Traffic dynamic management method based on traffic data dimensionality reduction reconstruction and multidimensional analysis

Abstract

This invention relates to the field of traffic data processing technology, specifically a traffic dynamic management method based on traffic data dimensionality reduction and reconstruction and multidimensional analysis. The method includes the following steps: constructing a traffic intersection area model, planning traffic paths within the area model, extracting information hierarchically from the original data of each area and performing dimensionality reduction and reconstruction using matrices and coordinates (specifically including data collection, road condition dimensionality reduction and reconstruction, vehicle condition dimensionality reduction and reconstruction, and storage of dimensionality reduction information); extracting hierarchical dimensionality reduction and reconstruction data from each area to form new coordinate curves or matrix data models; performing multidimensional feature analysis; and performing mixed calculations of multidimensional feature data before implementing dynamic traffic management. Compared with existing technologies, this method significantly reduces the storage capacity of traffic data, obtains quantifiable technical indicators through dimensionality reduction and reconstruction, and can more accurately identify data patterns, providing reliable technical support for intelligent traffic management, vehicle-road cooperation, and autonomous driving.

Description

Technical fields:

[0001] This invention relates to the field of traffic data processing technology, specifically a traffic dynamic management method based on traffic data dimensionality reduction and multidimensional analysis, which can effectively improve traffic management capabilities by performing dimensionality reduction and reconstruction of traffic data, efficient and accurate visualization and multidimensional indicator analysis, and intelligent output of control decisions. Background technology:

[0002] With the increase in the number of motor vehicles, traffic congestion and traffic accidents are becoming more and more common. However, due to the relatively backward traffic data processing technology, problems such as heavy traffic data storage pressure, low vehicle traffic efficiency, high energy consumption, high pollution, and untimely detection of traffic accidents are becoming increasingly prominent, putting traffic management work under tremendous pressure.

[0003] The main method for storing traffic data is multi-point storage of road traffic video and image data, but the storage cost is quite high. Current methods for alleviating traffic management pressure mainly include road expansion, controlling traffic volume (such as license plate restrictions), encouraging public transportation and cycling, tidal flow lanes, and staggered peak hours, but the results are not significant. Currently, there are also technologies that analyze historical traffic data to improve management experience and strategies, a typical example being the "green wave" technology. However, practice has shown that sudden events such as temporary parking obstructing traffic and traffic accidents often frequently cause traffic congestion. Therefore, analyzing historical data alone cannot yield the optimal traffic management strategy, and "green wave" is only applicable to specific traffic directions and speeds, making its practicality limited.

[0004] Currently, a small number of artificial intelligence algorithms are used to improve traffic control capabilities by collecting and analyzing traffic data in real time and optimizing traffic light timings. However, practice has shown that these methods suffer from two main problems: firstly, incomplete data feature mining, low data utilization efficiency, and relatively general technical methods; secondly, over-reliance on the "black box" operation of artificial intelligence, which makes it impossible to quantify and explain the content of data analysis and computation, and cannot avoid the problem of "AI uncertainty." If such technologies are widely applied to public safety fields such as traffic control, safety hazards are inevitable. Summary of the Invention:

[0005] This invention addresses the shortcomings and deficiencies of existing technologies by proposing a traffic dynamic management method based on traffic data dimensionality reduction and multidimensional analysis. This method enables real-time dimensionality reduction and reconstruction of traffic data, followed by the conversion of the data into matrix and coordinate data models, which are then analyzed and processed by artificial intelligence algorithms. This results in efficient and accurate output of management strategies, thereby effectively improving traffic management capabilities.

[0006] This invention achieves its purpose through the following measures:

[0007] A traffic dynamic management method based on traffic data dimensionality reduction reconstruction and multidimensional analysis is characterized by the following steps:

[0008] Step 1: Construct a traffic intersection area model. Using the stop line and its extension as boundaries, designate the central intersection area as the control zone and the road extension areas in multiple directions as the monitoring zone. Within each monitoring zone, designate the area from which data flows out of the monitoring zone to the control zone as the out zone and the area from which data flows in of the control zone to the monitoring zone as the in zone.

[0009] Step 2: Plan the traffic paths of the area model: Starting from each out zone of the monitoring area, connect it with the corresponding in zone in three directions: left, straight, and right, and construct a connection diagram between the out zone and the in zone;

[0010] Step 3: Extract information from the original data of each region in layers and use matrices and coordinates for dimensionality reduction and reconstruction. This includes data collection, road condition dimensionality reduction and reconstruction, vehicle condition dimensionality reduction and reconstruction, and storage of dimensionality reduction information.

[0011] Step 4: Extract hierarchical dimensionality reduction and reconstruction data from each region to form new coordinate curves or matrix data models, and perform multi-dimensional feature analysis, specifically including:

[0012] (1) Establish a monitoring data model for vehicles waiting to be released in the OUT area: monitor the vehicle condition matrix in each OUT area, calculate and extract the number of vehicles waiting to be released in the matrix at a certain time in real time and form a new matrix. Then, establish a coordinate system with time * quantity and generate a monitoring curve of the number of vehicles waiting to be released in the OUT area for feature analysis.

[0013] (2) Establish a monitoring data model for on-the-road traffic flow in OUT areas: monitor the vehicle condition matrix in each OUT area, calculate the number and location of on-the-road traffic flow in real time and form a new matrix, then establish a coordinate system with time * quantity, generate on-the-road traffic flow monitoring curves in OUT areas for feature analysis.

[0014] (3) Establish a fitting data model for the passage time of vehicles waiting to be released in the OUT area: monitor the vehicle condition matrix in each OUT area, calculate the number of vehicles waiting to be released in a certain batch in the matrix and the passage time, then establish a coordinate system with the number * time, record and fit multiple batches of data, and obtain the fitting data curve of the passage time of vehicles waiting to be released in the OUT area for feature analysis.

[0015] (4) Establish a vehicle outflow data model for the OUT area (the model principle can be used for vehicle inflow data analysis in the controlled area): Monitor the vehicle condition matrix in each OUT area, set the statistical duration t, calculate the number of vehicles outflowing in time intervals or cumulatively in units of t and form a new matrix, then establish a coordinate system with time * quantity to form the vehicle outflow data curve of the OUT area for feature analysis.

[0016] (5) Establish a vehicle inflow data model for IN zone (the model principle can be used for vehicle outflow data analysis in the controlled area): Monitor the vehicle condition matrix of each IN zone, set the statistical duration t, calculate the number of vehicles inflow in time-by-time or cumulatively in units of t and form a new matrix, then establish a coordinate system with time * quantity to form the vehicle inflow data curve of IN zone for feature analysis.

