A method for evaluating dust prevention in road construction

Abstract

The present invention discloses a road construction dust prevention assessment method, including constructing a macro traffic flow model based on cellular automata, taking visibility distribution field as input condition, simulating the driver's behavioral response under different visibility, including speed selection and distance maintenance, and obtaining the traffic flow state evolution process in the construction area; taking visibility, speed and distance as state input, using deep reinforcement learning algorithm to train micro driving behavior model, optimizing driving behavior strategy through reward and punishment mechanism, and realizing accurate description and prediction of driving behavior under different visibility conditions; according to the influence degree of road construction dust on traffic safety, calling preset traffic control strategy library, selecting appropriate traffic control scheme for dust prevention, including adjusting lane speed limit value and issuing warning information. The present invention effectively reduces the traffic safety risk caused by construction dust and ensures road traffic safety.

Description

Technical Field

[0001] The invention relates to the field of information technology, and in particular to a road construction dust prevention assessment method. Background Art

[0002] The assessment of the impact of road construction dust on traffic safety faces complex technical challenges. First, the layout of visibility sensors needs to consider the representativeness of spatial distribution and the accuracy of temporal resolution to accurately capture the dynamic changes of dust concentration. Second, the quantitative relationship between dust and visibility is affected by many factors, such as particle size distribution, humidity, wind speed, etc., and the establishment of an accurate model requires comprehensive consideration of these variables. Furthermore, the impact mechanism of dust on traffic flow is still unclear, and a multi-scale coupling model needs to be established to describe the interaction between dust, driving behavior and traffic flow. In addition, there may be interactive effects between the direct and indirect effects of dust on traffic safety. How to reasonably express this complex relationship in the model is a difficult point. Finally, due to the highly uneven temporal and spatial distribution of construction dust, how to design effective sampling strategies and data fusion methods to obtain representative evaluation results is also a technical problem that needs to be solved urgently. Therefore, it is particularly important to comprehensively and accurately reveal the impact of construction dust on road traffic safety by integrating multiple factors. Summary of the invention

[0003] The purpose of the present invention is to solve the problems in the prior art and to propose a road construction dust prevention assessment method.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A road construction dust prevention assessment method comprises the following steps:

[0006] S1, based on the dynamic change characteristics of dust concentration, a distributed visibility sensor network is used for real-time monitoring. The spatiotemporal distribution data of dust concentration in the construction area is obtained through multi-point sampling. The data between sampling points are spatially interpolated to obtain a continuous dust concentration distribution field;

[0007] S2, based on the influencing factors of particle size distribution, humidity and wind speed, a multivariate regression model is established to describe the quantitative relationship between dust concentration and visibility. The model parameters are trained with historical data, and the model is dynamically corrected using real-time monitoring data to calculate the continuous visibility distribution field in the construction area;

[0008] S3, construct a macro traffic flow model based on cellular automation, take the visibility distribution field as input, simulate the driver's speed selection and distance maintenance under different visibility, and obtain the traffic flow state evolution process in the construction area;

[0009] S4, takes visibility, speed and distance as state input, trains the micro-driving behavior model, and optimizes the driving behavior strategy through the reward and punishment mechanism;

[0010] S5, through the relationship between the micro-driving behavior model and the macro-traffic flow model, the obtained micro-behavior data is used as input to adjust the traffic flow model parameters, and according to the traffic flow parameters, the specific impact of dust on the traffic flow parameters is calculated;

[0011] S6, based on the data of dust concentration, visibility, driving behavior and traffic flow parameters, a Bayesian network model was constructed. By analyzing the direct relationship between dust and visibility, and the indirect relationship between driving behavior and traffic flow parameters affecting traffic safety, the influence probability between various factors was obtained. The influence degree of construction dust on traffic safety was obtained through the influence probability.

[0012] S7, according to the impact of construction dust on traffic safety, calls the preset traffic control strategy library and selects the appropriate traffic control plan for dust prevention.

[0013] Compared with the prior art, the present invention has the following advantages:

[0014] The present invention firstly monitors the dust concentration in the construction area in real time through a distributed visibility sensor network, thereby solving the problem of discontinuous and incomplete spatiotemporal distribution data of dust concentration in the construction area, and overcoming the difficulty of accurately quantifying the relationship between dust concentration and visibility, thereby improving the accuracy of dynamically capturing dust concentration; in addition, by analyzing the probability of the influence of dust, visibility, driving behavior and traffic flow parameters on traffic safety, the safety impact of construction dust is quantified, thereby solving the problem that it is difficult to simulate and predict the driver's behavior response under different visibilities; and by dealing with the difficulty of associating and interacting between micro-driving behavior and macro-traffic flow model, data interaction and parameter adjustment are realized, thereby solving the problem that it is difficult to accurately evaluate the comprehensive impact of construction dust on traffic safety.

[0015] In summary, the present invention achieves an accurate description of the relationship between construction dust, visibility, driving behavior, traffic flow and safety, provides technical support for precise and intelligent traffic safety management and control in road construction areas, effectively reduces the traffic safety risks caused by construction dust, and ensures road traffic safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flow chart of a road construction dust prevention assessment method proposed by the present invention;

[0017] Figure 2 A schematic diagram of a road construction dust prevention assessment method proposed by the present invention;

[0018] Figure 3This is a schematic diagram of a road construction dust prevention assessment method proposed by the present invention. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0020] Reference Figure 1-3 As shown, this embodiment discloses a road construction dust prevention assessment method, which specifically includes the following steps:

[0021] S1, according to the dynamic change characteristics of dust concentration, a distributed visibility sensor network is used for real-time monitoring. The spatiotemporal distribution data of dust concentration in the construction area are obtained through multi-point sampling. The data between sampling points are spatially interpolated using an interpolation algorithm to obtain a continuous dust concentration distribution field.

