A method for identifying road network congestion based on discontinuous electric police data

Abstract

The application discloses a method for identifying road network congestion based on discontinuous electric police data, and comprises the following steps: obtaining electric police kiosk photographing data, road network basic information data, electric police kiosk basic data, and obtaining vehicle travel trajectory data; performing electric police kiosk and road network matching based on the electric police kiosk basic data and the road network basic information data; performing complementation on the travel trajectory data of all vehicles based on the road network basic information data and the obtained electric police kiosk and road network matching information; performing full sampling on the travel trajectory of corresponding vehicles in the road network based on the complemented vehicle travel trajectory data, and calculating the average travel speed of road sections; and calculating the road network congestion index of corresponding road sections by using the free flow speed of a road section and the calculated average travel speed, combining with a planning time index, so as to realize road condition grading. The method can accurately complement the vehicle travel path, calculate the average travel speed and congestion index of road sections, and predict future congestion conditions.

Description

Technical Field

[0001] This invention relates to a method for identifying road network congestion based on discontinuous electronic police data, belonging to the field of traffic management technology. Background Technology

[0002] Currently, with the continuous increase in the number of motor vehicles in cities, road network congestion has become a core issue restricting traffic efficiency and exacerbating energy consumption and safety hazards. Intelligent transportation systems place higher demands on real-time identification, accurate classification, and spatiotemporal source tracing of congestion across the entire road network. Existing technologies mostly rely on single data sources such as loop detectors, video checkpoints, single-point GPS, or floating cars, which have shortcomings such as coverage blind spots, poor environmental robustness, data silos, and delayed congestion determination, making it difficult to support road network-level collaborative management and control.

[0003] Patent CN114067564B discloses a comprehensive traffic condition monitoring method based on YOLO, which includes a vehicle identification and traffic flow statistics scheme based on surveillance video and YOLO object detection. It determines congestion status through bounding box counting and speed estimation. However, this scheme has significant shortcomings: it only achieves single-point / single-section detection, fails to construct road segment associations and road network topology, and cannot identify congestion propagation or regional congestion. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a method for identifying road network congestion based on discontinuous electronic police data. It can accurately complete the vehicle travel path, calculate the average travel speed and congestion index of the road segment, and predict the future congestion situation when the data sampling is discontinuous.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is: a method for identifying road network congestion based on discontinuous electronic alarm data, the method comprising the following steps:

[0006] S01: Acquire traffic camera checkpoint photo data, road network basic information data, and traffic camera checkpoint basic data. Clean the traffic camera checkpoint photo data, break down vehicle travel, and obtain vehicle travel trajectory data according to the time sequence of different vehicles and checkpoint photos.

[0007] S02: Match the electronic police checkpoints with the road network based on the basic data of the electronic police checkpoints and the basic road network information data;

[0008] S03: Based on the aforementioned basic road network information data and the obtained electronic police checkpoint and road network matching information, complete the travel trajectory data of all vehicles;

[0009] S04: Based on the completed vehicle travel trajectory data, the travel trajectories of the corresponding vehicles in the road network are fully sampled and the average travel speed of the road segment is calculated.

[0010] S05: Obtain the 95th percentile travel time of all vehicles passing through a certain road segment and calculate its planning time index. Using the free-flow speed of a certain road segment and the calculated average travel speed, combined with the planning time index, calculate the road network congestion index of the corresponding road segment to achieve road condition classification.

[0011] S06: Based on The hybrid model predicts the road network congestion index for future periods.

[0012] Furthermore, in step S01, the traffic camera data is cleaned and the vehicle trips are broken down, including: removing time-based abnormal data and splitting the trajectories with time intervals greater than a set number of trips according to rules.

[0013] Furthermore, step S01 specifically includes:

[0014] S011: Group the electronic police checkpoint photo data according to the encrypted vehicle identification code, and sort them according to the photo time to form a separate electronic police checkpoint photo sequence for each vehicle.

[0015] S012: Calculate the vehicle In the and The time interval between each checkpoint being photographed And calculate according to the formula respectively. , , , , ;

[0016]

[0017]

[0018]

[0019]

[0020] in, For vehicles In the and The time interval between each checkpoint being photographed; For vehicles The third quartile of the time interval data indicates that 75% of the time interval data... Less than this value; For vehicles The third quartile of the time interval data indicates that 25% of the time interval data... Less than this value; Indicates vehicle The time interval data is distributed within the middle 50% range; For vehicles Determine the upper limit of outliers; For vehicles The lower limit for determining outliers;

[0021] S013: When the vehicle In the and The time interval between each checkpoint being photographed Greater than If it is assumed that the second trip has begun, the original travel trajectory is split into two trajectories to complete the trip splitting;

[0022] When the vehicle In the and The time interval between each checkpoint being photographed Less than If a photo is taken continuously at the same electronic police checkpoint, it is considered abnormal photo data, and only the data of the first photo taken at the electronic police checkpoint will be retained.