[0017] (6) Establish vehicle speed tracking data models for each region: monitor the vehicle condition matrix in each region, extract the speed data of each vehicle, calculate the average speed of the vehicle, and form a new matrix. Then, establish a coordinate system based on time * speed, and generate vehicle speed data curves in real time for feature analysis.

[0018] (7) Establish vehicle density distribution data model for each region: monitor the vehicle condition matrix of each region, calculate the vehicle coverage rate in the matrix in real time and form a new matrix, then establish a coordinate system with time * coverage rate, and generate vehicle density distribution curves in real time for feature analysis.

[0019] Step 5: Multi-dimensional feature data hybrid calculation: This includes calculating and acquiring vehicles en route and vehicles waiting to be released in the OUT area, accident monitoring in various areas, and congestion status monitoring in the IN and controlled areas. The calculation and prediction results will provide information and data support for dynamic traffic management.

[0020] Step 5-1: Using the data models in Step 4 (1) and (2), obtain or predict in real time the number of vehicles waiting to be released and the number of large traffic flows in the OUT area:

[0021] The calculated number of vehicles waiting to be released in the lane is N. await Vehicle status pending release z await :

[0022]

[0023] The calculated number of vehicles in the lane with heavy traffic is N. onway The traffic flow status is z. onway ,

[0024]

[0025] Step 5-2: Using the data models in Step 4 (6) and (7), obtain the vehicle speed characteristics and vehicle density information of the area changing over time. Set the density threshold as D and the duration threshold as P. By monitoring whether the duration of a sudden drop in vehicle speed to 0 and a vehicle density greater than D exceeds P, conduct accident monitoring in the monitoring and control areas. When the density in the area exceeds D and the duration exceeds P, or the number of vehicles in the area whose speed drops to 0 is ≥1, it is considered that a traffic accident has occurred; otherwise, it is considered that no traffic accident has occurred. The traffic accident status AD is as follows:

[0026]

[0027] Step 5-3: Using the data models (4), (5), (6), and (7) in Step 4, obtain the characteristic information of vehicle inflow and outflow, vehicle speed, and vehicle density changing over time in the IN zone and the control zone. Use machine learning algorithms to calculate the congestion status of the IN zone and the control zone. Specifically, the data models (4), (5), (6), and (7) in Step 4 are used in the control zone, and the data models (5), (6), and (7) in Step 4 are used in the IN zone. The calculation method for the congestion status of the control zone and the IN zone is the same.

[0028] Step 6: Dynamic traffic control, including traffic light control:

[0029] Step 6-1-1: Set initial state: Set all traffic lights to red (red light = 1) as the initial state;

[0030] Step 6-1-2: Determine the passable route: Based on Step 2, determine the combination of straight and left-turn lanes that can be allowed to pass simultaneously (red light = 0). The right-turn lane is then determined to be passable based on the traffic data of the aforementioned combination.

[0031] Step 6-1-3: Single-lane release quota calculation: Based on the values ​​or status outputs in steps 5-1, 5-2, and 5-3, calculate the single-lane release quota. The release priority and whether to allow passage are determined by the single-lane release quota value. The formula for calculating the single-lane release quota is as follows:

[0032] Let x n To calculate the total number of vehicles in a specific lane within the monitored area, the calculation is as follows:

[0033] x n =z await *N await +z onway *N onway

[0034] Let x spec This is a special vehicle (special operations vehicle) status.

[0035] Let x in The area is congested.

[0036] Let x cross The controlled area is congested.

[0037] Let x ad Current lane abnormal status,

[0038] Let y be the single-lane clearance index, then the expression for evaluating y is:

[0039] y = f(x) spec ,x in ,x cross ,x ad ,x n )

[0040] =C spec x spec +C in x in +C cross x cross +C ad x ad

[0041] +(1-x spec (1-x) in (1-x) cross (1-x) ad )*x n

[0042] Where, C spec C in C cross C ad C is a constant. spec >>x n C in >>x n C cross >>x n C ad >>x n .

[0043] Step 3 of this invention is specifically implemented through the following steps:

[0044] Step 3-1: Data Collection: Within the monitored and controlled areas, use maps, road engineering drawings, and cameras to collect road condition information by road segment, including lane length and width, lane line position, sidewalk position, roadside entrances and exits, intersections, road obstacles, road curvature, and slope information; collect video, images, and radar signal information on vehicle characteristics, traffic flow, and vehicle movement status by road segment;

[0045] Step 3-2: Road Condition Dimension Reconstruction: (1) First, establish a rectangular coordinate system and set the scale to make an accurate correspondence between the real road conditions and the coordinate system; (2) Set a line segment in the coordinate system to correspond to the parking line at the traffic intersection, and then use this line segment as the starting point to mark and reconstruct the length, width, lane position, sidewalk position, roadside entrances and exits, intersections, road obstacles, road curvature, and slope information of the out area in the coordinate system; (3) Divide the reconstructed road marked in the coordinate system into K areas and perform m i *n i Grid-based partitioning, with values ​​ranging from m. i >0、n i Given that K ≥ i > 0, and K ≥ i > 0, the information within the segmented grid is then stored using a matrix.

[0046] Step 3-3: Vehicle Condition Reconstruction: (1) Collect video, images, and radar signal information of the vehicle condition in real time according to K regions, and then use image recognition and signal recognition artificial intelligence algorithms to detect, track, and extract feature information indicators of vehicle characteristics, vehicle position, and vehicle movement status, and then according to m i *n i The matrix stores data, and after data overlay, a vehicle condition coordinate map can be generated in real time from the road condition map, thereby enabling a visual display of road condition and vehicle condition information;

[0047] Steps 3-4: Dimensional reduction information storage: Store the acquired multi-dimensional matrix data of road conditions and vehicle conditions. If there is little effective information in the matrix, the sparse matrix can be compressed and stored to further save storage space.

[0048] The calculation steps for the congestion status of the control area and IN area in step 5-3 of this invention are as follows:

[0049] Step 5-3-1: Construct the original input dataset: Using the data models in Step 4 (4), (5), (6), (7), obtain the vehicle location feature matrix, vehicle inflow and outflow feature curves, and vehicle speed feature matrix, and then define the congestion status corresponding to the features as labels;

[0050] Step 5-3-2: Feature preprocessing: Arrange the vehicle location feature matrix in chronological order to form a set of serialized matrix blocks, and then flatten each matrix block to obtain linear features; normalize the vehicle inflow and outflow feature curves; use the three preprocessed features as the input features of the model.

