-dimensional space coordinates of road surface sampling points.

[0062] In step S102, a point cloud processing algorithm combining statistical filtering and radius filtering is used to denoise the extracted point cloud data to obtain smoothed point cloud data.

[0063] Step S103, taking the three-dimensional space coordinates of the obtained smoothed point cloud data as the framework, a satellite image is fused with the smoothed point cloud data by using a SIFT feature matching-based image fusion algorithm, road feature points are extracted to establish a coordinate mapping relationship, and a high-resolution three-dimensional road model is generated; the road feature points include lane line inflection points and kerb end points.

[0064] Step S104, based on the generated three-dimensional road model, a geometric feature extraction algorithm is used to take equidistant cross sections at a preset interval along the road center line, measure the lane line spacing in the cross section to determine the lane width, fit the road center line curve to calculate the curvature, and obtain the road geometric parameters.

[0065] Step S105, environment data is obtained, which is integrated into the three-dimensional road model after data standardization processing, to obtain a lane-level modeling reference containing road geometric features and real-time environmental state; the environment data includes meteorological information and traffic flow information.

[0066] Specifically, the system accesses satellite images and laser radar scanning data, analyzes the laser radar scanning data by using a laser radar point cloud analysis algorithm, extracts point cloud data containing three-dimensional space coordinates (X, Y, Z axis coordinate values) of road surface sampling points, wherein the three-dimensional space coordinates are based on the geodetic coordinate system to ensure the spatial uniformity of the data. A point cloud processing algorithm combining statistical filtering and radius filtering is used to denoise the extracted point cloud data: statistical filtering calculates the distance mean and standard deviation of each point and a predetermined number (such as 50) of points in its neighborhood, and removes noise points whose distance mean exceeds 3 times the standard deviation; radius filtering deletes isolated points whose number of neighborhood points within a fixed search radius (such as 0.3 meters) is less than a predetermined threshold (such as 5), and obtains smoothed point cloud data after two-step processing.

[0067] Secondly, taking the three-dimensional space coordinates of the smoothed point cloud data as the spatial framework, an image fusion algorithm based on SIFT feature matching is adopted: SIFT feature points are extracted from the satellite image and the smoothed point cloud respectively, and the Euclidean distance between the feature points is calculated for preliminary matching, and then the RANSAC algorithm is used to remove the mis-matching points (the number of iterations is set to 200), so as to realize the registration and fusion of the satellite image and the smoothed point cloud data; road feature points such as lane line inflection points and curb end points are extracted, the coordinate mapping relationship of the feature points in the image and the point cloud is established, and a high-resolution three-dimensional road model is generated. Based on the generated three-dimensional road model, a geometric feature extraction algorithm is adopted: equidistant sections are intercepted along the road center line at a preset interval (such as 10 meters), and each section is perpendicular to the road center line; the distance between adjacent lane lines in the section is measured, and the average value of the measurement values of multiple sections is taken as the lane width; a cubic polynomial is used to fit the road center line curve, and the ratio of the second derivative to the square of the first derivative of the curve is calculated to obtain the road curvature value, and the lane width and the curvature are integrated to form the road geometric parameters. Environmental data are acquired, in which the meteorological information includes rainfall, visibility, and road surface temperature, and the traffic flow information includes the total number of vehicles and the vehicle type distribution within a unit time (such as 15 minutes); the environmental data are standardized, the meteorological information is converted into dimensionless index (such as 0-50 mm of rainfall corresponding to index 0-1), and the traffic flow information is converted into saturation (ratio of actual flow to design flow); the standardized environmental data and the extracted road geometric parameters are integrated into the three-dimensional road model to form a lane-level modeling benchmark containing road geometric features and real-time environmental state.

[0068] In this embodiment, through multi-source data fusion and standardization processing, high-precision digital modeling of road scene is realized. Compared with a single data source, the combination of laser radar point cloud and satellite image can more comprehensively capture road details; the combination of statistical filtering and radius filtering effectively improves the quality of point cloud data and reduces noise interference; SIFT feature matching and RANSAC algorithm ensure the registration accuracy of the image and the point cloud, and avoid spatial position deviation. At the same time, dynamic environmental data such as meteorology and traffic flow are integrated into the three-dimensional model, so that the modeling benchmark can reflect the actual road operating conditions. This method provides accurate and complete basic data for traffic flow simulation, risk identification and other applications, which helps to improve the scientificity and reliability of traffic engineering analysis, and has significant advantages in data precision, environmental adaptability and model integrity compared with traditional modeling methods.