[0023] S014: According to the trip splitting rules in step S013, the trips of all vehicles are split, and several trip trajectory data are finally obtained according to the time sequence of different vehicles and checkpoint photos.

[0024] Furthermore, in step S02, all electronic traffic enforcement checkpoints are first classified into five types: elevated entrances and exits and ramps, one-way elevated road sections, two-way elevated road sections, ground-level intersections, and ground-level road sections. Elevated entrances and exits and ramps, and ground-level intersections are directly matched using existing electronic traffic enforcement checkpoints and the latitude and longitude of the road section nodes. One-way elevated road sections are first matched with the linear sections of one-way elevated road sections based on their latitude and longitude coordinates. After a successful match, the one-way elevated road section... The checkpoints are matched with the start and end points of the matched road segments; ground road segments are first matched with linear road segments according to their latitude and longitude coordinates. After a successful match, the checkpoints of the ground road segments are matched with the start and end points of the matched road segments; elevated road segments are first matched with linear road segments according to their latitude and longitude coordinates. After a successful match, the segments are classified according to the direction of the entrance lanes and then matched with the start and end points of the matched road segments.

[0025] Furthermore, in step S03, the completion step specifically involves: extracting the average travel time of road segments based on basic road network information data; constructing a weighted directed graph based on the average travel time of road segments; generating K alternative routes using the Yen algorithm; calculating the average travel time of the alternative routes based on the weighted directed graph; calculating the time error by comparing the time interval between photos taken at two traffic enforcement checkpoints with the average travel time of all alternative routes; selecting the route with the smallest time error as the completion route to obtain the complete vehicle trajectory.

[0026] Furthermore, the specific steps to complete the task are as follows:

[0027] S031: Construct a weighted directed graph G;

[0028] Generate a weighted directed graph Node set :

[0029]

[0030]

[0031] in Represents a weighted directed graph Node coordinates in This refers to the latitude and longitude of the node.

[0032] Generate a weighted directed graph The set of directed edges :

[0033]

[0034] Each directed edge satisfies:

[0035]

[0036] in express ;

[0037] Weights for generating directed edges :

[0038] Extract from road network infrastructure data Average travel time and road segment length Therefore:

[0039]

[0040]

[0041] in Represents a directed edge The corresponding weights;

[0042] Final weighted directed graph Represented as:

[0043]

[0044] in Define road network nodes; Define a permissible directed edge; Define the average formation time of a directed edge, and represent the weight of the directed edge.

[0045] S032: Before solving using Yen's algorithm There are several acyclic shortest paths, i.e., alternative paths; first, a preliminary explanation of the symbol conventions is given:

[0046] Input: Directed weighted graph Source ,end Number of target paths ;

[0047] Simple path: from the source node To the target node path There are no duplicate nodes in the array, that is... ;

[0048] Travel time and route length for any route: The total travel time and length of the route are defined as follows:

[0049]

[0050]

[0051] Branch node: During the iteration of the Yen algorithm, for the currently known branch node... Shortest path Select one of the nodes As a branch node, this node is used to generate a new path that is the same as the previous path in the prefix but different in the suffix.

[0052] Prefix paths and suffix paths: For paths Branch nodes in ,definition:

[0053] Prefix path:

[0054]

[0055] Suffix path: from arrive The path must be such that it does not contain other nodes in the prefix path, thus avoiding loops;

[0056] The first shortest path calculation: Dijkstra's algorithm is used to calculate the path from... arrive shortest path Initialize the result set Initialize the candidate path set ;

[0057] Iterative generation of the first Path : For the known first Path Perform the following steps:

[0058] Branch node traversal: for path Each node :set up From arrive Prefix path;

[0059] Temporary graph construction: from the original graph Copy to get a temporary image ;delete For all paths in the array, their prefix AND The edge that is identical to and whose next edge is the same; from Remove from Except All external nodes should be checked to prevent loops.

