A method for real-time monitoring of traffic congestion based on multi-source data fusion from the Internet of Things

Abstract

This invention discloses a real-time traffic congestion monitoring method based on multi-source data fusion of the Internet of Things, belonging to the field of traffic congestion monitoring technology. The method includes the following steps: First, setting up geomagnetic coil detectors, roadside microwave radars, floating car GPS data acquisition terminals, and intersection video surveillance cameras on each monitoring road segment in the urban road network. Traffic status data for each monitoring road segment is collected according to their respective preset collection cycles. All types of data are unified into the same analysis time window for analysis to obtain standard data records for each monitoring road segment. The credibility of each data source is evaluated in real time using a cross-validation matrix method, and fusion weights are dynamically calculated for weighted fusion. Then, the fused speed and density are mapped onto a two-dimensional plane, and the traffic flow status is determined through trajectory movement direction analysis. Simultaneously, the speed spatial gradient between adjacent road segments is calculated to detect congestion boundaries and mark positions for tracking movement.

Description

Technical Field

[0001] This invention belongs to the field of traffic congestion monitoring technology, specifically a method for real-time monitoring of traffic congestion based on multi-source data fusion from the Internet of Things. Background Technology

[0002] With the acceleration of urbanization and the continuous growth of motor vehicle ownership, urban road traffic congestion has become increasingly serious, becoming a prominent problem that restricts urban development and affects residents' travel experience. Accurate and timely monitoring of traffic congestion is of great significance for traffic management departments to formulate diversion measures and guide travelers to choose reasonable routes. At present, various traffic detection devices have been deployed in urban road networks, including geomagnetic coil detectors buried under the road surface, microwave radar installed on the roadside, GPS terminals mounted on operating vehicles, and video surveillance cameras at intersections. These devices can collect traffic flow data from different angles, providing a rich source of data for traffic condition monitoring.

[0003] However, existing traffic congestion monitoring technologies typically determine traffic conditions by comparing the current speed or density value with preset thresholds. For example, a speed below 20 km / h is considered congested, while a speed above 40 km / h is considered smooth. This approach fails to reflect the direction of traffic flow evolution, i.e., whether it is worsening or improving. Misjudgments are prone to occur during the transition period of traffic flow state transitions. For instance, a road segment that is rapidly recovering to smooth traffic may still be classified as a congested transition state, or a road segment that is about to deteriorate into congestion may be misclassified as smooth. Based on this, a real-time traffic congestion status monitoring method based on multi-source data fusion from the Internet of Things is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, so as to solve the problems mentioned in the background art.

[0005] A method for real-time monitoring of traffic congestion based on multi-source data fusion from the Internet of Things includes:

[0006] Step 1: Set up geomagnetic coil detectors, roadside microwave radars, floating car GPS data acquisition terminals, and intersection video surveillance cameras on each monitoring section of the urban road network. Collect traffic status data for each monitoring section according to their respective preset collection cycles. Unify all types of data into the same analysis time window for analysis to obtain standard data records for each monitoring section. Each standard data record includes coil speed, coil flow, coil occupancy rate, equivalent density, radar speed, radar flow, GPS speed, GPS sample size, video queue length, and number of vehicles in the video.

[0007] Step 2: Pair the coil speed, radar speed and GPS speed in the standard data records of each monitoring section to construct a cross-verification matrix. Based on the cross-verification matrix, calculate the credibility score of each data source for coil speed, radar speed and GPS speed in different analysis time windows for each monitoring section.

[0008] Step 3: Based on the credibility scores of each data source within different analysis time windows for each monitored road segment, the speed values ​​of each data source are weighted and fused to obtain the fusion speed and fusion density of each monitored road segment within each analysis time window;

[0009] Step 4: Map the fusion speed and fusion density onto a two-dimensional plane with density as the horizontal axis and speed as the vertical axis to obtain the trajectory. Determine the trajectory movement direction based on the fusion speed and fusion density of the current analysis time window and the previous analysis time window. Distinguish the current traffic flow status of each monitored road segment and obtain the traffic flow status label of each monitored road segment.

[0010] Step 5: Obtain the upstream and downstream connection relationships of each monitored road segment in the road network, calculate the speed spatial gradient of each pair of adjacent road segments with upstream and downstream connection relationships, detect and track the congestion boundary between adjacent road segments, and obtain the calibrated position and moving speed of the congestion boundary.

[0011] Step Six: Determine the congestion level label based on the traffic flow status label and speed deviation of each road segment. At the same time, use the video queue length and the number of vehicles in the video to assist in the verification of the congestion level label, and integrate and output the real-time monitoring results of each monitored road segment.

[0012] As a further aspect of this invention: a unified analysis time window length of 5 minutes is set, and data with a collection timestamp greater than or equal to the start time of the window and less than the end time of the window are included in the corresponding analysis time window; for data from geomagnetic coil detectors, the cross-sectional traffic flow, average cross-sectional speed, and cross-sectional time occupancy rate are directly used as coil flow, coil speed, and coil occupancy rate, respectively; for data from roadside microwave radar, the average radar detection speed of each data point within the window is used as radar speed, and the sum of radar detection flow is used as radar flow; for floating car GPS data, the average speed of each operating vehicle within the window is first calculated, and then the average speed of all operating vehicles is averaged to obtain the GPS speed; the percentage value of coil occupancy rate is divided by 100, multiplied by 1000, and then divided by the preset average effective vehicle length to obtain the equivalent density.