[0022] Obtain the geographic information of the construction area, determine the best sampling point location using the grid method according to the law of dust diffusion, install distributed visibility sensors at the sampling points, and measure the concentration of suspended particulate matter in the air through the principle of light scattering, and transmit the measured data to the central data processing unit in real time. The central data processing unit receives the sensor data of each sampling point, performs median filtering and noise reduction on the sensor data, eliminates the interference of outliers, and obtains the cleaned dust concentration data. The ordinary Kriging interpolation algorithm is used, and the geographic coordinates of the sampling points are used as known points to divide the construction area into regular grids. The dust concentration estimation value of each unknown point in the grid is calculated to generate continuous dust concentration distribution field data. In the interpolation process, the experimental variation function is first constructed, and the exponential theoretical variation function is selected for fitting to estimate the range, base value, and nugget value. Based on the generated dust concentration distribution field data, the point-by-point tracking method is used to construct a two-dimensional dust concentration distribution contour map. The contour map is superimposed on the construction area map using ArcGIS software to intuitively present the spatial distribution of dust pollution and obtain a dust concentration distribution map. Set the data refresh interval to the preset time. When each preset time arrives, re-execute the interpolation calculation and contour drawing, update the dust concentration distribution map, and obtain a continuous dust concentration distribution field. To achieve real-time update and dynamic display of the dust concentration distribution field, set the data refresh interval to 5 minutes. Each refresh re-executes the interpolation calculation and contour drawing, and updates the dust distribution map on the display interface. Add a timeline control to the display interface to allow viewing the dust distribution at different time points. By sliding the timeline, the temporal and spatial variation trend of dust concentration can be dynamically displayed.

[0023] For example, when determining the location of the sampling point, a 10m×10m grid is used to divide the construction area, and the center point of each grid is selected as a potential sampling point. The representative score of each potential point is calculated according to the dust diffusion model, and the 30 points with the highest score are selected as the final sampling points. A TSI DustTrak II 8530 visibility sensor is installed at each sampling point. The sensor uses the 90° light scattering principle to measure the concentration of particulate matter in the range of 0.1-10μm with a measurement accuracy of ±0.1%. The sensor collects data once a minute and transmits it to the central server in real time through the 5G wireless network. After the server receives the raw data, it uses the median filter algorithm of the 5×5 window for noise reduction and filters out outliers. For the cleaned data, the ordinary Kriging interpolation algorithm is used for spatial interpolation. First, the experimental variation function is constructed, and the exponential theoretical variation function is selected for fitting. It is estimated that the range is 200 meters, the base value is 0.8, and the nugget value is 0.1. Then the entire construction area is divided into a regular grid of 2 meters×2 meters, and the dust concentration estimate of each grid point is calculated. Based on the interpolation results, the contour map was drawn using the point-by-point tracking method, and the contour interval was set to 10 μg / m 3 . The generated contour map was imported into ArcGIS 10.8 software, and it was superimposed with the construction area map through the layer overlay function to generate the final spatial distribution map of dust pollution. In order to achieve dynamic display, a 5-minute data refresh interval was set, and interpolation calculations and contour drawing were re-executed for each refresh. A timeline control was added to the WebGIS interface to display the dust distribution in the past 24 hours in hours. Users can view the dust distribution at different time points by sliding the timeline. The system automatically calculates and displays the rate of change of dust concentration, and intuitively displays the temporal and spatial evolution trend of dust pollution.

[0024] Step S2, based on the influencing factors of particle size distribution, humidity and wind speed, a multivariate regression model is established to describe the quantitative relationship between dust concentration and visibility. The model parameters are trained through historical data, and the model is dynamically corrected using real-time monitoring data to calculate the continuous visibility distribution field in the construction area.

[0025] According to the average particle size data measured by the particle size distribution instrument, the relative humidity data collected by the humidity sensor, the wind speed data recorded by the anemometer, and the visibility data measured by the visibility meter, the multivariate linear regression equation V = β0 + β1D + β2H + β3W is obtained, where V represents visibility, D represents average particle size, H represents relative humidity, W represents wind speed, and β0, β1, β2, and β3 are unknown coefficients. According to the multivariate linear regression equation, the least squares method is used to fit the historical monitoring data, and the estimated value of β is obtained by solving the normal equation (XX) β = XY, where X is the independent variable matrix and Y is the dependent variable vector; for the estimated value of β, the 10-fold cross-validation method is used to evaluate the fitting effect of the regression equation, and the data set is randomly divided into 10 parts, 9 parts are used as training sets and 1 part is used as validation sets in turn, and the average mean square error is calculated to obtain the initial dust concentration and visibility relationship model. If the mean square error is less than the preset threshold, the initial model is updated online using real-time monitoring data. The 60 most recent data points are selected using a 1-hour sliding window, and the model parameters are dynamically corrected using a recursive least squares algorithm. The update process of the recursive least squares algorithm is as follows: Kt = Pt-1xt / (1+xt

[0026] Pt-1xt), βt=βt-1+Kt(yt-xtβt-1), Pt=Pt-1-KtxtPt-1, where K is the gain matrix, P is the error covariance matrix, x is the input vector, and y is the observation value. Based on the calibrated multivariate regression model, combined with the real-time particle size distribution, humidity and wind speed data of each monitoring point in the construction area, the visibility value of the monitoring point is calculated. For the area between the monitoring points, the inverse distance weighted interpolation method is used for spatial interpolation. The spatial interpolation is based on the visibility values ​​of the monitoring points calculated based on the calibrated multivariate regression model, and then the continuous visibility distribution field of the entire construction area is obtained. The interpolation formula is V(x,y)=Σ(Vi / di^p) / Σ(1 / di^p), where V(x,y) is the visibility of the point to be interpolated, Vi is the visibility of the known point, di is the distance from the point to be interpolated to the known point, and p is the distance power, which is 2. This method is used to generate a continuous visibility distribution map of the entire construction area, realizing real-time dynamic monitoring and prediction of visibility.