[0060] Suffix path calculation: In Calculate from branch node To the target node shortest path ;

[0061] If they exist, they are combined into a complete candidate path:

[0062]

[0063] in This indicates a path concatenation operation;

[0064] Candidate path storage: If Not in Then add it. ;

[0065] Candidate path filtering: Choose the path with the shortest length. ,join in As ;from Delete the path;

[0066] Termination condition: If it has been generated If there is one path, then the process ends; if... Empty and not yet generated If there is no path, then stop and return to the current path. ;

[0067] S033: Calculate the average travel time for each alternative path using the weights of a weighted directed graph:

[0068]

[0069] in Indicates the first The average travel time for this route Indicates the first The first path One node;

[0070] S034: Calculation time error Utilizing time error Select the path that best reflects reality as the completed path, and calculate the average speed of all segments along that path. :

[0071]

[0072]

[0073] in The path for this road segment calculated using a weighted directed graph. Average travel time; The start and end points of the path to be completed for this segment. and The time interval captured at the electronic police checkpoint; The completed path for this segment The sum of the lengths of all road segments in the middle;

[0074] Will The data is then added between the two original traffic camera locations to form a complete trajectory.

[0075] Specifically, this embodiment makes the following assumption: the average speed of a vehicle is the same across all road segments in the path completed between two traffic camera images. Step S04 is as follows:

[0076] S041: Extract a specific road segment Road segment length And the average speed of all vehicles passing through the road segment calculated in step S034;

[0077] S042: Total traffic volume utilizing this section of road and the Average speed of vehicles on this road section This allows you to obtain the average travel speed for the current road segment. for:

[0078]

[0079] Furthermore, step S05 specifically includes:

[0080] S051: Based on the road segment obtained in step S034, the information obtained regarding the route passed by a certain vehicle. average speed of the car Utilizing the road section Road segment length This will give you the route a vehicle has taken. travel time :

[0081]

[0082] S052: Obtain information about a specific road segment. 95% percentile travel time for all vehicles And calculate its planning time index. :

[0083]

[0084] in For a certain section of road The free flow velocity;

[0085] S053: Obtain a specific road segment free flow velocity ratio :

[0086]

[0087] S054: Obtain the comprehensive congestion score for a certain road segment. The road network congestion index is:

[0088]

[0089] And graded according to scores: This indicates that the road conditions are good. This indicates moderate congestion. This indicates severe traffic congestion.

[0090] Furthermore, step S06 specifically includes:

[0091] S061: Constructing a data-driven segment transmission model ,include:

[0092] (a) Transmission model in traditional road sections Based on this, a closed-loop online learning and open-loop prediction mechanism is introduced;

[0093] (b) During the closed-loop online learning phase, the simulated road network is gradually calibrated using real-time traffic flow observation data, and correction parameters are introduced. Minimize the difference between the observed flow rate and the calculated flow rate;

[0094] (c) In the open-loop prediction stage, prediction is made based on the calibrated simulated road network. It is assumed that the boundary demand within the prediction time window is consistent with the previous time. The prediction results are obtained through mechanism simulation.

[0095] (d) A rolling time window mechanism is adopted to achieve continuous forecasting;

[0096] S062: Constructing a spatiotemporal depth tensor neural network ,include:

[0097] (a) Construct a three-dimensional depth tensor from traffic status data according to the time dimension, spatial dimension, and road segment dimension. ;

[0098] (b) The time dimension as Axis representation A historical step in time, spatial dimension The axis represents the closest point to the target road segment. Each road segment, depth dimension as The axis represents the total number of road segments. ;

[0099] (c) Input the 3D depth tensor X into the convolutional layer, pooling layer and fully connected layer to extract the spatiotemporal feature vector. ;

[0100] (d) Extract non-traffic flow input features as external feature vectors ;

[0101] (e) Introduce a historical periodic information vector of the congestion index corresponding to the predicted target time. ;

[0102] (f) will The deep learning prediction output is obtained by fusing the data according to preset weights. ;

[0103] S063: Fusion prediction, including:

[0104] (a) will Introduced as an additional input vector frame;

[0105] (b) Perform fusion prediction output as follows:

[0106]

[0107] in, for Predict weights, For deep learning to predict weights, For feature weighting parameters, For activation function, For Hadama's son;

[0108] (c) The backpropagation algorithm is used to jointly train all parameters and output the final short-term traffic flow prediction results.

[0109] Furthermore, the calibration process during the closed-loop online learning phase is achieved by minimizing the following difference:

[0110]

[0111] in, For road section exist Observational flow rate at any time For road section To the section Transmission traffic, For the road section A collection of connected upstream road segments;

[0112] And / or adopt a rolling time window mechanism to achieve continuous prediction, including: after each prediction task is completed, the time window is rolled forward by a fixed step, and the closed-loop online learning phase is re-entered using new traffic flow observation data;

[0113] and / or The depth dimension of the model's three-dimensional depth tensor X changes dynamically according to the road network scale, and the spatial dimension... The value is determined by the Euclidean distance matrix;

[0114] and / or external feature vectors Includes date, day of the week, holiday information, and weather conditions;

[0115] and / or historical cycle information vector This includes traffic flow status information for the day, week, or month preceding the predicted time.