[0013] As a further aspect of the present invention, the specific method for obtaining the reliability scores of each data source—coil velocity, radar velocity, and GPS velocity—within different analysis time windows for each monitored road segment is as follows:

[0014] For each monitored road segment within each analysis time window, a 3x3 cross-validation matrix is ​​constructed using three data sources: coil speed, radar speed, and GPS speed from the standard data records. The first data source is coil speed, the second is radar speed, and the third is GPS speed. The element in the i-th row and j-th column of the matrix represents the cross-validation consistency score between the i-th and j-th data sources. After the cross-validation matrix is ​​constructed, the sum of all off-diagonal elements in each data category's row is divided by the number of off-diagonal elements to obtain the consistency mean for the corresponding data. This yields the consistency mean Ai for coil speed, radar speed, and GPS speed. For coil speed and radar speed, a basic integrity score of 1 is given if the data is normal, and a score of 0 is given if the data is missing or abnormal. For GPS speed... The basic integrity score is equal to the minimum of the quotient obtained by dividing the GPS sample size by the sample size threshold and 1. The sample size threshold is equal to the number of lanes in the corresponding road segment multiplied by 3. Thus, the basic integrity scores Bi for coil speed, radar speed and GPS speed are obtained. The row consistency mean Ai of each data source is multiplied by 0.7 to obtain the first product. Then, the basic integrity scores Bi of each data source are multiplied by 0.3 to obtain the second product. Finally, the first product and the second product are added together to obtain the credibility score Qi of each data source. For each monitored road segment, the same analysis is performed on the three data sources of coil speed, radar speed and GPS speed in the standard data records in each analysis time window. Thus, the credibility scores of each data source of coil speed, radar speed and GPS speed in different analysis time windows of each monitored road segment are obtained.

[0015] As a further aspect of the present invention: the specific method for obtaining the mutual verification consistency score between the i-th type of data source and the j-th type of data source in the matrix is ​​as follows:

[0016] For the velocity value Vi of the i-th data source and the velocity value Vj of the j-th data source in the matrix, first subtract Vj from Vi and take the absolute value to obtain the deviation value Dij. Then divide Dij by the maximum value between Vi and Vj to obtain the relative deviation Rij. Finally, subtract Rij from 1 to obtain the mutual verification consistency score Mij. If the result of the Mij calculation is less than 0, it is taken as 0. The diagonal elements of the matrix are always taken as 1.

[0017] As a further aspect of the present invention, the specific method for obtaining the fusion speed and fusion density of each monitored road segment within each analysis time window is as follows:

[0018] For each monitoring segment within a single analysis time window, the quotient obtained by dividing the reliability score Qi of that data source by the sum of the reliability scores of all valid data sources participating in the fusion within the current window is used as the fusion weight Wi of that data source within that analysis time window for that monitoring segment. The product of the coil speed multiplied by the coil fusion weight, the product of the radar speed multiplied by the radar fusion weight, and the product of the GPS speed multiplied by the GPS fusion weight within that window is used as the fusion speed Vf within that analysis time window. When coil data is valid within that window, the equivalent density is directly used as the fusion density Kf within that window. When coil data is missing, the quotient obtained by multiplying the radar flow rate within that window by 12 and then dividing by the fusion speed is used as the fusion density Kf within that window. When both coil and radar data are missing, the fusion density within that window is marked as missing. The reliability scores of each data source for coil speed, radar speed, and GPS speed within different analysis time windows for each monitoring segment are analyzed in the same way to obtain the fusion speed and fusion density of each monitoring segment within each analysis time window.

[0019] As a further aspect of the present invention, the specific method for obtaining the traffic flow status tags of each monitored road segment is as follows:

[0020] When the fusion density is less than 0.5 times the critical density, it is considered a free-flow state; when the fusion density is greater than or equal to 0.5 times the critical density but less than the critical density, if the trajectory movement direction is the direction of deterioration, stability, abnormal decay, or free growth, it is considered a transitional flow state; if the trajectory movement direction is the direction of improvement, it is considered a free-flow state; when the fusion density is greater than or equal to the critical density and the fusion speed is greater than 0.2 times the free flow speed, it is considered a synchronous congestion state; when the fusion density is greater than or equal to the critical density and the fusion speed is less than or equal to 0.2 times the free flow speed, it is considered a severe congestion state.

[0021] As a further aspect of the present invention, the specific method for determining the trajectory movement direction is as follows:

[0022] When the absolute value of the density change is less than 2 vehicles per kilometer and the absolute value of the speed change is less than 1 kilometer per hour, it is considered a stable state; when the density change is greater than 0 and the speed change is less than 0, it is considered a deteriorating trend; when the density change is less than 0 and the speed change is greater than 0, it is considered an improving trend; when the density change is greater than 0 and the speed change is greater than 0, it is considered a free growth trend; when the density change is less than 0 and the speed change is less than 0, it is considered an abnormal decay trend.

[0023] As a further aspect of the present invention, the specific method for obtaining the calibrated position and moving speed of the congestion boundary is as follows:

[0024] For adjacent road segments with upstream and downstream connections, the speed difference is obtained by subtracting the fusion speed of the downstream road segment from the fusion speed of the upstream road segment. This speed difference is then divided by the distance between the center points of the two road segments, and the quotient is used as the speed spatial gradient of the pair of adjacent road segments. When the absolute value of the speed spatial gradient exceeds a gradient threshold, a congestion boundary is determined to exist between the pair of adjacent road segments. If the traffic flow status label of one side of the pair of adjacent road segments is synchronous congestion or severe congestion, while the traffic flow status label of the other side is free flow or transitional flow, then the congestion boundary is determined to be a congestion expansion type boundary. The fusion speed of the upstream road segment is labeled Vu, the fusion speed of the downstream road segment is labeled Vd, and the distance between the center points of the two road segments is labeled L. The distance is calculated using D = Vu ÷ (Vu + Vd). The offset distance D from the center point of the upstream road segment to the congestion boundary is calculated using ×L. The location reached by measuring this offset distance from the geographical coordinates of the upstream road segment's center point downstream is the designated position of the congestion boundary. The offset distance from the congestion boundary to the center point of the upstream road segment in the current analysis time window is subtracted from the offset distance in the previous analysis time window to obtain the boundary displacement. The absolute value of the boundary displacement is divided by 1000 to convert it to kilometers, then divided by the analysis time window length, and multiplied by 3600 to convert it to kilometers per hour to obtain the moving speed of the congestion boundary. When a road segment is adjacent to the non-congested side of a congestion expansion boundary, and the current traffic flow status label for that road segment is transitional, and the congestion boundary is moving towards that road segment, an impending congestion warning is added to that road segment.

[0025] As a further aspect of the present invention, the specific method for determining the congestion level label and performing auxiliary verification is as follows:

[0026] The speed ratio is obtained by dividing the fused speed of each monitored road segment in the current analysis time window by the free-flow speed. The speed deviation is then calculated by subtracting the speed ratio from 1. When the speed ratio is greater than 1, the speed deviation is set to 0. If the traffic flow status label is "free-flowing," the congestion level label is determined to be "smooth." If the traffic flow status label is "transitional flowing" and the speed deviation is less than 0.5, it is determined to be "basically smooth." If the speed deviation is greater than or equal to 0.5, it is determined to be "mildly congested." If the traffic flow status label is "synchronous congestion" and the speed deviation is less than 0.5, it is determined to be "mildly congested." If the speed deviation is greater than or equal to 0.5 and less than 0.7, it is determined to be "moderately congested." If the speed deviation is greater than or equal to 0.7, it is determined to be "severely congested." If the traffic flow status label is "severely congested," it is directly determined to be "severely congested." Regarding the fusion density... For road segments marked as missing, if the speed ratio is greater than or equal to 0.8, it is considered smooth; if it is greater than or equal to 0.5 and less than 0.8, it is considered basically smooth; if it is greater than or equal to 0.3 and less than 0.5, it is considered lightly congested; if it is greater than or equal to 0.15 and less than 0.3, it is considered moderately congested; and if it is less than 0.15, it is considered severely congested. When a road segment is marked with an impending congestion warning, if the congestion level label is smooth, it is adjusted to basically smooth; if it is basically smooth, it is adjusted to lightly congested. For road segments with a congestion level label of smooth or basically smooth, if the video queue length exceeds 0.5 times the length of the road segment, the congestion level label is raised by one level. For road segments with a congestion level label of moderately or severely congested, if the video queue length is less than 20 meters and the average number of vehicles in the video is less than 5, the congestion level label is lowered by one level.

[0027] The specific method for integrating and outputting the real-time monitoring results of each monitored road segment is as follows:

[0028] The output includes the segment number of each monitored road segment, the timestamp of the current analysis time window, the fusion speed, the fusion density, the traffic flow status label, the congestion level label, and the speed deviation value. For adjacent road segments with congestion boundaries, the output also includes the coordinates of the congestion boundary's location, the congestion boundary type, and the congestion boundary's moving speed. For road segments with an upcoming congestion warning sign, the output also includes the warning sign and the estimated congestion arrival time. The estimated congestion arrival time is equal to the distance from the center point of the road segment to the congestion boundary's location divided by the congestion boundary's moving speed.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] (1) In this invention, the loop speed, radar speed and GPS speed in the standard data records of each monitored road segment are paired and cross-verified to construct a cross-verification matrix. The cross-verification consistency score between each data source is calculated and the row consistency mean is obtained. At the same time, the row consistency mean is multiplied by 0.7 and the integrity score is multiplied by 0.3 to obtain the credibility score of each data source. The credibility score is then normalized into a fusion weight for weighted fusion. The fusion weight is calculated in real time through the comprehensive evaluation of cross-verification consistency and integrity. When a data source has abnormal data that is seriously inconsistent with other data sources due to equipment failure or environmental interference, its credibility score and fusion weight will be automatically reduced. This effectively avoids the problem that the fixed weight fusion method in the prior art cannot adapt to changes in data quality and causes abnormal data to pollute the fusion results. It ensures that the fusion speed can accurately reflect the real traffic status of the road segment.

[0031] (2) This invention maps the fusion speed and fusion density onto a two-dimensional plane with density as the horizontal axis and speed as the vertical axis, calculates the speed change and density change between the current analysis time window and the previous analysis time window, and determines the trajectory movement direction as a deterioration direction, improvement direction, free growth direction, abnormal decay direction or stable state based on the positive and negative combination of the change. It combines the density range where the current fusion density is located and the trajectory movement direction to comprehensively determine the traffic flow status label. By introducing the trajectory movement direction, the traffic status determination is extended from a static threshold determination at a single moment to a dynamic determination combined with the evolution trend. It can distinguish traffic states with the same current density but different evolution directions, such as a transitional state that is deteriorating and a transitional state that is improving, and provide more forward-looking state evolution information for traffic management decisions.

[0032] (3) This invention detects congestion boundaries by calculating the spatial gradient of speed between each pair of adjacent road segments with upstream and downstream connections in the road network, uses the fusion speed ratio of upstream and downstream road segments to perform linear interpolation calibration of the congestion boundary position, tracks the boundary displacement in a continuous time window to calculate the direction and speed of movement, and feeds back the congestion boundary propagation information to the traffic flow status determination to add warning marks to the road segments that are about to be affected by congestion. This invention achieves accurate positioning, dynamic tracking and early warning of congestion boundaries in the road network, so that traffic management departments not only know which road segments are congested, but also know which direction the congestion is spreading and how fast it is spreading. At the same time, the warning feedback mechanism makes each step form an information closed loop rather than a simple serial processing, which improves the system's ability to perceive and predict the spatial propagation of congestion.