[0069] In one of the embodiments, as shown in Figure 2 the particle filter algorithm is used to dynamically update the lane-level modeling benchmark, and the real-time vehicle trajectory and lane changing behavior information are fused to determine the initial state of the traffic flow, which can include the following steps:

[0070] Step S201, collect real-time vehicle trajectory data of the target section, use Kalman filtering combined with sliding window trajectory smoothing algorithm to denoise and correct the data, and generate a continuous and smooth trajectory sequence; the real-time vehicle trajectory data includes vehicle timestamp, spatial coordinates and instantaneous speed.

[0071] Step S202, based on the smooth trajectory sequence, extract the lane change intention features including lateral displacement rate of change and trajectory curvature mutation value, input the features into the pre-trained behavior classification model to output vehicle behavior labels, and filter to obtain lane change behavior information; the behavior classification model is constructed based on random forest algorithm.

[0072] Step S203, spatially correlate the vehicle behavior labels with the road geometry parameters in the lane-level modeling benchmark, and statistically calculate the number of vehicles per unit time in each lane by density estimation algorithm, and calculate the traffic flow density distribution combined with the lane size.

[0073] Step S204, based on the smooth trajectory sequence, lane change behavior information and traffic flow density distribution, adjust the lane-level modeling benchmark using particle filtering algorithm, and obtain the initial state of traffic flow through weight normalization and particle state estimation; the initial state of traffic flow includes lane-level traffic density, average speed of the section, lane-level traffic volume, lane-level lane change frequency, average vehicle following distance and road friction coefficient.

[0074] Specifically, through the roadside sensor array (including laser radar and high-definition camera) integrated by the electronic sand table, vehicle trajectory data of the target section is collected at a frequency of 10Hz, including timestamp, latitude and longitude coordinates and instantaneous speed. The data is connected to the sand table data center in real time, and Kalman filtering (process noise covariance 0.01) and 5-second sliding window smoothing algorithm are used to eliminate GPS drift and measurement error, and a continuous trajectory sequence aligned with the sand table three-dimensional coordinate system is generated, ensuring that the spatial position accuracy of the trajectory point in the sand table is ≤0.5 meters. The lateral displacement rate of change (threshold 0.8m / s) and the trajectory curvature mutation value (≥0.1 / m) are extracted from the smooth trajectory, and input into the pre-trained random forest classification model (50 trees, feature dimension 5) of the electronic sand table to label the lane change behavior in real time. The sand table visualization interface dynamically marks the lane change section with yellow arrows, synchronously generates a lane change time-space heat map, and superimposes it on the lane-level modeling benchmark, providing behavior layer data support for traffic flow analysis.

[0075] The vehicle behavior labels are spatially correlated with the lane-level model (including lane lines and curb three-dimensional coordinates) in the electronic sand table, and the number of vehicles in each lane is counted in a 15-minute time window. The traffic flow density is calculated by the built-in algorithm of the sand table according to the formula The result is displayed as a gradient color bar above each lane, with red representing high-density congestion and green representing free flow, realizing the sand table visualization of density distribution.

[0076] The trajectory data, lane changing information, and density distribution are input into the particle filter simulation engine (1000 particles per lane) of the electronic sand table to iteratively optimize the lane-level modeling benchmark. The particle state includes parameters such as lane line offset and curvature correction. After 100 resampling, the road geometry model in the sand table is updated (e.g., correcting lane line offset caused by settlement). Finally, the lane-level traffic density, average speed (built-in speed statistics device in the sand table), and lane changing frequency are directly extracted from the sand table to support subsequent risk simulation analysis and ensure real-time linkage update of data and sand table model.