[0027] Exemplarily, 10 monitoring points are arranged at the construction site, each equipped with a TSI8533 particle size distribution meter, a VaisalaHMP155 humidity sensor, a GillWindSonic anemometer, and a BiralVPF-730 visibility meter. The particle size distribution meter measures the PM2.5 and PM10 concentrations once a minute, and calculates the average particle size D; the humidity sensor records the relative humidity H once a minute; the anemometer measures the wind speed once a second, and takes the 1-minute average as the wind speed W; the visibility meter measures the visibility V once a minute. The data collector transmits these data to the central processing server in real time. After receiving the data, the central processing server first constructs the multivariate linear regression equation V=β0+β1D+β2H+β3W. Using the historical data of the past 30 days, a total of 43,200 data points, the initial coefficient values ​​are calculated by the least squares method. The model was evaluated using the 10-fold cross-validation method. The data was randomly divided into 10 parts, 9 parts were used for training each time, and 1 part was used for validation. This was repeated 10 times, and the average mean square error was calculated to be 0.15 km. 2 . After obtaining the initial model, the system receives new monitoring data once a minute, and uses a 1-hour sliding window of 60 data points to update the model parameters through a recursive least squares algorithm. The updated model is applied to the real-time data of 10 monitoring points to calculate the visibility value of each point. For the area between the monitoring points, the system uses the inverse distance weighted interpolation method for spatial interpolation, and selects the distance power p=2 during interpolation. Finally, a visibility distribution map with a grid resolution of 100m×100m is generated and updated every minute. The visibility distribution map is displayed in real time through the web interface. The color from red to green indicates visibility from low to high, and the specific values ​​of the 10 monitoring points are displayed at the same time.

[0028] Step S3, constructing a macroscopic traffic flow model based on cellular automation, taking the visibility distribution field as input condition, simulating the driver's behavioral response under different visibilities, including speed selection and distance keeping, and obtaining the traffic flow state evolution process in the construction area.

[0029] The construction area information with road network characteristics is received, and the construction area information is used to construct a macro traffic flow model based on cellular automata; the road is divided into a number of cells with a preset length according to the construction area information, and each cell state is represented by a preset value to indicate whether it is idle or occupied by a vehicle; the visibility distribution field data is obtained, and the visibility distribution field data is used as the input condition of the cellular automata model; the continuous visibility value is discretized into a plurality of preset levels according to the visibility distribution field data; for each cell, the corresponding maximum allowable speed and minimum safe vehicle distance parameters are set according to the visibility level of its position; the driver behavior model is used to describe the driver's reaction under different visibility conditions, and the driver behavior model includes speed selection and vehicle distance keeping behavior; the evolution process of traffic flow over time is simulated by iteratively calculating the cell state update, and the traffic flow state change corresponding to the preset time period is iteratively calculated; it is judged whether the traffic flow state change meets the preset conditions; if the traffic flow state change meets the preset conditions, the traffic flow dynamic characteristics of the entire construction area including the position, speed and acceleration information of each vehicle are obtained.

[0030] For example, according to the characteristics of the road network in the construction area, a macro traffic flow model based on cellular automata is constructed, and the road is divided into several cells with a length of 7.5 meters. The state of each cell is 0 for idle and 1 for occupied by vehicles. For a two-way four-lane road, each cross section contains 8 cells, and the number of longitudinal cells is determined by the actual length of the road section. The vehicle movement rule is defined as: if the n cells ahead, n is the speed, are idle, then the vehicle moves forward n cells; otherwise, it moves to the nearest idle cell. The update rule adopts a synchronous update method, and all vehicles update their positions simultaneously in each time step. The visibility distribution field data is used as the input condition of the cellular automata model, and the continuous visibility values ​​are discretized into 5 levels: 0-50 meters is level 1, 51-100 meters is level 2, 101-200 meters is level 3, 201-300 meters is level 4, and above 301 meters is level 5. For each cell, the corresponding maximum allowed speed and minimum safe distance parameters are set according to the visibility level of its location, such as level 1 visibility corresponds to a maximum speed of 20 km / h and a minimum safe distance of 15 meters. The driver behavior model is used to describe the driver's response under different visibility conditions, including speed selection and distance keeping behavior. The speed selection adopts a fuzzy inference algorithm, with the input variables being the current visibility level and the occupancy status of the 5 cells ahead, and the output variable being the expected speed. Fuzzy rule example If the visibility is low and the front is crowded, the expected speed is very low. Distance keeping is calculated based on the improved intelligent driving model, considering the change in the driver's reaction time under different visibility. The reaction time increases with the decrease in visibility level, ranging from 0.8 seconds to 2 seconds. The evolution of traffic flow over time is simulated by iteratively calculating the cell state update. Each iteration corresponds to 1 second of actual time, and 3600 iterations are performed continuously to obtain the traffic flow state changes within 1 hour. In terms of boundary processing, a dynamic vehicle generation and disappearance mechanism is adopted. Vehicles are generated at the entrance according to the preset traffic flow, and vehicles are directly removed at the exit. For complex road conditions, such as lane changes, they are realized by checking the spatial conditions of adjacent lanes; for intersections, signal light control rules are set to manage vehicle traffic. Finally, the dynamic characteristics of traffic flow in the entire construction area, including the position, speed and acceleration information of each vehicle, are obtained. A 1-kilometer-long two-way four-lane road is selected in the construction area for traffic flow micro-simulation. The road is divided into 1066 cells, each 7.5 meters long. Using the pre-acquired visibility distribution field data as input, the data shows that the visibility of the road section gradually decreases from the starting point to the end point, corresponding to five levels of 5, 4, 3, 2, and 1 respectively. According to the visibility level, the maximum allowable speed is set to 100km / h, 80km / h, 60km / h, 40km / h and 20km / h, and the minimum safe vehicle distance is 30 meters, 25 meters, 20 meters, 18 meters and 15 meters. At the traffic flow entrance, 0.5 vehicles are generated per second, corresponding to a traffic flow of 1,800 vehicles per hour.Fuzzy reasoning algorithm is used for speed selection, and 9 fuzzy rules are set. If the visibility is high and the road ahead is unobstructed, the expected speed is high. The improved intelligent driving model is used for vehicle distance maintenance, in which the driver's reaction time is set to 2 seconds, 1.6 seconds, 1.3 seconds, 1 second and 0.8 seconds respectively with visibility level from level 1 to level 5. Fixed-cycle signal lights are set at intersections, with green light time of 45 seconds, yellow light time of 3 seconds and red light time of 42 seconds. The traffic flow evolution process within 1 hour is obtained by 3600 iterations. The results show that in the sections with low visibility level 1-2, the average speed is reduced to 15-30km / h and the distance between vehicles is shortened to 10-15 meters, while in the sections with good visibility level 4-5, the average speed is maintained at 70-90km / h and the distance between vehicles is 25-35 meters. The traffic flow status is displayed in real time through a dynamic visual interface, with different colors indicating vehicle speeds. Red indicates low speeds of 0-20km / h, yellow indicates medium speeds of 21-60km / h, and green indicates high speeds of more than 61km / h.