[0033] (4) This invention determines the congestion level by cross-combining traffic flow status labels and speed deviation, then adjusts the level of road sections that are about to be affected by congestion according to the warning signs, and finally uses video queue length and vehicle number to independently assist in the verification and correction of the determined congestion level. When the queue length does not match the smooth flow status, the level is adjusted up; when the queue length and vehicle number do not match the congestion status, the level is adjusted down. This establishes a multi-layer congestion classification mechanism of traffic flow status acceleration deviation plus warning signs plus video verification, which can discover and correct the problem of overestimation or underestimation of the level caused by the systematic deviation of the fusion speed, ensuring that the final output congestion level is more in line with the actual road conditions, and improving the accuracy and reliability of congestion level determination. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the method framework structure of the present invention. Detailed Implementation

[0035] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Example 1: Please refer to Figure 1 This application provides a method for real-time monitoring of traffic congestion based on multi-source data fusion from the Internet of Things, including the following steps:

[0037] Step 1: Geomagnetic coil detectors, roadside microwave radars, floating car GPS data acquisition terminals, and intersection video surveillance cameras are installed on various monitoring road sections within the urban road network. Traffic status data for each monitoring road section is collected according to their respective preset collection cycles. All types of data are then analyzed within the same analysis time window to obtain standard data records for each monitoring road section. The specific method is as follows:

[0038] Here, the monitored road segment refers to a section of road between two adjacent intersections in the urban road network. The length of the road segment is usually between 200 meters and 2000 meters. This embodiment takes an urban arterial road containing five consecutive monitored road segments numbered R01 to R05 as an example for illustration.

[0039] The geomagnetic coil detector is buried under the road surface. When a vehicle passes over the coil, it causes a change in the magnetic field, which is then detected. It collects three parameters: cross-sectional traffic flow, average cross-sectional speed, and cross-sectional time occupancy rate. The cross-sectional traffic flow is the number of vehicles passing through the coil cross-section in five minutes, expressed as vehicles per five minutes. The average cross-sectional speed is the average speed of vehicles passing through the cross-section, expressed as kilometers per hour. The cross-sectional time occupancy rate is the percentage of time that vehicles occupy the coil detection area in five minutes, expressed as a percentage, with a value range of 0 to 100.

[0040] The roadside microwave radar is installed on a roadside pole in the middle of the road section. It detects vehicle movement by emitting microwave signals and receiving reflected signals from vehicles. It collects two parameters: radar detection speed and radar detection flow rate. The radar detection speed is the average speed of vehicles in the coverage area in kilometers per hour, and the radar detection flow rate is the number of vehicles passing through the detection area per minute in vehicles per minute. The data reporting cycle of the roadside microwave radar is 1 minute.

[0041] The floating car GPS terminal is installed in taxis and ride-hailing vehicles, collectively referred to as operating vehicles. It is used to collect and report the instantaneous speed and latitude and longitude coordinates of the operating vehicles every 30 seconds. The road segment to which each GPS data belongs is determined by matching the vehicle's latitude and longitude coordinates with the geographical range of the road segment, i.e., determining whether the coordinates fall within the spatial range of the road segment.

[0042] The intersection video surveillance camera is installed at the intersection at both ends of the road segment. It outputs the length of the vehicle queue from the stop line upstream every minute through video image analysis algorithm as the video detection queue length. At the same time, it uses the number of vehicles identified in the picture as the video detection vehicle number. The data processing cycle of the intersection video surveillance camera is 1 minute.

[0043] A unified analysis time window length of 5 minutes (300 seconds) is set. Starting from the real-time data acquisition time, multiple analysis time windows are continuously divided forward at 5-minute intervals. For example, if the real-time data acquisition time is 10:00, the first window is from 10:00 to 10:05, the second window is from 10:05 to 10:10, and so on, to obtain multiple analysis time windows. Then, based on the acquisition timestamp of each piece of raw data for each type of data, it is determined which analysis time window the data falls into. That is, data whose acquisition timestamp is greater than or equal to the start time of the window and less than the end time of the window are assigned to the corresponding analysis time window, thereby aligning all types of data into a unified analysis time window. Here, the basis for choosing 5 minutes as the window length is that this duration is consistent with the data reporting cycle of the geomagnetic coil detector, and at the same time, enough floating car GPS samples can be accumulated within 5 minutes to calculate a statistically reliable average road speed.

[0044] After aligning various data types to a unified analysis time window, for data from geomagnetic coil detectors, since the reporting cycle is exactly equal to the analysis time window length (5 minutes), and each window contains only one coil data point, the cross-sectional traffic flow, average cross-sectional speed, and cross-sectional time occupancy rate from that data point are directly used as the coil flow, coil speed, and coil occupancy rate for the corresponding window, respectively. For data from roadside microwave radar, the reporting cycle is 1 minute, meaning there are 5 radar data points within each 5-minute window. The average radar detection speed value of the 5 data points within the same window is used as the radar speed for the corresponding window, and the sum of the radar detection flow values ​​of the 5 data points within the same window is used as the radar flow for the corresponding window. For data from floating car GPS data acquisition terminals, the reporting cycle is 30 seconds, meaning there are a maximum of 10 data points per vehicle within each 5-minute window. First, each operating vehicle within the same window on the same road segment is analyzed. The arithmetic mean of all reported instantaneous speed values ​​is used to obtain the average speed of a single vehicle within the window. Then, the arithmetic mean of the average speeds of all operating vehicles on the same road segment within the window is used as the GPS speed of the corresponding road segment within the window. At the same time, the number of operating vehicles reporting data for the road segment within the window is recorded as the GPS sample size. When the GPS sample size of a certain road segment within a certain window is 0, that is, no operating vehicles report data, the GPS speed is marked as missing. In subsequent steps, the fusion calculation of that road segment and that window does not include the GPS data source. For the data of the intersection video surveillance camera, the processing cycle is 1 minute, that is, there are 5 video data in each 5-minute window. The maximum value of the video detection queue length of the 5 data in the same window is used as the video queue length in the corresponding window. At the same time, the average number of video detection vehicles of the 5 data in the same window is used as the average number of video vehicles in the corresponding window.