[0116] By adopting the above technical solution, the present invention has the following beneficial effects:

[0117] 1. Strong data adaptability: Applicable to sparse and discontinuous electronic traffic enforcement checkpoint data; 2. High path completion accuracy: Introduces weighted directed graphs and the Yen algorithm, compares multiple paths, and minimizes time error to select the true path; 3. More scientific congestion assessment: Constructs a congestion index by combining average road segment speed and PTI to distinguish different congestion levels; 4. Strong road segment congestion prediction capability: Combines mechanistic models and deep learning models. This enables the prediction of congestion trends. Attached Figure Description

[0118] Figure 1 This is a block diagram illustrating the principle of the method of the present invention. Detailed Implementation

[0119] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0120] like Figure 1 As shown, a method for identifying road network congestion based on discontinuous electronic traffic enforcement data includes the following steps:

[0121] S01: Acquire traffic camera checkpoint photo data, road network basic information data, and traffic camera checkpoint basic data. Clean the traffic camera checkpoint photo data, break down vehicle travel, and obtain vehicle travel trajectory data according to the time sequence of different vehicles and checkpoint photos.

[0122] S02: Match the electronic police checkpoints with the road network based on the basic data of the electronic police checkpoints and the basic road network information data;

[0123] S03: Based on the aforementioned basic road network information data and the obtained electronic police checkpoint and road network matching information, complete the travel trajectory data of all vehicles;

[0124] S04: Based on the completed vehicle travel trajectory data, the travel trajectories of the corresponding vehicles in the road network are fully sampled and the average travel speed of the road segment is calculated.

[0125] S05: Obtain the 95th percentile travel time of all vehicles passing through a certain road segment and calculate its planning time index. Using the free-flow speed of a certain road segment and the calculated average travel speed, combined with the planning time index, calculate the road network congestion index of the corresponding road segment to achieve road condition classification.

[0126] S06: Based on The hybrid model predicts the road network congestion index for future periods.

[0127] Furthermore, in step S01, the traffic camera data is cleaned and the vehicle trips are broken down, including: removing time-based abnormal data and splitting the trajectories with time intervals greater than a set number of trips according to rules.

[0128] Specifically, this embodiment uses the time interval between two consecutive captures by an electronic police checkpoint as the judgment criterion. When the time interval is too long, it is considered that two trips have occurred, and path splitting is performed at this point; when the time interval is too short, it is considered that it has been captured twice consecutively by a certain checkpoint, and only the information of the first capture is retained. Therefore, step S01 is specifically as follows:

[0129] S011: Group the electronic police checkpoint photo data according to the encrypted vehicle identification code, and sort them according to the photo time to form a separate electronic police checkpoint photo sequence for each vehicle.

[0130] S012: Calculate the vehicle In the and The time interval between each checkpoint being photographed And calculate according to the formula respectively. , , , , ;

[0131]

[0132]

[0133]

[0134]

[0135] in, For vehicles In the and The time interval between each checkpoint being photographed; For vehicles The third quartile of the time interval data indicates that 75% of the time interval data... Less than this value; For vehicles The third quartile of the time interval data indicates that 25% of the time interval data... Less than this value; Indicates vehicle The time interval data is distributed within the middle 50% range; For vehicles Determine the upper limit of outliers; For vehicles The lower limit for determining outliers;

[0136] S013: When the vehicle In the and The time interval between each checkpoint being photographed Greater than If it is assumed that the second trip has begun, the original travel trajectory is split into two trajectories to complete the trip splitting;

[0137] When the vehicle In the and The time interval between each checkpoint being photographed Less than If a photo is taken continuously at the same electronic police checkpoint, it is considered abnormal photo data, and only the data of the first photo taken at the electronic police checkpoint will be retained.

[0138] S014: According to the trip splitting rules in step S013, the trips of all vehicles are split, and several trip trajectory data are finally obtained according to the time sequence of different vehicles and checkpoint photos.

[0139] Specifically, in step S02, all electronic traffic enforcement checkpoints are first classified into five types: elevated entrances and exits and ramps, one-way elevated road sections, two-way elevated road sections, ground-level intersections, and ground-level road sections. Elevated entrances and exits and ramps, and ground-level intersections are directly matched using existing electronic traffic enforcement checkpoints and the latitude and longitude of the road section nodes. One-way elevated road sections are first matched with the linear sections of one-way elevated road sections based on their latitude and longitude coordinates. After a successful match, the one-way elevated road section... The checkpoints are matched with the start and end points of the matched road segments; ground road segments are first matched with linear road segments according to their latitude and longitude coordinates. After a successful match, the checkpoints of the ground road segments are matched with the start and end points of the matched road segments; elevated road segments are first matched with linear road segments according to their latitude and longitude coordinates. After a successful match, the segments are classified according to the direction of the entrance lanes and then matched with the start and end points of the matched road segments.