[0031] In step S4, visibility, vehicle speed and vehicle distance are used as state inputs, and a deep reinforcement learning algorithm is used to train the micro-driving behavior model. The driving behavior strategy is optimized through a reward and punishment mechanism to achieve accurate description and prediction of driving behavior under different visibility conditions.

[0032] Obtain visibility, vehicle speed and vehicle distance parameters as state space input, which includes five visibility levels, ten vehicle speed intervals and five vehicle distance levels. According to the state space input, acceleration, deceleration, maintaining current speed and lane change actions are determined as behavior space. The reward function is constructed using three evaluation indicators: safety, efficiency and comfort. Safety is calculated based on the time interval with the preceding vehicle, efficiency is calculated based on the difference between the current speed and the expected speed, and comfort is calculated based on the rate of change of acceleration. The driving behavior model is established using the deep Q network algorithm. The deep Q network includes an input layer, three hidden layers and an output layer. The number of input layer nodes is the state space dimension, and the number of output layer nodes is the action space dimension. The interaction data is stored and updated through the experience replay mechanism. The experience replay mechanism uses a circular queue to store experience tuples. When the replay buffer is full, the new data overwrites the oldest data. If the average reward value change rate of ten consecutive evaluations is less than the preset threshold, the training is stopped, and the optimal driving behavior strategy under different visibility conditions is obtained.

[0033] Exemplarily, a driving environment simulator is constructed based on actual traffic data, and parameters such as visibility, vehicle speed and vehicle distance are used as state space inputs. Visibility is divided into 5 levels, namely 0-50m, 51-100m, 101-200m, 201-300m, and above 300m; vehicle speed is discretized into 10 intervals, one interval every 20km / h; vehicle distance is divided into 5 levels, namely 0-10m, 11-20m, 21-30m, 31-50m, and above 50m. Actions such as acceleration, deceleration, maintaining the current speed and changing lanes are defined as behavior space. The driving behavior model is constructed using the deep Q network algorithm. The input layer has the number of nodes that is the state space dimension, which is 3 in this case. The three hidden layers have the number of nodes that are 64, 128 and 64 respectively. The output layer has the number of nodes that is the action space dimension, which is 4 in this case. The hidden layer uses the ReLU activation function, and the output layer uses the linear activation function. The Q value update formula is Q(s,a)=Q(s,a)+α[r+γ*max(Q(s',a'))-Q(s,a)], where α is the learning rate, γ is the discount factor, Q(s,a) represents the Q value corresponding to taking action a in state s, r is the immediate reward obtained by taking action a in state s, and max(Q(s',a')) represents the maximum Q value that can be obtained by taking all possible actions a' in the new state s'. The experience replay mechanism is used to store and update the interaction data, and the replay buffer size is set to 10000. Each training randomly samples data with a batch size of 64 for gradient descent. In the specific implementation, a circular queue is used to store the experience tuple (s,a,r,s'). When the buffer is full, the new data will overwrite the oldest data. The learning rate is set to 0.001 and the discount factor is 0.99. The ε value of the ε-greedy strategy decays linearly from 1 to 0.1, episode represents the current number of training rounds, total_episodes represents the total number of training rounds, and the decay formula is ε=max(0.1,1-episode / total_episodes). The driving behavior strategy is optimized through iterative training, and each training round contains 1000 time steps. The model performance is evaluated every 100 rounds, and the evaluation indicators include average reward value, safe driving time ratio and average speed. Training is stopped when the average reward value change rate of 10 consecutive evaluations is less than 1% or reaches the preset threshold of 500. Finally, the optimal driving behavior strategy for different visibility conditions is obtained, which is manifested as automatically reducing the speed and increasing the distance between vehicles in low visibility of 0-100m, and maintaining a higher speed and normal distance above 200m in high visibility. More specifically, a deep reinforcement learning algorithm is implemented on a simulated 10-kilometer highway section to optimize driving behavior. First, the road is divided into 100-meter grids, and each grid records the current visibility, average vehicle speed, and average vehicle distance.The visibility data is obtained from the weather station in real time and discretized into 5 levels: 0-50m is recorded as 1, 51-100m is recorded as 2, and so on. The speed data is measured by the roadside radar and discretized into 10 intervals: 0-20km / h is recorded as 1, 21-40km / h is recorded as 2, and so on. The distance between vehicles is calculated by camera recognition and divided into 5 levels. The constructed deep Q network contains 3 input nodes corresponding to visibility, vehicle speed and vehicle distance, 3 hidden layers with 64, 128 and 64 nodes respectively, and 4 nodes in the output layer corresponding to the four actions of acceleration, deceleration, speed maintenance and lane change. During the training process, 1,000 vehicles are simulated passing through the road section each time, and the state, action and reward obtained by each vehicle in each 100-meter grid are recorded. The reward function is designed to give a penalty of -10 when the distance to the vehicle in front is less than the safe distance; when the speed is close to the expected speed, a reward of +1 is given according to visibility; and a penalty of -1 is given for sudden acceleration or deceleration. A circular queue of size 10,000 is used to store these empirical data. After every 100 simulations, 64 sets of data are randomly selected for a network parameter update. The ε value starts at 1 and decreases by 0.1 every 1,000 simulations until it drops to 0.1. After about 10,000 simulations, or 10 million sets of data, the model converges. The final strategy shows that when visibility is less than 100m, the vehicle speed will automatically drop to below 40km / h and the distance between vehicles will increase to more than 50m; when visibility is 100-200m, the vehicle speed is maintained at 60-80km / h and the distance between vehicles is 30-50m; when visibility is greater than 200m, the vehicle speed can reach more than 100km / h and the distance between vehicles is maintained at 20-30m.