[0045] Then, the coil occupancy percentage is converted to a decimal value by dividing it by 100, and then multiplied by 1000 and divided by the preset average effective vehicle length. The quotient is then converted into an equivalent density for constructing a two-dimensional planar graph of speed and density, in units of vehicles per kilometer. The preset average effective vehicle length is 6 meters, which is based on the fact that the body length of a small vehicle on urban roads is about 4.5 meters, plus the length of the geomagnetic coil detection area is about 1.5 meters, totaling about 6 meters.

[0046] Then, the reasonable range for setting the speed parameters, namely the coil speed, radar speed, and GPS speed, is 0 to 150 kilometers per hour, and the reasonable range for the coil occupancy rate is 0 to 100%. Parameter values ​​that exceed the reasonable range are marked as abnormal values ​​and will not be included in the calculation of subsequent steps.

[0047] Finally, using the timestamp of the analysis time window as an index, the various parameter values ​​of the same road segment within the same window are combined into the standard data record of that road segment in that window. Each standard data record contains 10 parameters, including coil speed, coil flow, coil occupancy rate, equivalent density, radar speed, radar flow, GPS speed, GPS sample size, video queue length, and number of video vehicles; thus, the standard data record of each monitored road segment is obtained.

[0048] Step Two: Perform pairwise cross-verification of coil speeds, radar speeds, and GPS speeds in the standard data records for each monitoring section, construct a cross-verification matrix, and calculate the credibility score of each data source for coil speeds, radar speeds, and GPS speeds within different analysis time windows for each monitoring section based on the cross-verification matrix. The specific method is as follows:

[0049] For each monitored road segment within each analysis time window, a 3x3 cross-validation matrix is ​​constructed using three data sources: coil speed, radar speed, and GPS speed from the standard data records. The first data source is coil speed, the second is radar speed, and the third is GPS speed. The element in the i-th row and j-th column of the matrix represents the cross-validation consistency score between the i-th and j-th data sources. When a data source is marked as missing or abnormal within the current window, that data source does not participate in matrix construction, and the matrix dimension is reduced to 2 rows and 2 columns. If only one valid data source remains, matrix m cannot be constructed. The average row consistency of this data source is set to 0.5 by default, where i and j are both positive integers, ranging from 1 to 3.

[0050] The specific calculation method for the mutual verification consistency score is as follows: For the speed value Vi of the i-th data source and the speed value Vj of the j-th data source, first subtract Vj from Vi and take the absolute value to obtain the deviation value Dij. Then, divide Dij by the maximum value between Vi and Vj to obtain the relative deviation Rij. Finally, subtract Rij from 1 to obtain the mutual verification consistency score Mij. If the calculated result of Mij is less than 0, it is taken as 0. The value range of Mij is 0 to 1, where 1 indicates that the two data sources are completely consistent and 0 indicates that they are seriously inconsistent. Dividing by the maximum value of the two rather than the average value is to avoid the relative deviation being unreasonably amplified due to the denominator being too small in low-speed scenarios. The elements on the diagonal of the matrix, i.e., the mutual verification scores between the data source and itself, are always 1, thus obtaining the corresponding mutual verification consistency score values ​​at each element.

[0051] After the mutual verification matrix is ​​constructed, the sum of all off-diagonal elements in the row of each type of data is divided by the number of off-diagonal elements to obtain the consistency mean of the corresponding data. In this way, the consistency mean Ai of coil speed, radar speed and GPS speed is obtained. The closer Ai is to 1, the more likely it is to be accurate as the corresponding data source is highly consistent with other data sources. The lower Ai is, the more likely there is a deviation as it is generally inconsistent with other data sources.

[0052] Then, for coil speed and radar speed, if the data is normal, the basic integrity score of the corresponding data source is 1; if it is missing or abnormal, the basic integrity score of the corresponding data source is 0. The minimum value between the quotient obtained by dividing the GPS sample size by the sample size threshold and 1 is taken as the basic integrity score of the GPS speed. The sample size threshold is equal to the number of lanes in the corresponding road segment multiplied by 3, that is, at least 3 operating vehicles are required for each lane to reliably represent the speed level of that lane. Thus, the basic integrity scores Bi of coil speed, radar speed, and GPS speed are obtained. Finally, the mean consistency Ai of coil speed, radar speed, and GPS speed are combined with the basic integrity score Bi to obtain the credibility score Qi of each data source for coil speed, radar speed, and GPS speed. The specific method is as follows:

[0053] First, the mean row consistency Ai of each data source is multiplied by 0.7 to obtain the first product. Then, the basic integrity score Bi of each data source is multiplied by 0.3 to obtain the second product. Finally, the first product and the second product are added together to obtain the reliability score Qi of each data source. Mutual verification consistency is given a high weight of 0.7 because it is the core factor in judging data accuracy, while integrity is given a weight of 0.3 as a basic guarantee. For each monitoring section, the same analysis is performed on the three data sources of coil speed, radar speed and GPS speed in the standard data records within each analysis time window, thereby obtaining the reliability score of each data source of coil speed, radar speed and GPS speed within different analysis time windows for each monitoring section.