[0140] Specifically, in step S03, the completion step is as follows: extract the average travel time of road segments based on the road network basic information data, construct a weighted directed graph based on the average travel time of road segments; generate K alternative paths using the Yen algorithm; calculate the average travel time of the alternative paths based on the weighted directed graph; calculate the time error by comparing the time interval between the photos taken by two traffic police checkpoints with the average travel time of all alternative paths, select the path with the smallest time error as the completion path, and obtain the complete vehicle trajectory.

[0141] Specifically, the completion steps are as follows:

[0142] S031: Construct a weighted directed graph G;

[0143] Generate a weighted directed graph Node set :

[0144]

[0145]

[0146] in Represents a weighted directed graph Node coordinates in This refers to the latitude and longitude of the node.

[0147] Generate a weighted directed graph The set of directed edges :

[0148]

[0149] Each directed edge satisfies:

[0150]

[0151] in express ;

[0152] Weights for generating directed edges :

[0153] Extract from road network infrastructure data Average travel time and road segment length Therefore:

[0154]

[0155]

[0156] in Represents a directed edge The corresponding weights;

[0157] Final weighted directed graph Represented as:

[0158]

[0159] in Define road network nodes; Define a permissible directed edge; Define the average formation time of a directed edge, and represent the weight of the directed edge.

[0160] S032: Before solving using Yen's algorithm There are several acyclic shortest paths, i.e., alternative paths; first, a preliminary explanation of the symbol conventions is given:

[0161] Input: Directed weighted graph Source ,end Number of target paths ;

[0162] Simple path: from the source node To the target node path There are no duplicate nodes in the array, that is... ;

[0163] Travel time and route length for any route: The total travel time and length of the route are defined as follows:

[0164]

[0165]

[0166] Branch node: During the iteration of the Yen algorithm, for the currently known branch node... Shortest path Select one of the nodes As a branch node, this node is used to generate a new path that is the same as the previous path in the prefix but different in the suffix.

[0167] Prefix paths and suffix paths: For paths Branch nodes in ,definition:

[0168] Prefix path:

[0169]

[0170] Suffix path: from arrive The path must be such that it does not contain other nodes in the prefix path, thus avoiding loops;

[0171] The first shortest path calculation: Dijkstra's algorithm is used to calculate the path from... arrive shortest path Initialize the result set Initialize the candidate path set ;

[0172] Iterative generation of the first Path : For the known first Path Perform the following steps:

[0173] Branch node traversal: for path Each node :set up From arrive Prefix path;

[0174] Temporary graph construction: from the original graph Copy to get a temporary image ;delete For all paths in the array, their prefix AND The edge that is identical to and whose next edge is the same; from Remove from Except All external nodes should be checked to prevent loops.

[0175] Suffix path calculation: In Calculate from branch node To the target node shortest path ;

[0176] If they exist, they are combined into a complete candidate path:

[0177]

[0178] in This indicates a path concatenation operation;

[0179] Candidate path storage: If Not in Then add it. ;

[0180] Candidate path filtering: Choose the path with the shortest length. ,join in As ;from Delete the path;

[0181] Termination condition: If it has been generated If there is one path, then the process ends; if... Empty and not yet generated If there is no path, then stop and return to the current path. ;

[0182] S033: Calculate the average travel time for each alternative path using the weights of a weighted directed graph:

[0183]

[0184] in Indicates the first The average travel time for this route Indicates the first The first path One node;

[0185] S034: Calculation time error Utilizing time error Select the path that best reflects reality as the completed path, and calculate the average speed of all segments along that path. :

[0186]

[0187]

[0188] in The path for this road segment calculated using a weighted directed graph. Average travel time; The start and end points of the path to be completed for this segment. and The time interval captured at the electronic police checkpoint; The completed path for this segment The sum of the lengths of all road segments.

[0189] Specifically, step S04 is as follows:

[0190] S041: Extract a specific road segment Road segment length And the average speed of all vehicles passing through the road segment calculated in step S034;

[0191] S042: Total traffic volume utilizing this section of road and the Average speed of vehicles on this road section This allows you to obtain the average travel speed for the current road segment. for:

[0192]

[0193] Furthermore, step S05 specifically includes:

[0194] S051: Based on the road segment obtained in step S034, the information obtained regarding the route passed by a certain vehicle. average speed of the car Utilizing the road section Road segment length This will give you the route a vehicle has taken. travel time :

[0195]

[0196] S052: Obtain information about a specific road segment. 95% percentile travel time for all vehicles And calculate its planning time index. :

[0197]

[0198] in For a certain section of road The free flow velocity;

[0199] S053: Obtain a specific road segment free flow velocity ratio :

[0200]

[0201] S054: Obtain the comprehensive congestion score for a certain road segment. The road network congestion index is:

[0202]

[0203] And graded according to scores: This indicates that the road conditions are good. This indicates moderate congestion. This indicates severe traffic congestion.