[0034] Step S5, through the relationship between the micro driving behavior model and the macro traffic flow model, the obtained micro behavior data is used as input to adjust the traffic flow model parameters, and according to the traffic flow parameters, the specific impact of dust on the traffic flow parameters is calculated, including the change data of the average vehicle speed, flow and density indicators.

[0035] According to the vehicle trajectory data obtained by the micro-driving behavior model, the micro-features such as the instantaneous speed, acceleration and headway of each vehicle within a 1-second interval are extracted from the vehicle trajectory data; the driving behavior is classified into different modes based on the micro-features using the K-means clustering algorithm, and the number of clusters is determined by the silhouette coefficient method, and 3-5 categories are finally selected; the proportion distribution of each behavior mode is calculated, such as the proportion of radical, conservative and moderate types. The traffic flow model based on cellular automata is used to describe the macro traffic flow characteristics, and the road is divided into several cells, each with a length of 7.5 meters. According to the proportion of driving behavior modes, the parameters in the cellular automata model are adjusted, such as the deceleration probability p and the random slowing probability p0. The specific adjustment method is p=w1p1+w2p2+w3p3, where w1, w2, and w3 are the proportions of the three behavior modes, and p1, p2, and p3 are the corresponding benchmark probability values. The relationship model between dust concentration and traffic flow parameters is established by using the multivariate regression analysis method. The independent variables of the relationship model include dust concentration data and visibility level data obtained from the environmental monitoring station. The dust concentration data include PM2.5 and PM10, and the dependent variables are average vehicle speed, flow rate and density. The significant variables are selected by stepwise regression method, and the regression coefficient is estimated by the least square method to obtain the influence function of dust on traffic flow parameters. The adjusted cellular automaton model is combined with the dust influence function for iterative calculation. The initial condition is set as the current measured traffic state of the road, and each iteration corresponds to 1 minute of actual time. In each iteration, the change of traffic flow parameters is first calculated according to the current dust conditions, and then the changed parameters are substituted into the cellular automaton model for evolution. If the average vehicle speed change rate for 10 consecutive iterations is less than the preset threshold value such as 1%, it is determined that the system has reached a stable state and the iteration is stopped. Finally, the spatiotemporal distribution data of indicators such as average vehicle speed, flow rate and density are calculated to form a quantitative description of the impact of dust on traffic flow.

[0036] For example, the technical solution is implemented on a 10-kilometer-long urban main road. First, the driving data of 1,000 vehicles within 1 hour are collected through roadside radars and cameras, and the position, speed and acceleration are recorded once a second. The K-means clustering algorithm is used to analyze these data, and the number of clusters is set to 3, 5, and 7. The optimal number of clusters is finally determined to be 5 by calculating the silhouette coefficient. Five driving behavior modes are obtained: aggressive type 20%, robust type 35%, conservative type 25%, following type 15% and lane change type 5%. The road is divided into 1,333 cells, each cell is 7.5 meters long. According to the distribution of behavior patterns, the parameters of the cellular automaton model are calculated: deceleration probability p = 0.2 × 0.1 + 0.35 × 0.3 + 0.25 × 0.5 + 0.15 × 0.4 + 0.05 × 0.2 = 0.325, random slowing probability p0 = 0.2 × 0.05 + 0.35 × 0.1 + 0.25 × 0.15 + 0.15 × 0.1 + 0.05 × 0.2 = 0.1075. The PM2.5, PM10 concentration and visibility data for 24 hours on the day were obtained from the environmental monitoring station, with a total of 1440 data points. At the same time, the corresponding average speed, flow and density data were obtained from the traffic monitoring system. The multivariate regression model was constructed using the stepwise regression method. The final model included three independent variables: PM2.5, visibility and time, and R 2 The value is 0.82. Set the initial traffic state to the current measured average speed of 40km / h, flow of 1800 vehicles / hour, and density of 45 vehicles / km. Perform iterative calculations, with each iteration representing 1 minute of real time. In the first iteration, when PM2.5 is 150μg / m 3 , when visibility is 2km, the regression model predicts that the average speed will drop to 35km / h. Substitute this new speed into the cellular automaton model to simulate the traffic state after 1 minute. Repeat this process until the average speed change rate is less than 1% for 10 consecutive iterations. The final results show that in heavy pollution PM2.5>150μg / m 3 Under these conditions, the average vehicle speed stabilized at 28 km / h, the traffic volume dropped to 1,200 vehicles / hour, and the density increased to 60 vehicles / km.