[0051] Step 5-3-3: Build a deep learning model: (1) Build a self-attention mechanism network, input the preprocessed vehicle position feature matrix into this network, and output the linear feature layer result; (2) Build a convolutional neural network with residual structure, and finally output a linear layer, input the vehicle speed feature matrix into this network, and output the linear feature layer result; (3) Combine the normalized vehicle inflow and outflow feature curve with the two linear layer features, input the fully connected network and activate it to predict the congestion status;

[0052] Step 5-3-4: Model Training: Feed the dataset into the deep learning model training in Step 5-3-3, use the cross-entropy loss function to calculate the loss of the network model. During the training process, if the difference in loss decrease between the first n (n≥1) epochs and the last n epochs is less than a fixed value (fixed value>0), then the model training ends and the calculation parameters in the model are saved.

[0053] Step 5-3-5: Congestion Calculation: Real-time acquisition of vehicle location feature matrix, vehicle inflow / outflow feature curves, and vehicle speed feature matrix. After preprocessing in Step 5-3-2, these are input into the network to obtain the calculation result PC. The calculation result represents the congestion probability. A threshold PC is set. th Similarly, when calculating congestion status in the IN area, the result is PI, and the threshold is set to PI. th .

[0054] Step 6 of this invention also includes overflow prevention and control, i.e., controlling the inflow of vehicles into the in zone. This is achieved through the calculation in step 6-1, obtaining an index for the in zone. When x exists... in When = 1, traffic congestion information data for the in section is output, and red lights are used to control vehicle flow.

[0055] Step 6 of this invention also includes accident monitoring and control. Based on the calculations in step 6-1, when a lane has an x... ad When = 1, traffic accident information data will be output.

[0056] Step 6 of this invention also includes congestion management and control. Through the calculations in step 6-1, indicators for the control area are obtained. When x exists... cross When = 1, traffic congestion information data of the controlled area road section will be output, and red lights will be used to control vehicle inflow in the corresponding out area.

[0057] Step 6 of this invention also includes inefficient traffic restriction management. Through step 6-1, the index of the out zone is obtained. When the green light state is in effect, if the release index of a certain lane is 0, the red light is switched and other lanes are reselected for release.

[0058] Compared with existing technologies, this invention (1) reduces the storage capacity and cost of traffic data by dimensionality reduction and reconstruction of real-time traffic data such as videos, pictures, and radar signals, thereby improving the data processing efficiency. Secondly, it obtains a series of quantifiable technical indicators through dimensionality reduction and reconstruction, making traffic data processing more accurate and standardized. Thirdly, it combines a series of quantifiable technical indicators with technologies such as digital twins and virtual engines to virtually restore real historical traffic data. (2) By using coordinate curves and matrix data models to perform multi-dimensional feature analysis on technical indicator data, it can more accurately obtain data patterns, discover data anomalies, and realize the visualization of traffic data processing. This can solve the problems of "black box" operation and "AI uncertainty" in artificial intelligence, and broaden the path for the widespread application of artificial intelligence technology in public safety fields such as traffic management. (3) The coordinate curves, matrix data models, and algorithms innovatively launched in this invention can be deeply integrated with artificial intelligence technology and products. Through continuous accumulation and optimization, they can provide reliable technical support for intelligent traffic management, vehicle-road collaboration, and autonomous driving. Attached image description:

[0059] Appendix Figure 1 This is a system flowchart of the present invention.

[0060] Appendix Figure 2 This is a regional map of the cross-shaped traffic intersection in Embodiment 1 of the present invention.

[0061] Appendix Figure 3 This is the intersection traffic path diagram in Embodiment 1 of the present invention.

[0062] Appendix Figure 4 This is a schematic diagram of the road conditions in the monitoring area in Embodiment 1 of the present invention.

[0063] Appendix Figure 5 This is the full road condition information matrix diagram in Embodiment 1 of the present invention.

[0064] Appendix Figure 6 This is a schematic diagram of the vehicle condition in the monitoring area in Embodiment 1 of the present invention.

[0065] Appendix Figure 7 This is a schematic diagram of the vehicle conditions in the controlled area in Embodiment 1 of the present invention.

[0066] Appendix Figure 8 This is a matrix diagram of vehicle location information in the monitoring area in Embodiment 1 of the present invention.

[0067] Appendix Figure 9 This is a matrix diagram of vehicle location information in the control area in Embodiment 1 of the present invention.

[0068] Appendix Figure 10 This is a schematic diagram of the vehicle inflow data curve in the controlled area in Embodiment 1 of the present invention.

[0069] Appendix Figure 11 This is a sequence diagram of vehicle inflow data in the controlled area in Embodiment 1 of the present invention.

[0070] Appendix Figure 12 This is a schematic diagram of the vehicle outflow data curve in the controlled area in Embodiment 1 of the present invention.

[0071] Appendix Figure 13 This is a schematic diagram of vehicle speed tracking data in the control area in Embodiment 1 of the present invention.

[0072] Appendix Figure 14 This is a data matrix diagram of vehicle speed tracking in the control area in Embodiment 1 of the present invention.

[0073] Appendix Figure 15 This is a schematic diagram of the vehicle density distribution data curve in the controlled area in Embodiment 1 of the present invention.

[0074] Appendix Figure 16 This is the lane condition matrix diagram in Embodiment 2 of the present invention.

[0075] Appendix Figure 17 This is the lane width road condition matrix diagram in Embodiment 2 of the present invention.

[0076] Appendix Figure 18 This is a schematic diagram of the vehicle inflow data curve in the IN area in Embodiment 2 of the present invention.

[0077] Appendix Figure 19 This is a schematic diagram of the vehicle speed tracking data curve in the IN area of ​​Embodiment 2 of the present invention.

[0078] Appendix Figure 20 This is a schematic diagram of the average speed tracking data curve of vehicles in the IN zone in Embodiment 2 of the present invention.

[0079] Appendix Figure 21 This is the vehicle speed tracking data matrix diagram in the IN area of ​​Embodiment 2 of the present invention.

[0080] Appendix Figure 22 This is a schematic diagram of the vehicle density distribution data curve in the IN area of ​​Embodiment 2 of the present invention.

[0081] Appendix Figure 23 This is a schematic diagram of the monitoring data curve of vehicles waiting to be released in the OUT area in Embodiment 3 of the present invention.

[0082] Appendix Figure 24 This is a schematic diagram of the on-the-go traffic monitoring data curve in the OUT area in Embodiment 3 of the present invention.