[0054] For each monitored road segment within each analysis time window, the three independent speed data sources—coil speed, radar speed, and GPS speed—are paired for comparison. A cross-validation matrix is ​​constructed by calculating the cross-validation consistency score between each pair of data sources. The average of the off-diagonal elements in each data source's row is then used to obtain the row consistency mean, which reflects the overall consistency between that data source and other data sources. Finally, the row consistency mean and the basic integrity score, reflecting data integrity, are weighted at 0.7 and 0.3 respectively to calculate the reliability score of that data source. The beneficial effect of this step is the establishment of a real-time cross-validation quality assessment mechanism between data sources. When a sensor experiences drift or data offset, its detection speed consistency with other sensors automatically decreases, thus lowering its reliability score. Unreliable data sources can be automatically identified without relying on manual inspection, providing a dynamic and adaptive quality basis for subsequent weighted fusion.

[0055] Step 3: Based on the credibility scores of the coil speed, radar speed, and GPS speed data sources within different analysis time windows for each monitored road segment, the speed values ​​from each data source are weighted and fused to obtain the fused speed and fused density for each monitored road segment within each analysis time window. The specific method is as follows:

[0056] For each monitoring segment within a single analysis time window, the quotient obtained by dividing the reliability score Qi of that data source by the sum of the reliability scores of all valid data sources participating in the fusion within the current window is used as the fusion weight Wi of that data source within that analysis time window for that monitoring segment. The sum of the fusion weights of the coil speed, radar speed, and GPS speed within that analysis time window is equal to 1. The fusion speed Vf within that analysis time window is equal to the product of the coil speed multiplied by the coil fusion weight, plus the product of the radar speed multiplied by the radar fusion weight, plus the product of the GPS speed multiplied by the GPS fusion weight. The sum of the products corresponding to each valid data source is the fusion speed. When a data source is missing or abnormal, that data source does not participate in the above summation calculation.

[0057] Meanwhile, when coil data is valid within the window, the equivalent density is directly used as the fusion density Kf within that window. When coil data is missing, the quotient obtained by multiplying the radar flow rate within the window by 12 and then dividing by the fusion speed is used as the fusion density Kf within the window. Multiplying by 12 converts 5 minutes per vehicle to vehicles per hour. When both coil and radar data are missing, the fusion density within the window is marked as missing. The reliability scores of the coil speed, radar speed, and GPS speed data from different analysis time windows for each monitored road segment are used to weight and fuse the speed values ​​from each data source to obtain the fusion speed and fusion density of each monitored road segment within each analysis time window.

[0058] The credibility scores of each data source obtained in step two are normalized and used as fusion weights. Data sources with higher credibility receive greater weights in the fusion calculation, while those with lower credibility have their weights automatically reduced. The fusion speed is then obtained by weighted summation of the velocity values ​​from each data source using these weights. Simultaneously, the fusion density is determined based on the availability of coil data or radar traffic data. The beneficial effect of this step is that it achieves adaptive weighted fusion driven by data quality. The fusion weights are entirely determined by the consistency relationship between the data in the current window, without relying on pre-set fixed weights or historical statistical parameters. When the quality of a data source declines, the system automatically reduces its fusion weight, ensuring that the fusion result always converges towards higher-quality data sources, thus improving the accuracy of the fusion speed and the system's fault tolerance.

[0059] Step 4: Obtain the fusion speed and fusion density of each adjacent two analysis time windows in each monitored road segment. Map the fusion speed and fusion density onto a two-dimensional plane with density as the horizontal axis and speed as the vertical axis to obtain the trajectory. Determine the trajectory movement direction based on the fusion speed and fusion density of the current and previous analysis time windows in each monitored road segment. Identify the current traffic flow status of each monitored road segment to obtain the traffic flow status label for each monitored road segment. The specific method is as follows:

[0060] For a single monitored road segment, a two-dimensional plane is first constructed with vehicle density as the horizontal axis and vehicle speed as the vertical axis. On this plane, the fusion speed and fusion density of each analysis time window are used to form a data point. The data points of consecutive windows are connected to form the trajectory of the monitored road segment unit.

[0061] The difference between the fusion speed of the current analysis time window and the previous analysis time window of the monitored road segment is obtained as the speed change, and the difference between the fusion density of the current analysis time window and the previous analysis time window is obtained as the density change.

[0062] When the absolute value of the density change is less than 2 vehicles per kilometer and the absolute value of the speed change is less than 1 kilometer per hour, the trajectory movement direction of the traffic flow state of the monitored road segment is determined to be in a stable state, that is, the trajectory has not undergone meaningful movement; if the stable state condition is not met, the trajectory movement direction is determined according to the following rules.