[0037] Step S6, based on the data of dust concentration, visibility, driving behavior and traffic flow parameters after changes, a Bayesian network model is constructed. By analyzing the direct relationship between dust and visibility, and the indirect relationship between driving behavior and traffic flow parameters affecting traffic safety, the influence probability between various factors is obtained, and the degree of influence of construction dust on traffic safety is obtained by calculating the influence probability.

[0038] The dust concentration, visibility, driving behavior and traffic flow parameters were obtained to construct the Bayesian network structure of the parameters. The initial network structure was determined through an expert questionnaire survey, and the connections between nodes were adjusted in combination with data correlation analysis. According to the Bayesian network structure, the K2 algorithm was used for structural optimization. The maximum number of parent nodes was set to 3, and the BIC scoring function was used. The optimized structure output by the K2 algorithm was used as the input for parameter learning. For the optimized Bayesian network structure, the conditional probability table of each node in the Bayesian network was calculated using the maximum likelihood estimation method. For complete data, the frequency was directly calculated as the probability estimate; for missing data, the EM algorithm was used for parameter learning. The EM algorithm initialized the parameters to uniform distribution, set the maximum number of iterations to 100, and the convergence threshold to 0.001. The conditional probability distribution between nodes in the network was obtained through multiple iterative optimizations. According to the conditional probability table, the variable elimination algorithm was used to calculate the probability of the direct impact of dust concentration on visibility. The variable elimination algorithm also calculated the probability of the indirect impact of dust concentration on traffic safety by affecting visibility, driving behavior and traffic flow parameters. The variable elimination order was carried out from leaf nodes to root nodes according to the network topology. For each variable, the probability and of all possible values ​​are calculated. By using marginal probability calculation and chain rule, the overall impact of construction dust on traffic safety is calculated by comprehensively considering direct and indirect impacts. If the direct impact probability and indirect impact probability are obtained, the log-likelihood ratio method is used to quantify the impact. The calculation formula of the log-likelihood ratio method is LLR = the logarithm of the ratio of the safety probability under dust conditions to the non-dust conditions, that is, LLR = log(P(safety|dust) / P(safety|no dust)). The LLR value is mapped to between 0 and 1 through the sigmoid function to obtain the final impact score. Set the impact threshold: 0-0.2 for low risk, 0.2-0.5 for medium risk, 0.5-0.8 for high risk, and 0.8-1 for extremely high risk.

[0039] For example, the technical solution is implemented in a large road construction project. First, 10 environmental monitoring points and 20 traffic monitoring points are arranged at the construction site, and data is collected every minute. The environmental monitoring points collect PM2.5 in the range of 0-500μg / m 3 、PM10 range 0-600μg / m 3and visibility range 0-10km data; traffic monitoring points collected data for speed range 0-120km / h, headway range 0-10s, and traffic flow range 0-3000 vehicles / hour. The network structure was initially determined through an expert questionnaire survey, and then the node connection was adjusted using Pearson correlation coefficient analysis, with the correlation coefficient threshold set to 0.3. The K2 algorithm was used to optimize the network structure, setting the maximum number of parent nodes to 3, using the BIC scoring function, and iterating 500 times to obtain the optimal structure. For about 80% of the complete data set, the frequency was directly calculated as the probability estimate; for missing data, the EM algorithm was used for parameter learning, the initialization parameters were uniformly distributed, the maximum number of iterations was set to 100, and the convergence threshold was 0.001. In 90% of cases, the algorithm converged within 50 iterations. The variable elimination algorithm was used to calculate the impact probability, and the elimination was carried out in the order of traffic flow → headway → speed → visibility → PM10 → PM2.5. The overall impact of construction dust on traffic safety was obtained through marginal probability calculation and chain rule. The log-likelihood ratio method is used to quantify the impact, and the calculated LLR value is mapped to the 0-1 interval through the sigmoid function. The final results show that when the PM2.5 concentration exceeds 150μg / m 3 When the traffic safety risk level rises from 0.3 to 0.7, it enters the high-risk range.

[0040] Step S107, based on the impact of road construction dust on traffic safety, call the preset traffic control strategy library and select a suitable traffic control plan for dust prevention, including adjusting lane speed limit values ​​and issuing warning information.