[0077] The trajectory data is accurately aligned with the sand table coordinate system after filtering and smoothing, ensuring spatial consistency. The lane changing behavior is intuitively presented with dynamic arrows and heat maps, improving the readability of behavior analysis. The traffic density is visualized in real time through color gradient, facilitating quick identification of congestion areas. The particle filter engine drives the dynamic update of the sand table model, ensuring real-time linkage of road geometry parameters and traffic flow state. This reduces data conversion steps, improves modeling accuracy and analysis efficiency, and enhances data interpretation ability through visual interaction, providing a high-precision and perceptual digital foundation for subsequent risk simulation and decision-making.

[0078] In one embodiment, the traffic density distribution can be calculated by the following formula:

[0079]

[0080] ρ = { ρ1, ρ2,..., ρN} L}

[0081] where ρ represents the traffic density distribution of the target road section, ρ l represents the traffic density of the lth lane, N l represents the average number of vehicles in the lth lane within a unit time T, L v represents the standard vehicle length, D represents the length of the target road section, and W represents the width of a single lane.

[0082] This embodiment quantifies the traffic density of different lanes accurately in units of lanes, combined with parameters such as standard vehicle length, road section length, and single lane width, avoiding the problem of ignoring lane differences in traditional overall density calculation, ensuring that the density data matches the spatial granularity of the lane-level BIM model of the electronic sand table. At the same time, it directly interfaces with real-time traffic data collected by the system to quickly generate dynamically updated density distribution results, presenting the congestion status of each lane in the electronic sand table in a visual form, providing accurate quantitative basis for subsequent traffic flow simulation and high-risk area identification, effectively improving the system's fine-grained perception and analysis ability of traffic running status, and supporting the scientificity and pertinence of traffic control and decision-making.

[0083] In one of the embodiments, the simulation risk identification module 102 can further include:

[0084] The risk area identification unit 1021 is configured to:

[0085] Step S301, extract the real-time road friction coefficient and the average vehicle speed of the road section from the initial state of the traffic flow, input the pre-trained neural network model, and calculate the braking distance prediction value.

[0086] Step S302, adjust the road section reference capacity based on the braking distance prediction value using a correction formula to obtain an adjusted capacity index.

[0087] Step S303, combine the capacity index with the road geometry parameters and environmental data in the lane-level modeling reference to generate a model input parameter set containing normal, rainfall, and snow scenarios.

[0088] Step S304, import the model input parameters into a traffic simulation software to build a multi-scenario traffic flow simulation model, simulate the traffic flow operating state under different scenarios, and obtain an operating feature data set including vehicle delay, congestion duration, and conflict point distribution data.

[0089] Step S305, based on a preset risk determination threshold, identify the high-risk data association area in the operating feature data set by a spatial clustering algorithm, and determine the high-risk area distribution represented by polygon boundary coordinates.

[0090] The traffic flow optimization unit 1022 is configured to:

[0091] Step S306, extract risk level data including average braking distance extension rate and vehicle conflict probability from the high-risk area distribution, and generate dynamic traffic control instructions including speed limit, lane control, and warning prompt according to a preset rule.

[0092] Step S307, adjust the parameters of the traffic flow simulation model according to the dynamic traffic control instructions, combine the adjusted model parameters with road geometry constraints to redistribute the traffic flow of the surrounding road network associated with the high-risk area, compare the total delay time, the number of high-risk areas, and the vehicle conflict probability before and after optimization, and verify the optimized high-risk area distribution; the road geometry constraints include the number of lanes, the width of single lane, the radius of road curvature, and the design slope of the road section.

[0093] Specifically, the real-time road surface friction coefficient (such as ice and snow road μ = 0.2) and the average vehicle speed (unit: km / h) are extracted from the initial state of traffic flow, and input into the pre-trained BP neural network model (three-layer structure, 2 neurons in the input layer, 8 neurons in the hidden layer, and the activation function is Leaky ReLU, α = 0.01). The predicted braking distance is output by matrix operation and activation function calculation. Based on the predicted value, the road section reference capacity is adjusted by using the correction formula (such as wherein C0 is the designed capacity, S pred is the predicted braking distance, S0 is the standard braking distance) to obtain the corrected capacity index. The index is fused with the road geometry parameters (lane width, curvature radius) and environmental data (rainfall, temperature) in the lane-level modeling reference to generate the model input parameter set of normal, rainfall, and ice and snow scenarios. The parameters are imported into the traffic simulation software (such as VISSIM 11.0) to run the simulation model to simulate traffic flow under different scenarios and obtain running characteristic data such as vehicle delay time, congestion duration, and conflict point coordinates. The risk judgment threshold (such as delay time ≥ 120 seconds, conflict point density ≥ 5 per 100㎡) is set, and the DBSCAN spatial clustering algorithm (neighborhood radius 50 meters, minimum point number 5) is used for clustering analysis of the running characteristic data to determine the distribution of high-risk areas represented by polygon boundary coordinates.