[0083] Appendix Figure 25 This is a schematic diagram of the fitting data curve of the time of vehicles waiting to be released in the OUT area in Embodiment 3 of the present invention. Detailed implementation method:

[0084] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0085] Example 1:

[0086] This example demonstrates how to partition traffic conditions into zones, construct a connectivity graph of travel routes, and then use algorithms to monitor and manage congestion in the controlled area in real time by performing dimensionality reduction and multi-dimensional feature analysis on traffic information such as road conditions. The specific process is as follows:

[0087] Step 1: (e.g.) Figure 2 As shown, a regional model of a cross-shaped traffic intersection is constructed. The central intersection area of ​​the traffic intersection is set as the control zone, and the road extension areas in multiple directions are set as monitoring zones. Then, within each monitoring zone, the area from which data flows out of the monitoring zone to the control zone is set as the OUT zone, and the area from which data flows in of the control zone to the monitoring zone is set as the IN zone. In this embodiment, east, south, west, and north are represented by the letters E, S, W, and N, respectively. For example, OUTE represents the OUT zone from west to east flowing out of the control zone, and INE represents the IN zone from west to east flowing in of the control zone to the monitoring zone.

[0088] Step 2: (e.g.) Figure 3 (As shown) Plan the traffic path model for this intersection: Starting from each OUT zone of the monitoring area, connect with the corresponding IN zones in three directions: left, straight, and right, and construct a connection diagram between the OUT zones and IN zones.

[0089] Step 3: Dimensionality Reduction and Reconstruction of Original Data: This involves four steps: data acquisition, road condition dimensionality reduction and reconstruction, vehicle condition dimensionality reduction and reconstruction, and storage of dimensionality-reduced information.

[0090] Step 3-1: Data Collection: Within the monitored and controlled areas, use maps, road engineering drawings, and cameras to collect road condition information by road segment, including the length and width of the road lanes, lane line positions, sidewalk positions, roadside entrances and exits, intersections, road obstacles, road curvature, and slope information; collect video, images, and radar signal information on vehicle characteristics, traffic flow, and vehicle movement status by road segment;

[0091] Step 3-2: Road condition dimensionality reduction and reconstruction (e.g.) Figure 4 , Figure 5 As shown): (1) First, establish a rectangular coordinate system and set the scale so that there is an accurate correspondence between the actual road conditions and the coordinate system. In this example, as shown Figure 4 As shown, suppose the actual road length L of a certain road segment (out area) is 1400 meters and the maximum road width W is 20 meters. Assume each scale of the x-axis of the coordinate system maps to 50 meters of actual road length, and each scale of the y-axis maps to 2 meters of actual road width. Then, as shown... Figure 4As shown, taking 28 marks on the x-axis and 10 marks on the y-axis can completely map the real road conditions. Figure 5 This is a schematic diagram using a matrix to store the boundary dimensions and parking line positions of this road section. Figure 5 In the coordinate system, 0 represents driving lane information, 1 represents roadside information, and 2 represents parking line information; (2) Set a line segment in the coordinate system to correspond to the parking line at the traffic intersection, and then use this line segment as the starting point to mark and reconstruct the road condition length, width, lane line position, sidewalk position, roadside entrance and exit, intersection, road obstacles, road curvature, and slope information of the out area in the coordinate system. In this example, set a line segment at the position of x=0, y=10 to correspond to the parking line at the traffic intersection, and then use this line segment as the starting point to mark and reconstruct the road condition information such as the length, width, lane line position, and uphill and downhill slope of the out area in the coordinate system; (3) Divide the road marked and reconstructed in the coordinate system into K areas and perform m i *n i The grid is divided into sections, and the information within each section is stored in a matrix. In this example, we set K=28, meaning the grid is divided into 28 equal parts according to the horizontal axis scale. If we want to analyze the road conditions in the 16th grid region (i=16), such as... Figure 4 , Figure 5 The given area corresponds to a real road length of 50 meters and a width of 12 meters. We can further divide this area using a 2m x 2m grid, thus forming a 25m x 6 matrix (m...). 16 =25, n 16 =6) Used to store road condition information for the 16th grid area;

[0092] Step 3-3: Vehicle condition downgrading and reconstruction (e.g.) Figure 6 and Figure 8 (As shown) (1) Real-time collection of vehicle status video, images, and radar signal information in K regions, followed by the use of image recognition and signal recognition artificial intelligence algorithms to detect, track, and extract feature information indicators of vehicle characteristics, vehicle position, and vehicle movement, and then according to m i *n i The matrix stores data, and after data overlay, a vehicle condition coordinate map can be generated in real time from the road condition map, thus visualizing road and vehicle condition information. In this example, we continue from step 3-2, taking the 16th network area of ​​this road segment as an example. Figure 8 As shown, this road segment is in a 25*6 matrix (m 16 =25, n 16 =6) Based on the recorded road condition information, then use the same 25*6 matrix (m 16 =25, n 16 =6) Record vehicle location information, in Figure 8In the diagram, 0 represents a space not occupied by a vehicle, and 1 represents a space occupied by a vehicle.

[0093] Steps 3-4: Dimensionality reduction information storage: (e.g.) Figure 5 , Figure 8 , Figure 9 , Figure 11 , Figure 14 , Figure 16 , Figure 17 , Figure 21 As shown, the acquired multi-dimensional matrix data of road and vehicle conditions is stored. By comparing the matrix data with video and image data, it can be found that the storage space required for the matrix data is greatly reduced.

[0094] Step 4: Extract the dimensionality-reduced and reconstructed data of the control area to form a new coordinate curve or matrix data model, and perform multi-dimensional feature analysis, specifically including:

[0095] (1) Establish a data model for vehicle inflow into the controlled area: (e.g.) Figure 10 (As shown) A vehicle condition matrix is ​​used to monitor the area. A statistical duration t is set, and the number of vehicles flowing in is calculated in units of t (e.g., t = 1 second in this example). The inflow volume at each moment is summed to form a new sequence. Then, a coordinate system is established using time * quantity to generate a vehicle inflow data curve for feature analysis (see sequence). Figure 11 );

[0096] (2) Establish a data model for vehicle outflow from the controlled area: (e.g.) Figure 12 (As shown) Monitor the vehicle condition matrix within the area, set the statistical duration t (e.g., t = 1 second in this example), calculate the number of vehicles leaving in time intervals or cumulatively in units of t to form a new matrix, and then establish a coordinate system with time * quantity to form a vehicle outflow data curve for feature analysis;

[0097] (3) Establish a vehicle speed tracking data model for the controlled area: (e.g.) Figure 13 (As shown) Monitor the vehicle condition matrix in each area, extract the speed data of each vehicle, and calculate the continuous speeds of multiple vehicles to form a new matrix. Then, establish a coordinate system based on time * speed, and generate vehicle speed data curves in real time for feature analysis. The newly formed matrix is ​​defined as u*v (e.g., u = 11, v = 16 in this example), which consists of multiple time-varying vehicle speed sequences. u represents the length of the vehicle speed time sequence, and v represents the number of vehicles (see...). Figure 14 (The value -1 in the matrix represents that the vehicle does not exist at that moment);

[0098] (4) Establish a vehicle density distribution data model for the controlled area: (e.g.) Figure 15(As shown) The vehicle condition matrix within the monitoring area is used to calculate the vehicle coverage rate in the matrix in real time to form a new matrix. Then, a coordinate system is established with the time * coverage ratio, and a vehicle density distribution curve is generated in real time for feature analysis.