[0204] Specifically, step S06 is as follows:

[0205] S061: Constructing a data-driven segment transmission model ,include:

[0206] (a) Transmission model in traditional road sections Based on this, a closed-loop online learning and open-loop prediction mechanism is introduced;

[0207] (b) During the closed-loop online learning phase, the simulated road network is gradually calibrated using real-time traffic flow observation data, and correction parameters are introduced. Minimize the difference between the observed flow rate and the calculated flow rate;

[0208] (c) In the open-loop prediction stage, prediction is made based on the calibrated simulated road network. It is assumed that the boundary demand within the prediction time window is consistent with the previous time. The prediction results are obtained through mechanism simulation.

[0209] (d) A rolling time window mechanism is adopted to achieve continuous forecasting;

[0210] S062: Constructing a spatiotemporal depth tensor neural network ,include:

[0211] (a) Construct a three-dimensional depth tensor from traffic status data according to the time dimension, spatial dimension, and road segment dimension. ;

[0212] (b) The time dimension as Axis representation A historical step in time, spatial dimension The axis represents the closest point to the target road segment. Each road segment, depth dimension as The axis represents the total number of road segments. ;

[0213] (c) Input the 3D depth tensor X into the convolutional layer, pooling layer and fully connected layer to extract the spatiotemporal feature vector. ;

[0214] (d) Extract non-traffic flow input features as external feature vectors ;

[0215] (e) Introduce a historical periodic information vector of the congestion index corresponding to the predicted target time. ;

[0216] (f) will The deep learning prediction output is obtained by fusing the data according to preset weights. ;

[0217] S063: Fusion prediction, including:

[0218] (a) will Introduced as an additional input vector frame;

[0219] (b) Perform fusion prediction output as follows:

[0220]

[0221] in, for Predict weights, For deep learning to predict weights, For feature weighting parameters, For activation function, For Hadama's son;

[0222] (c) The backpropagation algorithm is used to jointly train all parameters and output the final short-term traffic flow prediction results.

[0223] Specifically, the calibration process in the closed-loop online learning phase is achieved by minimizing the following difference:

[0224]

[0225] in, For road section exist Observational flow rate at any time For road section To the section Transmission traffic, For the road section A collection of connected upstream road segments;

[0226] The rolling time window mechanism enables continuous prediction by: after each prediction task, the time window is rolled forward by a fixed step, and the system re-enters the closed-loop online learning phase using new traffic flow observation data;

[0227] The depth dimension of the model's three-dimensional depth tensor X changes dynamically according to the road network scale, and the spatial dimension... The value is determined by the Euclidean distance matrix;

[0228] External feature vectors Includes date, day of the week, holiday information, and weather conditions;

[0229] Historical cycle information vector This includes traffic flow status information for the day, week, or month preceding the predicted time.

[0230] The specific embodiments described above further illustrate the technical problems, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying road network congestion based on discontinuous electronic traffic enforcement data, characterized in that, The method includes the following steps: S01: Acquire traffic camera checkpoint photo data, road network basic information data, and traffic camera checkpoint basic data. Clean the traffic camera checkpoint photo data, break down vehicle travel, and obtain vehicle travel trajectory data according to the time sequence of different vehicles and checkpoint photos. S02: Match the electronic police checkpoints with the road network based on the basic data of the electronic police checkpoints and the basic road network information data; S03: Based on the aforementioned basic road network information data and the obtained electronic police checkpoint and road network matching information, complete the travel trajectory data of all vehicles; S04: Based on the completed vehicle travel trajectory data, the travel trajectories of the corresponding vehicles in the road network are fully sampled and the average travel speed of the road segment is calculated. S05: Obtain the 95th percentile travel time of all vehicles passing through a certain road segment and calculate its planning time index. Using the free-flow speed of a certain road segment and the calculated average travel speed, combined with the planning time index, calculate the road network congestion index of the corresponding road segment to achieve road condition classification. S06: Based on The hybrid model predicts the road network congestion index for future periods.

2. The method according to claim 1, characterized in that, In step S01, the traffic camera data is cleaned and the vehicle trips are segmented, including: removing time-based abnormal data and segmenting trajectories with time intervals greater than a set number of trips according to rules.