[0094] From the identified high-risk area distribution, the average braking distance extension rate (actual braking distance to standard value ratio), vehicle conflict probability (number of conflicts per unit time / vehicle number), and other risk level data are extracted, and dynamic traffic control instructions are generated according to the preset rules (such as high conflict probability triggering speed limit), including speed limit (such as 60 km / h), lane control (closing the inner lane), warning prompt (turning on the electronic warning sign), etc. The instructions are converted into traffic flow simulation model parameters (such as adjusting the upper limit of vehicle speed, lane access right), combined with the road geometry constraint conditions (number of lanes, single lane width, curvature radius, slope), and the traffic flow of the high-risk area surrounding road network is simulated again. By comparing the total delay time, the number of high-risk areas, and the vehicle conflict probability before and after optimization, the effectiveness of the adjustment scheme is verified, and the optimized high-risk area distribution is finally determined to provide quantitative decision basis for traffic management.

[0095] The embodiment is based on the initial state of traffic flow, uses neural network to predict braking distance and dynamically corrects traffic capacity, constructs a multi-scenario simulation model combining road geometry and environmental data, simulates the running state of traffic flow under different conditions, and obtains key data such as vehicle delay and conflict point distribution; through spatial clustering algorithm and preset threshold, quickly positioning high-risk areas to avoid the subjectivity of manual identification. In the optimization link, the system generates dynamic control instructions according to the risk level, redistributes traffic flow through simulation simulation combined with road geometric constraints, and verifies the optimization effect by indexes such as total delay time and conflict probability, forming a closed-loop system of "risk identification-strategy generation-simulation verification", which significantly improves the scientificity of traffic risk control and the efficiency of road network operation compared with traditional methods.

[0096] In one embodiment, the braking distance prediction value can be calculated by the following formula:

[0097]

[0098] wherein, represents the braking distance prediction value, x j represents the input layer variable, j = 1, 2, corresponding to the real-time road surface friction coefficient μ and the average speed v of the section, N i represents the number of input layer neurons, N i = 2, N h represents the number of hidden layer neurons, ω j,i represents the weight matrix from the input layer to the hidden layer, ω i,out represents the weight from the hidden layer to the output layer, b i , b out represents the bias vector of the hidden layer and the output layer, σ hid represents the hidden layer activation function, α represents the hyperparameter, σ out represents the output layer activation function, adopts a linear activation function to ensure that the output is a continuous value σ out (z) = z.

[0099] The embodiment can mine the potential relationship between variables through multi-layer nonlinear transformation, and is more suitable for complex road conditions such as ice and snow, rain, etc. The output layer adopts a linear activation function to ensure that the prediction value is a continuous variable, effectively avoiding discretization errors. It can dynamically correct the road traffic capacity and provide key parameter support for traffic flow simulation and risk area identification, significantly improving the accuracy of traffic risk prediction and the scientificity of traffic control decision-making, and reducing the risk of traffic accidents caused by braking distance estimation deviation.

[0100] In one of the embodiments, the Monte Carlo simulation algorithm is used to calculate the risk quantization value of the high-risk area distribution, extract the key interaction factors to update the weight matrix of the neural network model, and trigger the traffic warning signal according to the risk level, which can include the following steps:

[0101] Step S401, extracting the space-time distribution characteristics from the traffic flow data of the high-risk area distribution; the space-time distribution characteristics include the traffic flow and the speed fluctuation.

[0102] Step S402, using the Monte Carlo simulation algorithm to calculate the risk quantization value of the space-time distribution characteristics, and obtaining the risk level quantization result.

[0103] Step S403, extracting the key interaction factors affecting the risk through mutual information analysis on the risk level quantization result; the key interaction factors include the intersection congestion degree and the signal light switching frequency.