[0041] A multi-level fuzzy comprehensive evaluation model is established based on dust concentration, visibility and traffic flow parameters. The model contains three levels: the first level is dust concentration, the second level is visibility and traffic flow parameters, and the third level is comprehensive impact degree. The data-driven method is used to determine the membership function. The distribution characteristics of each indicator at different levels are obtained through statistical analysis of historical data, and the membership function is constructed accordingly. The impact degree is divided into three levels: low, medium and high. The degree to which the current state belongs to each level is calculated through the membership function. A preset traffic control strategy library is constructed. The strategy library contains lane speed limit adjustment, warning information release, vehicle diversion, temporary detour, sprinkler dispatch and dust net deployment strategies. Each strategy corresponds to different dust impact levels and traffic flow conditions. The decision tree algorithm is used to learn historical data and generate strategy selection rules. The training features of the decision tree include dust concentration, visibility, traffic flow, average vehicle speed and headway. The CART algorithm is used to construct a binary decision tree, and the Gini index is used as the splitting criterion. The fuzzy comprehensive evaluation results and the rules generated by the decision tree are used to select the most suitable traffic control plan for the current situation from the strategy library. Considering multiple factors such as dust concentration, visibility, traffic flow and vehicle speed, the weight of each factor is determined by the hierarchical analysis method, and the applicability score of each candidate plan is calculated by the weighted sum method. If the scores of multiple candidate plans are similar, such as the difference is less than 0.1, the plan with the least impact on traffic is selected; the selected traffic control plan is executed through the road traffic management system, which includes sending new speed limit values ​​to variable speed limit signs, displaying warning information on variable information boards, adjusting traffic light timing plans, and pushing relevant information through traffic broadcasts and navigation software APIs. A distributed coordination mechanism is used to handle the coordination problems between multiple traffic management systems, and a master control node is set to uniformly dispatch each subsystem. Based on the execution results, the effectiveness of the management and control strategy is evaluated every 15 minutes, and the strategy parameters are dynamically adjusted when necessary.

[0042] For example, the technical solution is implemented in a 10-kilometer-long highway construction section. First, 20 dust monitoring points and 50 traffic monitoring points are arranged to collect data once a minute. In the multi-level fuzzy comprehensive evaluation model, the dust concentration is divided into 5 levels including 0-50, 51-100, 101-150, 151-200, >200μg / m 3, visibility is divided into 4 levels including <50, 50-200, 200-500, >500m, and traffic flow parameters include 3 levels each for vehicle speed and flow. Through statistical analysis of 3 months of historical data, the membership function of each indicator is constructed. The decision tree algorithm adopts the CART method, using 80% of the historical data, about 86,400 records, for training, and the remaining 20% ​​for verification. Finally, a decision tree with a depth of 8 is generated with an accuracy rate of 85%. The factor weights determined by the hierarchical analysis method are dust concentration 0.4, visibility 0.3, traffic flow 0.2, and vehicle speed 0.1. The strategy library contains 20 preset schemes, such as reducing the speed limit by 20% + starting a sprinkler truck, diverting vehicles by 30% + setting up dust nets, etc. When the system detects that the PM2.5 concentration exceeds 150μg / m 3 , when visibility drops to 300m, the fuzzy evaluation results in a high impact of 0.75. The decision tree selects three candidate solutions, with weighted scores of 0.82, 0.79, and 0.78, respectively. The solution with the highest score is selected: reduce the speed limit from 100km / h to 80km / h, start four sprinkler trucks, and set up temporary dust nets within a range of 3km. The main control node sends instructions to the five subsystems to coordinate the implementation of the control plan. After 15 minutes of evaluation, it was found that the dust concentration dropped by 10%, the traffic flow decreased by 5%, and the average vehicle speed decreased by 15km / h. The system determines that the strategy is effective and continues to be implemented.

[0043] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A road construction dust prevention assessment method, characterized in that: The following steps are involved: S1, based on the dynamic change characteristics of dust concentration, a distributed visibility sensor network is used for real-time monitoring. The spatiotemporal distribution data of dust concentration in the construction area is obtained through multi-point sampling. The data between sampling points are spatially interpolated to obtain a continuous dust concentration distribution field; S2, based on the influencing factors of particle size distribution, humidity and wind speed, a multivariate regression model is established to describe the quantitative relationship between dust concentration and visibility. The model parameters are trained with historical data, and the model is dynamically corrected using real-time monitoring data to calculate the continuous visibility distribution field in the construction area; S3, construct a macro traffic flow model based on cellular automation, take the visibility distribution field as input, simulate the driver's speed selection and distance maintenance under different visibility, and obtain the traffic flow state evolution process in the construction area; S4, takes visibility, speed and distance as state input, trains the micro-driving behavior model, and optimizes the driving behavior strategy through the reward and punishment mechanism; S5, through the relationship between the micro-driving behavior model and the macro-traffic flow model, the obtained micro-behavior data is used as input to adjust the traffic flow model parameters, and according to the traffic flow parameters, the specific impact of dust on the traffic flow parameters is calculated; S6, based on the data of dust concentration, visibility, driving behavior and traffic flow parameters, a Bayesian network model was constructed. By analyzing the direct relationship between dust and visibility, and the indirect relationship between driving behavior and traffic flow parameters affecting traffic safety, the influence probability between various factors was obtained. The influence degree of construction dust on traffic safety was obtained through the influence probability. S7, according to the impact of construction dust on traffic safety, call the preset traffic control strategy library and select the appropriate traffic control plan for dust prevention; S5 includes: Acquiring vehicle trajectory data, and extracting microscopic features of instantaneous speed, acceleration, and headway from the vehicle trajectory data; classifying driving behaviors into different modes according to the micro-features; A macroscopic traffic flow model based on cellular automata is used to describe the macroscopic traffic flow characteristics, and the parameters in the cellular automata model are adjusted according to the proportion of driving behavior patterns. Establish a relationship model between dust concentration and traffic flow parameters. The independent variables of the relationship model include dust concentration data and visibility level data obtained from environmental monitoring stations, and the dependent variables are average vehicle speed, flow rate and density. The adjusted cellular automation model is combined with the dust influence function for iterative calculation. If the average vehicle speed change rate of 10 consecutive iterations is less than the preset threshold, the system is judged to have reached a stable state. The spatiotemporal distribution data of average vehicle speed, flow rate and density indicators are calculated to form a quantitative description of the impact of dust on traffic flow.