[0099] Step 5: Multi-dimensional feature data hybrid calculation: Monitor the congestion status of the controlled area. The calculation and prediction results of this part will serve as a prerequisite for dynamic traffic control.

[0100] Step 5-1: Using the data models in Step 4 (3) and (4), obtain the vehicle speed characteristics that change over time within the controlled area (see details). Figure 13 Features include the duration of the speed drop to 0, the number of vehicles with a speed of 0, etc., and vehicle density information (see details). Figure 15 If the density threshold is set to D = 0.8 and the duration threshold is P = 4, and multiple vehicles with a sudden drop in speed to 0 are detected and the duration of the vehicle density being greater than D exceeds P, an accident may occur in the controlled area, and the accident status is AD = 1.

[0101] Step 5-2: Using the data models (1), (2), (3), and (4) in Step 4, obtain the characteristic information of vehicle inflow and outflow, vehicle speed, and vehicle density changing over time within the controlled area. Use machine learning algorithms to calculate the congestion status of the controlled area. The steps are as follows: 1. Obtain the traffic flow characteristics of the controlled area over a period of time from the data models (1) and (2) in Step 4, including the traffic flow information flowing into and out of the controlled area (see details). Figure 10 and Figure 12 See sequence Figure 11 ); 2. Obtain the vehicle speed curve (or matrix) within the control area from the data model in step 4(3) (see details). Figure 13 See the schematic matrix. Figure 14 ), and then extract the speed change features for 10 seconds from the curve; 3. Obtain the vehicle density in a certain area of ​​the controlled area at a certain time from the data model in step 4 (4) (see details). Figure 15 ) and vehicle location matrix (see vehicle condition diagram) Figure 7 The position matrix is ​​shown below. Figure 9 Based on the above three pieces of information, set the congestion threshold PC for the controlled area. th =0.8, using a multi-input deep learning network for training and prediction, the control area is judged to be in a congested state.

[0102] Step 6: Relieve congestion (initialize red lights, clear data):

[0103] Let x cross The controlled area is congested. Determine that lane x exists cross=1, therefore, output traffic congestion information data of the controlled area road section, and use red lights to control vehicle flow (initialize red lights);

[0104] Example 2:

[0105] This example addresses congestion in the IN area and prevents overflow by performing the following operations:

[0106] By dividing road conditions into zones and constructing a connectivity map of travel routes, and by performing dimensionality reduction and reconstruction of traffic information such as road conditions and conducting multi-dimensional feature analysis, algorithms are used to monitor the congestion status of the IN zone in real time and implement dynamic traffic control.

[0107] Specifically as follows:

[0108] Step 1: Same as Step 1 in Example 1.

[0109] Step 2: Same as step 2 in Example 1.

[0110] Step 3: This embodiment is basically the same as step 3 in embodiment 1, but the difference is that this embodiment does not use [a specific method] when recording road condition information features. Figure 5 The matrix shown (which integrates roadside information, lane lines, and driving lane information into a single matrix) is not like... Figure 16 and Figure 17 As shown, lane lines and driving lane information are recorded using different matrices. Figure 16 In the matrix, 2 represents lane line information, and 0 represents information outside the lane lines. Figure 17 In the matrix, 1 represents roadside information and 0 represents lane information.

[0111] Step 4: Extract the dimensionality-reduced and reconstructed data from the IN region to form a new coordinate curve or matrix data model, and perform multi-dimensional feature analysis, specifically including:

[0112] (1) Establish a data model for vehicle inflow in the IN zone: (e.g.) Figure 18 (As shown) The vehicle condition matrix in the monitoring area is set, the statistical duration is set to t (t = 1 second in this example), the number of vehicles flowing in is calculated in units of t, the inflow at each moment is summed to form a new sequence, and then a coordinate system is established with time * quantity to form a vehicle inflow data curve for feature analysis.

[0113] (2) Establish a vehicle speed tracking data model for the IN zone: (e.g.) Figure 19 and Figure 20(As shown) Monitor the vehicle condition matrix in each area, extract the speed data of each vehicle, and calculate the continuous speeds of multiple vehicles to form a new matrix. Then, establish a coordinate system based on time * speed, and generate vehicle speed data curves in real time for feature analysis. The newly formed matrix is ​​defined as u*v (e.g., u = 11, v = 32 in this example), which consists of multiple time-varying vehicle speed sequences. u represents the length of the vehicle speed time sequence, and v represents the number of vehicles (see...). Figure 21 , Figure 21 The image only displays the first 16 columns of the numerical matrix (where -1 indicates that the vehicle does not exist at that moment).

[0114] (3) Establish a data model for vehicle density distribution in the IN zone: (e.g.) Figure 22 (As shown) The vehicle condition matrix within the monitoring area is used to calculate the vehicle coverage rate in the matrix in real time to form a new matrix. Then, a coordinate system is established with the time * coverage ratio, and a vehicle density distribution curve is generated in real time for feature analysis.

[0115] Step 5: Using the data models (1), (2), and (3) in Step 4, obtain the characteristic information of vehicle inflow and outflow, vehicle speed, and vehicle density changing over time in the IN zone. Use machine learning algorithms to calculate the congestion status of the IN zone. The steps are as follows: 1. Obtain the vehicle inflow characteristics of the IN zone over a period of time from the data model in Step 4 (1) (see details). Figure 18 ); 2. Obtain the vehicle speed curve (or matrix) in the IN region from the data model in step 4(2) (see details). Figure 19 See matrix Figure 21 Then, extract the speed change characteristics for 10 seconds from the curve; 3. Obtain the vehicle density in a certain area of ​​the controlled area at a certain time from the data model in step 4(3) (see details). Figure 22 ) and vehicle position matrix (see matrix) Figure 8 Based on the above three pieces of information, set the IN area congestion threshold PI. th =0.8, using a multi-input deep learning network for training and prediction to determine congestion status.