[0104] Step S404, updating the neural network model parameters according to the key interaction factors to obtain the optimized weight matrix.

[0105] Step S405, if the weight matrix meets the preset convergence condition, triggering the corresponding level of traffic warning signal according to the risk level.

[0106] Firstly, from the identified high-risk area traffic flow data, the space-time distribution characteristics such as traffic flow (number of vehicles passing per unit time) and speed fluctuation (standard deviation) are extracted to provide basic data for risk quantization. Secondly, the Monte Carlo simulation algorithm is used to calculate the characteristic data for more than 1000 iterations, construct a probability distribution model and output the risk quantization value, divide it into 5 risk levels according to the quantile, and realize the numerical expression of the risk degree. Next, the mutual information analysis method is used to select the key interaction factors such as intersection congestion degree (actual flow / design flow) and signal light switching frequency with 0.3 bits as the threshold, and the core variables affecting the risk are determined. Then, the key factors are input into the neural network model, and the back propagation algorithm (learning rate 0.001, loss function MSE) is used to update the model parameters, and the prediction logic is optimized by adjusting the weight matrix from the input layer to the hidden layer and from the hidden layer to the output layer. Finally, when the weight matrix gradient norm ≤0.01 meets the convergence condition, the system automatically triggers the corresponding warning signal according to the risk level, such as no warning for level 1-2, yellow warning for level 3, speed limit of 60km / h for level 4, and closure of part of the lane for level 5.

[0107] The embodiment converts complex traffic risks into quantifiable and interpretable indicators through Monte Carlo simulation and mutual information analysis, which is more objective than traditional experience judgment. The dynamic optimization of neural networks by back propagation algorithm enables the model to adapt to the changes of risk characteristics in different periods and under different road conditions, avoiding the limitations of "one-size-fits-all" early warning. In the electronic sand table visualization interface, the risk level and the changes of key factors can be presented in real time to help managers quickly locate the risk source. The hierarchical early warning mechanism realizes the precise allocation of traffic control resources, such as adjusting the signal timing of high-risk intersections first. It provides data-driven scientific decision support for traffic engineering planning and real-time scheduling, effectively reduces the incidence of traffic accidents, and improves the efficiency of road network operation.

[0108] In one embodiment, the risk quantification value of a single simulation of the Monte Carlo simulation algorithm can be calculated by the following formula:

[0109]

[0110] where R represents the risk quantification value, with a value range of [0, 1], q represents the current traffic flow, q max represents the historical maximum traffic flow of the target road section, represents the speed fluctuation standard deviation, represents the maximum speed fluctuation standard deviation observed in history, and ω1 and ω2 represent weight coefficients, with ω1 + ω2 = 1.

[0111] The embodiment quantifies two core indicators, traffic flow and speed fluctuation, providing a scientific and objective numerical basis for traffic risk assessment. The formula takes the ratio of current traffic flow to historical maximum traffic flow and the ratio of speed fluctuation standard deviation to historical maximum value as the basis variables, and combines weight coefficients for weighted calculation, which can comprehensively reflect the real-time pressure and instability of traffic flow. Compared with traditional qualitative assessment methods, the formula accurately quantifies the risk differences in different periods and under different road conditions through specific numerical comparison, making it easy to directly match with risk level standards. In the electronic sand table system, the risk quantification value calculated based on the formula can be intuitively presented in the three-dimensional scene, helping managers quickly locate high-risk areas. At the same time, its results as key parameters drive the system to automatically trigger corresponding levels of traffic management measures, realizing a closed-loop management from risk identification to response, effectively improving the accuracy and efficiency of traffic risk management.

[0112] It should be understood that, although the steps in the flowcharts related to the embodiments described above are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other sequences. Moreover, at least some of the steps in the flowcharts related to the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of the steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or stages.

[0113] In an embodiment, a computer device is provided, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the BIM-based traffic engineering electronic sand table simulation and analysis system as described above when executing the computer program.

[0114] In an embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps of the above method embodiments.