3. The method according to claim 2, characterized in that, Step S01 is as follows: S011: Group the electronic police checkpoint photo data according to the encrypted vehicle identification code, and sort them according to the photo time to form a separate electronic police checkpoint photo sequence for each vehicle. S012: Calculate the vehicle In the and The time interval between each checkpoint being photographed And calculate according to the formula respectively. , , , , ; ； ； ； ； in, For vehicles In the and The time interval between each checkpoint being photographed; For vehicles The third quartile of the time interval data indicates that 75% of the time interval data... Less than this value; For vehicles The third quartile of the time interval data indicates that 25% of the time interval data... Less than this value; Indicates vehicle The time interval data is distributed within the middle 50% range; For vehicles Determine the upper limit of outliers; For vehicles The lower limit for determining outliers; S013: When the vehicle In the and The time interval between each checkpoint being photographed Greater than If it is assumed that the second trip has begun, the original travel trajectory is split into two trajectories to complete the trip splitting; When the vehicle In the and The time interval between each checkpoint being photographed Less than If a photo is taken continuously at the same electronic police checkpoint, it is considered abnormal photo data, and only the data of the first photo taken at the electronic police checkpoint will be retained. S014: According to the trip splitting rules in step S013, the trips of all vehicles are split, and several trip trajectory data are finally obtained according to the time sequence of different vehicles and checkpoint photos.

4. The method according to claim 1, characterized in that, In step S02, all electronic traffic enforcement checkpoints are first classified into five types: elevated entrances and exits and ramps, one-way elevated road sections, two-way elevated road sections, ground intersections, and ground road sections. Elevated entrances and exits and ramps, and ground intersections are directly matched with existing electronic traffic enforcement checkpoints and road segment nodes based on their latitude and longitude. One-way elevated road sections are first matched with their linear segments based on their latitude and longitude coordinates. After a successful match, the checkpoints for the one-way elevated road section are matched with the start and end points of the matched segment. Ground road sections are first matched with their linear segments based on their latitude and longitude coordinates. After a successful match, the checkpoints for the ground road section are matched with the start and end points of the matched segment. Two-way elevated road sections are first matched with their linear segments based on their latitude and longitude coordinates. After a successful match, the sections are categorized according to the direction of the approach lane and then matched with the start and end points of the matched segment.

5. The method according to claim 1, characterized in that, In step S03, the completion step specifically involves: extracting the average travel time of road segments based on basic road network information data; constructing a weighted directed graph based on the average travel time of road segments; generating K alternative paths using the Yen algorithm; calculating the average travel time of the alternative paths based on the weighted directed graph; calculating the time error by comparing the time interval between photos taken at two traffic enforcement checkpoints with the average travel time of all alternative paths; selecting the path with the smallest time error as the completion path to obtain the complete vehicle trajectory.

6. The method according to claim 5, characterized in that, The specific steps for completing the task are as follows: S031: Construct a weighted directed graph G; Generate a weighted directed graph Node set : ； ； in Represents a weighted directed graph Node coordinates in This refers to the latitude and longitude of the node. Generate a weighted directed graph The set of directed edges : ； Each directed edge satisfies: ； in express ; Weights for generating directed edges : Extract from road network infrastructure data Average travel time and road segment length Therefore: ； ； in Represents a directed edge The corresponding weights; Final weighted directed graph Represented as: ； in Define road network nodes; Define a permissible directed edge; Define the average formation time of a directed edge, and represent the weight of a directed edge; S032: Before solving using Yen's algorithm There are several acyclic shortest paths, i.e., alternative paths; first, a preliminary explanation of the symbol conventions is given: Input: Directed weighted graph Source ,end Number of target paths ; Simple path: from the source node To the target node path There are no duplicate nodes in the array, that is... ; Travel time and route length for any route: The total travel time and length of the route are defined as follows: ； ； Branch node: During the iteration of the Yen algorithm, for the currently known branch node... Shortest path Select one of the nodes As a branch node, this node is used to generate a new path that is the same as the previous path in the prefix but different in the suffix. Prefix paths and suffix paths: For paths Branch nodes in ,definition: Prefix path: ； Suffix path: from arrive The path must be such that it does not contain other nodes in the prefix path, thus avoiding loops; The first shortest path calculation: Dijkstra's algorithm is used to calculate the path from... arrive shortest path Initialize the result set Initialize the candidate path set ; Iterative generation of the first Path : For the known first Path Perform the following steps: Branch node traversal: for path Each node :set up From arrive Prefix path; Temporary graph construction: from the original graph Copy to get a temporary image ;delete For all paths in the array, their prefix AND The edge that is identical to and whose next edge is the same; from Remove from Except All external nodes should be checked to prevent loops. Suffix path calculation: In Calculate from branch node To the target node shortest path ; If they exist, they are combined into a complete candidate path: ； in This indicates a path concatenation operation; Candidate path storage: If Not in Then add it. ; Candidate path filtering: Choose the path with the shortest length. ,join in As ;from Delete the path; Termination condition: If it has been generated If there is one path, then the process ends; if... Empty and not yet generated If there is no path, then stop and return to the current path. ; S033: Calculate the average travel time for each alternative path using the weights of a weighted directed graph: ； in Indicates the first The average travel time for this route Indicates the first The first path One node; S034: Calculation time error Utilizing time error Select the path that best reflects reality as the completed path, and calculate the average speed of all segments along that path. : ； ； in The path for this road segment calculated using a weighted directed graph. Average travel time; The start and end points of the path to be completed for this segment. and The time interval captured at the electronic police checkpoint; The completed path for this segment The sum of the lengths of all road segments.