[0116] Step 6: Prevent overflow (control vehicle inflow into the in zone):

[0117] Let x in The area is congested. Determine that lane x exists in =1, output traffic congestion information data for the in section of the road, and use red lights to control the inflow of vehicles into the lane.

[0118] Example 3:

[0119] This example addresses a road condition where there are no vehicles in the OUT zone and traffic is inefficiently prohibited. The following operations are performed:

[0120] By dividing road conditions into zones and constructing a traffic path connectivity map, and through dimensionality reduction and multi-dimensional feature analysis of traffic information such as road conditions, algorithms are used to capture the idle status of lanes in the OUT zone in real time and implement dynamic traffic control. Specifically:

[0121] Step 1: Same as Step 1 in Example 1.

[0122] Step 2: Same as step 2 in Example 1.

[0123] Step 3: Same as step 3 in Example 1.

[0124] Step 4: Extract the OUT region data, reduce its dimensions, and reconstruct it to form a new coordinate curve or matrix data model. Perform multi-dimensional feature analysis, specifically including:

[0125] (1) Establish a data model for monitoring vehicles waiting to be released in the OUT area: (e.g.) Figure 23 (As shown) The vehicle condition matrix within the monitoring area is used to calculate and extract the number of vehicles waiting to be released at a certain moment in real time to form a new matrix. Then, a coordinate system is established with time * quantity to generate a monitoring curve of the number of vehicles waiting to be released for feature analysis.

[0126] (2) Establish a data model for monitoring on-the-go traffic flow in the OUT area: (e.g.) Figure 24 (As shown) The vehicle condition matrix in the monitoring area is used to calculate the number and location of on-the-road traffic in real time to form a new matrix. Then, a coordinate system is established by time * quantity to generate on-the-road traffic monitoring curves for feature analysis.

[0127] (3) Establish a data fitting model for the passage time of vehicles waiting to be released in the OUT area: (e.g.) Figure 25 (As shown) The vehicle condition matrix in the monitoring area is used to calculate the number of vehicles waiting to be released and their passage time in a certain batch within the matrix. Then, a coordinate system is established with the number * time. Data from multiple batches is recorded and fitted to obtain the fitted data curve of the passage time of vehicles waiting to be released for feature analysis.

[0128] Step 5: Using the data models (1) and (2) in Step 4, obtain or predict in real time the number of vehicles waiting to be released in the OUT area and the number of large traffic flows on the way. Specifically, obtain the number N of vehicles waiting to be released in the left-turn, right-turn, and straight-ahead lanes in the four directions (east, west, south, and north) from the data model (1). await Information and waiting vehicle status z await Information (details can be found here) Figure 23 The number of vehicles N in transit in the four directions (east, west, south, and north) is obtained from the data model (2). onway Information and status of traffic flow on the road onway Information (details can be found here) Figure 24 The pending traffic flow N is obtained from the data model (3). await Predicted release time (see details) Figure 25 Based on the above three pieces of information, the release time and release time are determined.

[0129] Step 6: Inefficient traffic restrictions (solving the problem of idle green lights):

[0130] Set initial state: Set all traffic lights to red (red = 1) initially, and set the initial vehicle clearance threshold to N. start If the total number of vehicles in a certain lane of the monitoring area obtained in step 5-1 exceeds this threshold, then priority can be given to allowing passage. Let x be the threshold value. n The total number of vehicles to be calculated for a specific lane within the monitoring area is calculated as follows:

[0131] x n =z await *N await +z onway *N onway

[0132] When the light is green, determine lane x. n =0, therefore the light for that lane turns red, and other lanes are allowed to proceed.

[0133] Example 4:

[0134] By dividing road conditions into zones and constructing a connectivity map of traffic routes, and through dimensionality reduction and multi-dimensional feature analysis of traffic information such as road conditions, algorithms are used to monitor the congestion status of the controlled area in real time and implement dynamic traffic control. This example uses the same steps as Example 1, with the specific details of the machine learning algorithm in step 5-3 as follows:

[0135] The steps for using machine learning algorithms to predict traffic congestion are as follows:

[0136] Step 1: Construct the original input dataset: Extract three features from the data processing model to describe the congestion status of the control area, including the vehicle location feature matrix, the vehicle inflow and outflow feature curves, and the vehicle speed feature matrix or feature map matrix, and define the congestion status corresponding to each feature as labels; among them, the vehicle inflow and outflow features are obtained through the inflow and outflow curves, with an acquisition length of 10 seconds in history; the vehicle speed feature is the vehicle speed map or the feature matrix after dimensionality reduction and reconstruction of the original data, which contains the speed records of multiple vehicles in this area, with an acquisition length of 10 seconds in history; the labels are defined by setting thresholds: if the scatter plot coverage of the vehicle location matrix is ​​greater than 70%, the difference between the vehicle inflow and outflow in the historical records is greater than 4 for more than 5 seconds, and the average vehicle speed in the vehicle speed map drops below 10 kilometers per hour, then it is defined as congested; otherwise, it is considered smooth.

[0137] Step 2: Feature preprocessing: First, the vehicle location feature matrix is ​​preprocessed and arranged into a set of sequential matrix blocks in chronological order. Then, the matrix is ​​flattened to obtain a sequence, so that each time step is represented by a linear sequence. Second, the vehicle inflow and outflow feature curves are normalized. Finally, the preprocessed features are used as the input features of the model.

[0138] Step 3: Building the Deep Learning Model: 1) First, build a fully attention-based network. Input the preprocessed vehicle position feature matrix into the fully connected attention-based network. This process involves two encoder layers. The first sub-layer connection structure includes a multi-head self-attention sub-layer, a normalization layer, and a residual connection. The second sub-layer connection structure includes a feedforward fully connected sub-layer, a normalization layer, and a residual connection. Then, the matrix output from the encoder is passed through a linear layer to output the linear feature layer result. 2) Second, build a convolutional neural network with a residual structure. Input the vehicle speed feature matrix or vehicle speed feature map into the convolutional neural network with a residual structure. First, all data are fed into the same two-dimensional convolutional layer. The filter count is 32, the stride is 2, and the kernel size is 7*7. Then, a max pooling layer with a stride of 2 and a kernel size of 3*3 is applied. Next, two residual modules are passed, each containing two convolutional layers with identical parameters, 32 filters, and a kernel size of 3*3. After processing by each residual module, the unprocessed information is concatenated and fed into the next residual module. Finally, an average pooling layer is applied, followed by a fully connected layer to output the linear feature layer result. 3) Finally, the normalized vehicle inflow / outflow feature curve is concatenated with the two linear layer features and input into the fully connected network. Finally, it is fed into a Softmax layer for congestion prediction.