[0115] For the device embodiment, since it basically corresponds to the method embodiment, the relevant part is described in the part of the method embodiment. The device embodiments described above are only schematic, and the components described as separate components can or can not be physically separate, and the components displayed as a unit can or can not be a physical unit, i.e. can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the present disclosure according to the actual needs. Those skilled in the art can understand and implement it without creative labor.

[0116] The above-described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the application. It should be pointed out that, for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A BIM-based traffic engineering electronic sand table simulation and analysis system, characterized in that, The system comprises: An initial state modeling module is configured to acquire satellite images and laser radar scanning data, obtain road geometric parameters through point cloud processing, image fusion and feature extraction, and construct a lane-level modeling benchmark in combination with environmental data; and is further configured to dynamically update the lane-level modeling benchmark by using a particle filtering algorithm, fuse real-time vehicle trajectories and lane-changing behavior information to determine the initial state of traffic flow; A simulation risk identification module is configured to predict braking distances under low road friction conditions by using a neural network model based on the initial state of traffic flow, and obtain adjusted traffic capacity indicators; and is further configured to build a multi-scenario traffic flow simulation model in combination with the environmental data and traffic capacity indicators to simulate multi-scenario traffic flow and determine the distribution of high-risk areas; A risk quantification early warning module is configured to calculate risk quantification values for the distribution of high-risk areas by using a Monte Carlo simulation algorithm, extract key interaction influencing factors to update the weight matrix of the neural network model, and trigger traffic warning signals according to risk levels.

2. The system of claim 1, wherein, The acquisition of satellite images and laser radar scanning data, the obtaining of road geometric parameters through point cloud processing, image fusion and feature extraction, and the construction of a lane-level modeling benchmark in combination with environmental data comprises: Satellite images and laser radar scanning data are acquired, and point cloud data is extracted from the laser radar scanning data by using a laser radar point cloud analysis algorithm; the point cloud data contains three-dimensional spatial coordinates of road surface sampling points; A point cloud processing algorithm combining statistical filtering and radius filtering is used to denoise the extracted point cloud data to obtain smoothed point cloud data; The three-dimensional spatial coordinates of the obtained smoothed point cloud data are used as a framework, an image fusion algorithm based on SIFT feature matching is used to register and fuse the satellite images and the smoothed point cloud data, road feature points are extracted to establish a coordinate mapping relationship, and a high-resolution three-dimensional road model is generated; the road feature points include lane line inflection points and curb end points; Based on the generated three-dimensional road model, a geometric feature extraction algorithm is used to take equidistant cross sections at a preset interval along the road center line, measure the lane line spacing within the cross section to determine the lane width, fit the road center line curve to calculate the curvature, and obtain the road geometric parameters; After data standardization processing, the environmental data is integrated into the three-dimensional road model together with the extracted road geometric parameters to obtain a lane-level modeling benchmark containing road geometric features and real-time environmental states; the environmental data includes meteorological information and traffic flow information.

3. The system of claim 1, wherein, The dynamic updating of the lane-level modeling benchmark by using a particle filtering algorithm and the determination of the initial state of traffic flow by fusing real-time vehicle trajectories and lane-changing behavior information comprises: Real-time vehicle trajectory data of a target road section is collected, a Kalman filtering combined with a sliding window trajectory smoothing algorithm is used to denoise and correct the data, and a continuous and undulating smooth trajectory sequence is generated; the real-time vehicle trajectory data includes vehicle timestamps, spatial coordinates and instantaneous speeds; The lane-changing intention features including lateral displacement change rate and trajectory curvature mutation value are extracted based on the smooth trajectory sequence, the vehicle behavior label is output by inputting the features into a pre-trained behavior classification model, and lane-changing behavior information is obtained by screening; the behavior classification model is constructed based on a random forest algorithm; The vehicle behavior label is spatially associated with road geometric parameters in the lane-level modeling benchmark, the vehicle quantity per unit time in each lane is counted by a density estimation algorithm, and the traffic flow density distribution is calculated by combining lane size; The lane-level modeling benchmark is adjusted based on the smooth trajectory sequence, the lane-changing behavior information and the traffic flow density distribution, and the traffic flow initial state is obtained by weight normalization and particle state estimation; the traffic flow initial state includes lane-level traffic flow density, road section average speed, lane-level traffic volume, lane-level lane-changing frequency, vehicle following distance mean value and road surface friction coefficient.