[0063] When the density change is greater than 0 and the speed change is less than 0, it indicates that the vehicle density of the monitored road segment is increasing and the vehicle speed is decreasing, meaning there are more and more vehicles on the road and their speed is getting slower. The trajectory movement direction is determined to be the deteriorating direction, indicating that the traffic flow condition of the monitored road segment is worsening. When the density change is less than 0 and the speed change is greater than 0, it indicates that the vehicle density of the monitored road segment is decreasing and the vehicle speed is increasing, meaning there are fewer vehicles on the road and their speed is recovering. The trajectory movement direction is determined to be the improving direction, indicating that the traffic flow condition of the monitored road segment is improving. When the density change is greater than 0 and the speed change is greater than 0, it indicates that both the vehicle density and vehicle speed of the monitored road segment are increasing, meaning there are more vehicles on the road but their speed is still increasing. The trajectory movement direction is determined to be the free growth direction. When the density change is less than 0 and the speed change is less than 0, it indicates that the vehicle density and speed of the monitored road segment are decreasing simultaneously, meaning that the number of vehicles on the road is decreasing but the speed is also decreasing. The trajectory movement direction is determined to be an abnormal decay direction, indicating that the monitored road segment is in the initial stage of traffic flow collapse or is affected by external factors. When the density change is equal to 0 and the speed change is not equal to 0, or when the speed change is equal to 0 and the density change is not equal to 0, and the above stable state conditions are not met, if the speed change is less than 0, the trajectory movement direction is determined to be a deteriorating direction; if the speed change is greater than 0, the trajectory movement direction is determined to be an improving direction; if the speed change is equal to 0, the trajectory movement direction is determined to be a stable state.

[0064] For each monitored road segment, the historical average speed from 2:00 AM to 5:00 AM is obtained from historical data as the free-flow speed of each monitored road segment. The free-flow speed refers to the average speed that vehicles can reach when the traffic flow on that road segment is extremely sparse. At the same time, a preset critical density is set for each monitored road segment. The preset critical density is between 80 and 120 vehicles per kilometer. The critical density refers to the density inflection point where the traffic flow on that road segment changes from free flow to congested flow, which is determined by the inflection area of ​​the speed vs. density scatter plot in the historical data.

[0065] Traffic flow is categorized into four states: free-flowing, transitional, synchronous congestion, and severe congestion. Free-flowing means vehicles can freely choose their speed without interfering with each other. Transitional flow means vehicles begin to influence each other and their speeds are limited, but congestion has not yet formed. Synchronous congestion means vehicles are forced to move in sync with the vehicle in front, forming a dense traffic flow but still moving slowly. Severe congestion means vehicles are essentially stationary, moving in stop-and-go traffic. The specific criteria for determining these states are as follows:

[0066] When the fusion density of the current analysis time window is less than 0.5 times the critical density of the road segment, the traffic flow status label of the monitored road segment is determined to be free-flowing. At this time, the vehicle density is low enough that there is no risk of congestion, and the free-flowing status is determined regardless of the trajectory movement direction. When the fusion density of the current analysis time window is greater than or equal to 0.5 times the critical density of the road segment but less than the critical density of the road segment, if the trajectory movement direction is deteriorating, stable, or abnormally decaying, the traffic flow status label of the monitored road segment is determined to be transitional. If the trajectory movement direction is improving, the traffic flow status label of the monitored road segment is determined to be free-flowing, because the traffic flow is leaving the congested area and recovering towards a smooth flow. If the direction of free growth is indicated, the traffic flow status label for the monitored road segment is determined to be transitional flow, because although the speed is still increasing, the density has entered the transition range and there is a risk of saturation. If the fused density of the current analysis time window is greater than or equal to the critical density of the road segment and the fused speed of the current analysis time window is greater than 0.2 times the free flow speed of the road segment, the traffic flow status label for the monitored road segment is determined to be synchronous congestion, indicating that the density has exceeded the critical value but vehicles are still moving slowly. If the fused density of the current analysis time window is greater than or equal to the critical density of the road segment and the fused speed of the current analysis time window is less than or equal to 0.2 times the free flow speed of the road segment, the traffic flow status label for the monitored road segment is determined to be severe congestion, indicating that vehicles are basically at a standstill.

[0067] The same criteria are applied to each monitored road segment within each analysis time window to obtain the traffic flow status label for each monitored road segment.

[0068] The fusion velocity and fusion density of continuous analysis time windows are mapped onto a two-dimensional plane with density as the horizontal axis and velocity as the vertical axis to form a motion trajectory. The trajectory's movement direction is determined by calculating the positive and negative combinations of density and velocity changes between the current and previous windows, categorized into five types: deterioration direction, improvement direction, free growth direction, abnormal decay direction, and stable state. The trajectory's movement direction is then combined with density partitions based on critical density division to determine the traffic flow status label. The beneficial effect of this step is that it achieves direction-sensitive traffic flow status identification, distinguishing between deteriorating and improving trends within the same density range and assigning different status labels. This avoids the problem of traditional static area division methods misclassifying improving road segments as congested or deteriorating road segments as unobstructed during traffic flow state transition periods, making the status identification results more closely reflect the actual dynamic evolution of traffic flow.

[0069] Step 5: Obtain the upstream and downstream connectivity of each monitored road segment in the road network. Based on the fused speed and traffic flow status labels of each monitored road segment, calculate the spatial speed gradient for each pair of adjacent road segments with upstream and downstream connectivity in the road network. Simultaneously, detect and track the congestion boundaries between adjacent road segments to obtain the calibrated location and moving speed of the congestion boundaries. The specific method is as follows:

[0070] Here, the congestion boundary refers to the dividing position between congested and non-congested road segments in the road network space. This dividing position will move along the road direction over time, and its direction and speed of movement reflect the spread of congestion, that is, in which direction the congestion is expanding and how fast it is expanding. The upstream and downstream connection relationship refers to the connection relationship between each monitored road segment in the road network, that is, which road segment's exit connects to which road segment's entrance. This relationship is pre-configured in the system.

[0071] Based on the pre-configured upstream and downstream connections of each monitored road segment in the road network, the speed spatial gradient is calculated for each pair of adjacent road segments with upstream and downstream connections. For the upstream and downstream road segments in the adjacent road segments with upstream and downstream connections, the speed difference is obtained by subtracting the fusion speed of the downstream road segment from the fusion speed of the upstream road segment. The quotient obtained by dividing the speed difference by the distance between the center points of the two road segments is used as the speed spatial gradient of the pair of adjacent road segments. The distance between the center points of the two road segments is in kilometers, and the speed spatial gradient is in kilometers per hour per kilometer.