[0139] Step 4: Model Training: Feed the dataset into the network from Step 3 for model training. Calculate the network model loss using the cross-entropy loss function until the model loss is less than a predetermined threshold L. Then, model training ends, and the computational parameters within the model are saved. The cross-entropy loss function is used to calculate the network model loss. Let the prediction result be... The true label is y, and the cross-entropy loss function formula is as follows:

[0140]

[0141] During training, if the difference in loss decrease between the first 10 epochs and the last 10 epochs is less than 0.1, the model training ends, and the computational parameters within the model are saved for real-time prediction.

[0142] Step 5: Congestion Prediction: In real time, three features describing the congestion status of the controlled area are obtained from the data processing model: vehicle location feature matrix, vehicle inflow / outflow feature curve, and vehicle speed feature matrix. After preprocessing in Step 2, these features are input into the network to obtain the prediction result PC, which represents the congestion probability. Then, a threshold PC is set. th =0.8. If it is greater than the threshold, it is considered congested; if it is less than the threshold, it is considered smooth, in order to predict the congestion status.

[0143] Compared with existing technologies, this invention (1) reduces the storage capacity and cost of traffic data by dimensionality reduction and reconstruction of real-time traffic data such as videos, pictures, and radar signals, thereby improving the data processing efficiency. On the other hand, it obtains a series of quantifiable technical indicators through dimensionality reduction and reconstruction, making traffic data processing more accurate and standardized. (2) By using coordinate curves and matrix data models to perform multi-dimensional feature analysis on technical indicator data, it can more accurately obtain data patterns, discover data anomalies, and realize the visualization of traffic data processing. This can solve the problems of "black box" operation and "AI uncertainty" in artificial intelligence, and broaden the path for the widespread application of artificial intelligence technology in public safety fields such as traffic control. (3) The coordinate curves, matrix data models, and algorithms innovatively launched in this invention can be deeply integrated with artificial intelligence technology and products. Through continuous accumulation and optimization, they can provide reliable technical support for intelligent traffic control, vehicle-road collaboration, and autonomous driving.