4. The system of claim 3, wherein, The traffic flow density distribution is calculated by the following formula: p = {p1, p2,..., p L} wherein p represents the traffic flow density distribution of the target road section, p l represents the traffic flow density of the lth lane, N l represents the average number of vehicles in the lth lane within a unit time T, L v represents the standard vehicle length, D represents the length of the target road section, and W represents the width of a single lane.

5. The system of claim 1, wherein, The simulation risk identification module further includes: A risk area identification unit is configured to: The real-time road surface friction coefficient and road section average speed are extracted from the traffic flow initial state, input into a pre-trained neural network model, and a braking distance prediction value is calculated; The road section benchmark traffic capacity is adjusted based on the braking distance prediction value to obtain an adjusted traffic capacity index; The traffic capacity index is combined with road geometric parameters and environmental data in the lane-level modeling benchmark to generate a model input parameter set including normal, rainfall and ice and snow scenarios; The model input parameters are imported into a traffic simulation software to build a multi-scenario traffic flow simulation model, simulate traffic flow running states in different scenarios, and obtain a running feature data set including vehicle delay, congestion duration and conflict point distribution data; Based on a preset risk judgment threshold, a high-risk data association area in the running feature data set is identified by a spatial clustering algorithm to determine a high-risk area distribution represented by polygon boundary coordinates; A traffic flow optimization unit is configured to: Risk level data including average braking distance extension rate and vehicle conflict probability are extracted from the high-risk area distribution, and dynamic traffic control instructions including speed limit, lane control and warning prompt are generated according to a preset rule; The parameters of the traffic flow simulation model are adjusted according to the dynamic traffic control instructions, the adjusted model parameters are combined with road geometric constraints to redistribute the traffic flow of the surrounding road network associated with the high-risk area, and the total delay time, the number of high-risk areas and the vehicle conflict probability before and after optimization are compared to verify the optimized high-risk area distribution; the road geometric constraints include the number of lanes, single-lane width, road curvature radius and road section design slope.

6. The system of claim 5, wherein, The braking distance prediction value is calculated by the following formula: wherein, denotes the braking distance prediction value, x j denotes the input layer variables, j = 1, 2, corresponding to real-time road surface friction coefficient μ and road segment average speed v, N i denotes the number of input layer neurons, N i = 2, N h denotes the number of hidden layer neurons, ω j,i denotes the weight matrix from input layer to hidden layer, ω i,out denotes the weight from hidden layer to output layer, b i , b out denotes the bias vector of hidden layer and output layer, σ hid denotes the hidden layer activation function, α denotes the hyperparameter, σ out denotes the output layer activation function, a linear activation function is adopted to ensure the output to be continuous σ out (z) = z.

7. The system of claim 1, wherein, The risk quantization value of the high-risk area distribution is calculated by using the Monte Carlo simulation algorithm, the key interaction influencing factors are extracted to update the weight matrix of the neural network model, the traffic warning signal is triggered according to the risk level, and the traffic warning signal includes: The space-time distribution features are extracted from the traffic flow data of the high-risk area distribution; the space-time distribution features include traffic volume and speed fluctuation. Adopting a Monte Carlo simulation algorithm to calculate a risk quantization value of the spatio-temporal distribution characteristics, to obtain a risk grade quantization result; Extracting key interaction factors influencing the risk from the risk grade quantization result through mutual information analysis; the key interaction factors include intersection congestion degree and signal light switching frequency; Updating neural network model parameters according to the key interaction factors to obtain an optimized weight matrix; If the weight matrix satisfies a preset convergence condition, triggering a corresponding level of traffic warning signal according to the risk grade.

8. The system of claim 7, wherein, The risk quantization value of a single simulation of the Monte Carlo simulation algorithm is calculated by the following formula: wherein R represents a risk quantification value, with a value range [0, 1], q represents a current traffic flow, q max represents a historical maximum traffic flow of a target road section, represents a speed fluctuation standard deviation, represents a maximum speed fluctuation standard deviation observed in history, and ω1, ω2 represent weight coefficients, with ω1+ω2=1. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the system in any one of claims 1 to 8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the system in any one of claims 1 to 8.

Images