2. A road construction dust prevention assessment method according to claim 1, characterized in that: The S1 specifically includes the following steps: S101, obtain geographic information of the construction area, determine the best sampling point location according to the dust diffusion law, and install distributed visibility sensors at the sampling points; S102, the distributed visibility sensor measures the concentration of suspended particles in the air and transmits the measurement data to the central data processing unit in real time; S103, the central data processing unit receives the sensor data of each sampling point, performs median filtering and noise reduction processing on the sensor data, and obtains the dust concentration data after cleaning; S104, taking the geographical coordinates of the sampling points as known points, dividing the construction area into regular grids, and interpolating and calculating the estimated value of dust concentration at each unknown point in the grid; S105, if the dust concentration estimation value is obtained, constructing a two-dimensional dust concentration distribution contour map; S106, superimposing the contour map onto the construction area map to obtain a dust concentration distribution map; S107, setting a preset time interval for data refresh, re-executing interpolation calculation and contour drawing when each preset time arrives, updating the dust concentration distribution map, and obtaining a continuous dust concentration distribution field.

3. A road construction dust prevention assessment method according to claim 1, characterized in that: The S2 specifically includes the following steps: S201, obtaining a multivariate linear regression equation V=β0+β1D+β2H+β3W, where V represents visibility, D represents average particle size, H represents relative humidity, and W represents wind speed; S202, fitting the historical monitoring data according to the multivariate linear regression equation, and obtaining an estimated value of β by solving the normal equation (XᵀX)β=XᵀY, where X is an independent variable matrix and Y is a dependent variable vector; S203, evaluating the fitting effect of the regression equation for the estimated value of β, and calculating the average mean square error; S204, if the average mean square error is less than a preset threshold, the initial model is updated online using the real-time monitoring data, and the model parameters are dynamically corrected; S205, performing spatial interpolation using an interpolation method, and obtaining a continuous visibility distribution field of the entire construction area based on the visibility values ​​of the monitoring points calculated by the calibrated multivariate regression model.

4. A road construction dust prevention assessment method according to claim 1, characterized in that: The S3 includes: S301, receiving construction area information with road network characteristics, and constructing a macro traffic flow model based on cellular automata; S302, dividing the road into a number of cells with a preset length according to the construction area information, and each cell state is represented by a preset value to indicate whether it is idle or occupied by a vehicle; S303, obtaining the visibility distribution field data in S2, and using the visibility distribution field data as input conditions of a cellular automation model; S304, discretizing the continuous visibility values ​​into a plurality of preset levels according to the visibility distribution field data; S305, for each cell, setting corresponding maximum allowable vehicle speed and minimum safe vehicle distance parameters according to the visibility level of the location where the cell is located; S306, using a driver behavior model to describe the driver's response under different visibility conditions, wherein the driver behavior model includes vehicle speed selection and vehicle distance keeping behavior; S307, simulating the evolution of traffic flow over time by iteratively calculating the cell state update, and iteratively calculating the traffic flow state change corresponding to the preset time period; S308, determining whether the traffic flow state change satisfies a preset condition. If the traffic flow state change satisfies the preset condition, obtaining the dynamic characteristics of the traffic flow in the entire construction area including the position, speed and acceleration information of each vehicle.

5. A road construction dust prevention assessment method according to claim 1, characterized in that: In S4, the state input includes five visibility levels, ten vehicle speed intervals, and five vehicle distance levels. According to the state input, acceleration, deceleration, maintaining the current speed, and lane change actions are determined as the behavior space, and the reward function is constructed using evaluation indicators of safety, efficiency, and comfort. The safety is calculated based on the time interval with the preceding vehicle, the efficiency is calculated based on the difference between the current speed and the expected speed, and the comfort is calculated based on the acceleration change rate. A driving behavior model is established using a deep Q network algorithm, wherein the deep Q network includes an input layer, three hidden layers, and an output layer, wherein the number of nodes in the input layer is the state space dimension, and the number of nodes in the output layer is the action space dimension; The interaction data is stored and updated through an experience replay mechanism, which uses a circular queue to store experience tuples. When the replay buffer is full, new data overwrites the oldest data. If the change rate of the average reward value of ten consecutive evaluations is less than the preset threshold, the training is stopped and the optimal driving behavior strategy under different visibility conditions is obtained.

6. A road construction dust prevention assessment method according to claim 1, characterized in that: The S6 includes: Obtain the dust concentration in S1, visibility in S2, driving behavior in S4, and adjusted traffic flow parameters in S5, and construct a Bayesian network structure; According to the Bayesian network structure, structural optimization is performed, and for the optimized Bayesian network structure, a conditional probability table of each node in the Bayesian network is calculated; According to the conditional probability table, the probability of the direct impact of dust concentration on visibility is calculated, and the probability of the indirect impact of dust concentration on traffic safety by affecting the visibility, the driving behavior and the traffic flow parameters is also calculated; For the direct impact probability and the indirect impact probability, the log-likelihood ratio method is used to quantify the impact degree, wherein the calculation formula of the log-likelihood ratio method is LLR=logarithm of the ratio of the safety probability under dust conditions to that under no dust conditions.

7. A road construction dust prevention assessment method according to claim 1, characterized in that: S7 includes: A multi-level fuzzy comprehensive evaluation model was established based on the dust concentration in S1, visibility in S2, and adjusted traffic flow parameters in S5; The membership function is determined by using data-driven methods, and the distribution characteristics of each indicator at different levels are analyzed through historical data statistics; Build a preset traffic control strategy library containing multiple strategies, use decision tree algorithm to learn historical data, and generate strategy selection rules; Using the fuzzy comprehensive evaluation result and the strategy selection rule, the most suitable traffic control scheme for the current situation is selected from the strategy library. If multiple candidate schemes have similar scores, the scheme with the least impact on traffic is selected. The selected traffic control scheme will be implemented through the road traffic management system.

Images