Claims

1. A traffic dynamic management and control method based on traffic data dimension reduction reconstruction and multi-dimensional analysis, characterized in that, Comprising the following steps: Step 1: Construct the traffic intersection area model, with the parking line and the extension line of the parking line as the boundary, set the traffic intersection center intersection area as the control area, and set the extended area of the road in multiple directions as the monitoring area; in each monitoring area, set the area flowing out of the monitoring area to the control area as the out area, and set the area flowing into the monitoring area from the control area as the in area; Step 2: Plan the traffic path of the area model: take each out area of the monitoring area as the starting point, connect with the corresponding in area in the left, straight and right directions, and construct the connection relationship diagram of the out area and the in area; Step 3: Extract information from the original data of each area by layering and use matrix and coordinates for dimension reduction reconstruction, including data collection, road condition dimension reduction reconstruction, vehicle condition dimension reduction reconstruction and dimension reduction information storage; Step 4: Extract the layered dimension reduction reconstruction data of each area to form a new coordinate curve or matrix data model, and perform multi-dimensional feature analysis, including: (1) Establish an out area waiting to release vehicle monitoring data model: monitor the vehicle condition matrix in each out area, real-time calculate and extract the number of vehicles waiting to release at a certain time in the matrix and form a new matrix, then establish a coordinate system with time*quantity, generate an out area waiting to release vehicle number monitoring curve for feature analysis; (2) Establish an out area in-transit vehicle flow monitoring data model: monitor the vehicle condition matrix in each out area, real-time calculate the number and position of in-transit vehicle flow and form a new matrix, then establish a coordinate system with time*quantity, generate an out area in-transit vehicle flow monitoring curve for feature analysis; (3) Establish an out area waiting to release vehicle passing time fitting data model: monitor the vehicle condition matrix in each out area, calculate the number and passing time of a batch of waiting to release vehicles in the matrix, then establish a coordinate system with quantity*time, record and fit multiple batches of data to obtain the fitting data curve of the out area waiting to release vehicle passing time for feature analysis; (4) Establish an out area vehicle outflow data model, the model principle is used for control area vehicle inflow data analysis: monitor the vehicle condition matrix in each out area, set the statistical time t, calculate the vehicle outflow number in units of t or accumulate to form a new matrix, then establish a coordinate system with time*quantity, form an out area vehicle outflow data curve for feature analysis; (5) Establish an in area vehicle inflow data model, the model principle is used for control area vehicle outflow data analysis: monitor the vehicle condition matrix in each in area, set the statistical time t, calculate the vehicle inflow number in units of t or accumulate to form a new matrix, then establish a coordinate system with time*quantity, form an in area vehicle inflow data curve for feature analysis; (6) Establish a vehicle speed tracking data model for each area: monitor the vehicle condition matrix in each area, extract the speed data of each vehicle, calculate the average speed of the vehicle, and form a new matrix, then establish a coordinate system with time*speed, real-time generate a vehicle speed data curve for feature analysis; (7) Establishing a vehicle density distribution data model of each region: monitoring the vehicle condition matrix of each region, calculating the vehicle coverage rate in the matrix in real time and forming a new matrix, then establishing a coordinate system with time * coverage rate, generating a vehicle density distribution curve in real time for feature analysis; Step 5: Multi-dimensional feature data mixed calculation: including the calculation and acquisition of out-zone in-transit vehicles and vehicles to be released, accident monitoring in each region, congestion state monitoring in in-zone and control zone, and the calculation and prediction results will provide information data support for traffic dynamic control: Step 5-1: Real-time acquisition or prediction of the number of vehicles to be released in the out area and the number of large vehicle flows in transit using the data models of (1) and (2) in step 4: the number of vehicles to be released in the lane is calculated as , and the state of the vehicle to be released is : ， The number of large vehicle flow vehicles in the lane in the way is calculated as The in-the-way large vehicle flow state is , , Step 5-2: Using the data models in step 4 (6), (7), obtain the vehicle speed characteristics and vehicle density information changing with time in the region, set the density threshold D and the duration threshold P, monitor whether the vehicle speed drops to 0 and the duration of the vehicle density greater than D exceeds P, and monitor the accident in the monitoring zone and the control zone, when the density in the region exceeds D and the duration exceeds P, or the number of vehicles with speed dropping to 0 in the region is greater than or equal to 1, it is determined that a traffic accident has occurred, otherwise no traffic accident has occurred, and the traffic accident state AD is as follows: ， Step 5-3: Using the data models in step 4 (4), (5), (6), (7), obtain the feature information of vehicle inflow and outflow, vehicle speed and vehicle density changing with time in in-zone and control zone, use machine learning algorithm to calculate the congestion state of in-zone and control zone, the control zone specifically uses the data models in step 4 (4), (5), (6), (7), and the in-zone specifically uses the data models in step 4 (5), (6), (7), the calculation method of congestion state of control zone and in-zone is the same; Step 6: Traffic dynamic control, including signal light control: Step 6-1-1: Setting the initial state: setting the initial state of the signal light as all red, red = 1; Step 6-1-2: Determining the passable path: determining the combination of straight and left turn lanes that can be released at the same time according to step 2, red = 0, and determining whether the right turn lane can pass according to the passable data of the combination; Step 6-1-3: Single-lane release index calculation: according to the values or state outputs in steps 5-1, 5-2, and 5-3, the single-lane release index is calculated, and according to the single-lane release index value, it is determined whether to release and the priority of release. The single-lane release index calculation formula is as follows: let In order to monitor the total number of vehicles required for a certain lane in the area, the calculation is as follows: ， Set For special vehicle states, ; Set For in zone congestion state, ; Set To control the congestion state of the area, , Set Current lane abnormal state, ; Let the single-lane release index be Then the expression for evaluating is: , 。 2.The traffic dynamic management and control method based on traffic data dimension reduction reconstruction and multi-dimensional analysis according to claim 1, characterized in that, Step 3 is implemented through the following steps: Step 3-1: Data acquisition: in the monitoring zone and the control zone, using maps, road engineering drawings and cameras, collecting road condition information by road section, including lane length and width, lane line position, sidewalk position, road side entrance and exit, fork, road obstacle, road curvature and slope information; collecting vehicle features, traffic volume, video, picture and radar signal information of vehicle motion state by road section; Step 3-2: Dimensionality reduction reconstruction of road conditions: (1) First, establish a rectangular coordinate system and set the scale, so that the real road conditions and the coordinate system form an exact correspondence; (2) In the coordinate system, set a line segment corresponding to the stop line of the traffic intersection, then take this line segment as the starting point to mark and reconstruct the length and width dimensions of the out area, as well as the lane line position, sidewalk position, road side entrance and exit, fork intersection, road obstacle, road curvature, slope information in the coordinate system; (3) Divide the marked and reconstructed road in the coordinate system into K regions, and m i *n i Grid segmentation, value range m i >0, n i >0, K≥i>0, then store the information in the segmented grid with a matrix; Step 3-3: Dimensionality reduction reconstruction of vehicle condition: (1) Real-time acquisition of video, picture, radar signal information of vehicle condition according to K regions, and then detection, tracking and feature information index extraction of vehicle characteristics, vehicle position and vehicle motion condition are carried out by using image recognition, signal recognition artificial intelligence algorithm, and then m i *n i Matrix data storage is carried out, and thus vehicle condition coordinate graph can be generated in real time in the road condition graph after data superposition, so as to realize visual display of road condition and vehicle condition information; Step 3-4: Dimension reduction information storage: storing the obtained multi-dimensional matrix data of road condition and vehicle condition, if the effective information in the matrix is less, compressing and storing the sparse matrix to save storage space. 3.The traffic dynamic management and control method based on traffic data dimension reduction reconstruction and multi-dimensional analysis according to claim 1, characterized in that, The calculation steps of congestion state of control zone and in-zone in step 5-3 are as follows: Step 5-3-1: Building an original input data set: using the data models in step 4 (4), (5), (6), (7) to obtain the vehicle position feature matrix, vehicle inflow and outflow feature curve and vehicle speed feature matrix, and then defining the congestion state corresponding to the feature as a label; Step 5-3-2: Feature pre-processing: arrange the vehicle position feature matrix in chronological order to form a set of serialized matrix blocks, and then serialize and tile each matrix block to obtain linear features; normalize the vehicle inflow and outflow feature curve; the three pre-processed features are used as input features of the model; Step 5-3-3: Build a deep learning model: (1) build a self-attention mechanism network, input the pre-processed vehicle position feature matrix into the network, and output the linear feature layer result; (2) build a convolutional neural network with residual structure, and finally output a linear layer, input the vehicle speed feature matrix into the network, and output the linear feature layer result; (3) concatenate and merge the normalized vehicle inflow and outflow feature curve with the two linear layer features, input into the fully connected network and activate, and predict the congestion state; Step 5-3-4: Model training: send the data set into the deep learning model in step 5-3-3 for training, use the cross-entropy loss function to calculate the loss of the network model, and during the training process, if the loss difference between the first n Epoch and the last n Epoch is less than a fixed value, n≥1, fixed value>0, then the model training is completed, and the calculation parameters in the model are saved; Step 5-3-5: Congestion state calculation: Real-time acquisition of vehicle position feature matrix, vehicle inflow and outflow feature curve and vehicle speed feature matrix, after preprocessing in step 5-3-2, input into the network to obtain the calculation result The calculation result represents the congestion probability, and the threshold is set ; similarly, when the in area performs congestion state calculation, the calculation result is , and the threshold is set to .

4. The traffic dynamic management and control method based on traffic data dimension reduction reconstruction and multi-dimensional analysis according to claim 1, characterized in that, Step 6 also includes prevention of overflow control, that is, controlling the inflow of vehicles in the in area, wherein By the calculation in step 6, the index of in area is obtained, when there is then the in area section traffic congestion information data is output, and the vehicle flow is controlled by red light.

5. The traffic dynamic management and control method based on traffic data dimension reduction reconstruction and multi-dimensional analysis according to claim 1, characterized in that, The accident monitoring and control is also included in step 6. When there is an accident in a certain lane, the traffic accident information data is outputted through the calculation in step 6. ​ 6.The traffic dynamic management and control method based on traffic data dimension reduction reconstruction and multi-dimensional analysis according to claim 1, characterized in that, In step 6, the traffic congestion control is also included. Through the calculation in step 6, the index of the control area is obtained. When there is , the traffic congestion information data of the control area is output, and the vehicle flow into the corresponding out area is controlled by using the red light.

7. The traffic dynamic management and control method based on traffic data dimension reduction reconstruction and multi-dimensional analysis according to claim 1, characterized in that, Step 6 also includes low-efficiency no-entry control, which calculates the indicators of the out area through step 6, and when the release indicator of a lane is 0 under the green light state, the red light is switched, and other lanes are selected for release.

Images