7. The method according to claim 6, characterized in that, Step S04 is as follows: S041: Extract a specific road segment Road segment length And the average speed of all vehicles passing through the road segment calculated in step S034; S042: Total traffic volume utilizing this section of road and the Average speed of vehicles on this road section This allows you to obtain the average travel speed for the current road segment. for: 。 8. The method according to claim 7, characterized in that, Step S05 is as follows: S051: Based on the road segment obtained in step S034, the information obtained regarding the route passed by a certain vehicle. average speed of the car Utilizing the road section Road segment length This will give you the route a vehicle has taken. travel time : ； S052: Obtain information about a specific road segment. 95% percentile travel time for all vehicles And calculate its planning time index. : ； in For a certain section of road The free flow velocity; S053: Obtain a specific road segment free flow velocity ratio : ； S054: Obtain the comprehensive congestion score for a certain road segment. The road network congestion index is: ； And graded according to scores: This indicates that the road conditions are good. This indicates moderate congestion. This indicates severe traffic congestion.

9. The method according to claim 1, characterized in that, Step S06 is as follows: S061: Constructing a data-driven segment transmission model ,include: (a) Transmission model in traditional road sections Based on this, a closed-loop online learning and open-loop prediction mechanism is introduced; (b) During the closed-loop online learning phase, the simulated road network is gradually calibrated using real-time traffic flow observation data, and correction parameters are introduced. Minimize the difference between the observed flow rate and the calculated flow rate; (c) In the open-loop prediction stage, prediction is made based on the calibrated simulated road network. It is assumed that the boundary demand within the prediction time window is consistent with the previous time. The prediction results are obtained through mechanism simulation. (d) A rolling time window mechanism is adopted to achieve continuous forecasting; S062: Constructing a spatiotemporal depth tensor neural network ,include: (a) Construct a three-dimensional depth tensor from traffic status data according to the time dimension, spatial dimension, and road segment dimension. ; (b) The time dimension as Axis representation A historical step in time, spatial dimension The axis represents the closest point to the target road segment. Each road segment, depth dimension as The axis represents the total number of road segments. ; (c) Input the 3D depth tensor X into the convolutional layer, pooling layer and fully connected layer to extract the spatiotemporal feature vector. ; (d) Extract non-traffic flow input features as external feature vectors ; (e) Introduce a historical periodic information vector of the congestion index corresponding to the predicted target time. ; (f) will The deep learning prediction output is obtained by fusing the data according to preset weights. ; S063: Fusion prediction, including: (a) will Introduced as an additional input vector frame; (b) Perform fusion prediction output as follows: ； in, for Predict weights, For deep learning to predict weights, For feature weighting parameters, For activation function, For Hadama's son; (c) The backpropagation algorithm is used to jointly train all parameters and output the final short-term traffic flow prediction results.

10. The method according to claim 9, characterized in that, The calibration process during the closed-loop online learning phase is achieved by minimizing the following difference: ； in, For road section exist Observational flow rate at any time For road section To the section Transmission traffic, For the road section A collection of connected upstream road segments; And / or adopt a rolling time window mechanism to achieve continuous prediction, including: after each prediction task is completed, the time window is rolled forward by a fixed step, and the closed-loop online learning phase is re-entered using new traffic flow observation data; and / or The depth dimension of the model's three-dimensional depth tensor X changes dynamically according to the road network scale, and the spatial dimension... The value is determined by the Euclidean distance matrix; and / or external feature vectors Includes date, day of the week, holiday information, and weather conditions; and / or historical cycle information vector This includes traffic flow status information for the day, week, or month preceding the predicted time.

Images