[0072] When the speed spatial gradient is negative and its absolute value exceeds the gradient threshold, it indicates a sharp increase in speed from upstream to downstream, meaning the upstream is congested while the downstream is unobstructed. This suggests that the congestion boundary is located between two road segments and the congestion is propagating from downstream to upstream. When the speed spatial gradient is positive and its absolute value exceeds the gradient threshold, it indicates a sharp decrease in speed from upstream to downstream, meaning the upstream is unobstructed while the downstream is congested. The gradient threshold is set at 15 km / h. When the absolute value of the speed spatial gradient exceeds this threshold, a congestion boundary is determined to exist between the adjacent road segments. Simultaneously, the type of congestion boundary is determined based on the traffic flow status labels on both sides of the adjacent road segments: if the traffic flow status label on one side of the adjacent road segments is the same... If the traffic flow status label on the other side of the road is labeled as free-flowing or transitional flow, then the congestion boundary is determined to be a congestion expansion type boundary, meaning there is a clear state difference between the congested area and the non-congested area. In other cases, the congestion boundary is determined to be a general type boundary, and only its position and movement speed are recorded, but no congestion expansion label is attached, and the warning feedback in step four is not triggered. Here, the basis for selecting 15 as the gradient threshold is that the speed difference corresponding to the absolute value of the gradient of 15 between two adjacent road segments with a center distance of about 0.5 kilometers is about 7.5 kilometers per hour. This speed difference rarely occurs in normal traffic flow and usually means that there is a qualitative difference in the traffic status of the two road segments.

[0073] The detected congestion boundary is located. The distance between the center points of the two road segments is in meters. First, the fusion speed of the upstream road segment and the fusion speed of the downstream road segment are added together to obtain the sum of the fusion speeds of the two road segments. Then, the fusion speed of the upstream road segment is divided by the sum of the fusion speeds of the two road segments to obtain the speed ratio of the upstream road segment. Finally, the speed ratio of the upstream road segment is multiplied by the distance between the center points of the two road segments to obtain the offset distance of the congestion boundary from the center point of the upstream road segment. Since the geographical coordinates of the center point of each monitored road segment and the road direction are pre-configured in the system, the road position reached by measuring the offset distance from the geographical coordinates of the center point of the upstream road segment along the road direction towards the downstream road segment is the location of the congestion boundary of the pair of adjacent road segments. The calculation formula is expressed as follows: Let the fusion speed of the upstream road segment be Vu, the fusion speed of the downstream road segment be Vd, and the distance between the center points of the two road segments be L. Then, the formula for calculating the offset distance D of the congestion boundary from the center point of the upstream road segment is D=Vu÷(Vu+Vd)×L.

[0074] It should be noted that the principle behind this calibration method is that the speed ratio on the side of the road segment with lower speed, i.e. the more congested road segment, is smaller, so the calculated offset distance is shorter. The calibration position of the congestion boundary is closer to the center point of the road segment with lower speed, i.e. closer to the congested area. On the side of the road segment with higher speed, i.e. the more unobstructed road segment, the speed ratio is larger, so the congestion boundary is farther from the center point of the unobstructed road segment. The calibration position of the congestion boundary in this way conforms to the actual traffic pattern that the edge of the congested area tends to be biased towards the congested side.

[0075] The same calibration process is performed on each pair of adjacent road segments in the road network that have upstream and downstream connections and whose absolute value of speed spatial gradient exceeds the gradient threshold, thereby obtaining the calibrated position of the congestion boundary of each pair of adjacent road segments.

[0076] The changes in the location of the congestion boundary of the same pair of adjacent road segments are tracked within a continuous analysis time window. The boundary displacement is obtained by subtracting the offset distance of the congestion boundary from the center point of the upstream road segment in the previous analysis time window from the offset distance of the congestion boundary from the center point of the upstream road segment in the current analysis time window. When the boundary displacement is negative, it indicates that the congestion boundary is moving upstream, that is, the congestion is spreading to the upstream road segment. When the boundary displacement is positive, it indicates that the congestion boundary is moving downstream, that is, the congestion is dissipating. The absolute value of the boundary displacement is divided by 1000 to convert it to kilometers, then divided by the length of the analysis time window, that is, 300 seconds, to convert it to kilometers per second. Finally, it is multiplied by 3600 to convert it to kilometers per hour to obtain the moving speed of the congestion boundary.

[0077] When the absolute value of the spatial velocity gradient of a pair of adjacent road segments no longer exceeds the gradient threshold in a certain analysis time window, it is determined that the congestion boundary between the pair of adjacent road segments has disappeared, that is, the congestion has completely dissipated or has expanded beyond the range of the pair of road segments, and the tracking of the congestion boundary of the pair of adjacent road segments is stopped.

[0078] This step also feeds back congestion boundary information to the traffic flow status determination in step four. When a road segment is adjacent to the non-congested side of a congestion expansion boundary and the current traffic flow status label for that road segment is transitional, and the congestion boundary is moving towards that road segment, a warning sign indicating impending congestion is added to that road segment. The specific method for determining whether the congestion boundary is moving towards a road segment is as follows: when the road segment is located upstream of the congestion boundary (i.e., the road segment is upstream), if the boundary displacement is negative (i.e., the congestion boundary is moving upstream), then it is determined that the congestion boundary is moving towards that road segment. The congestion boundary is moved in the direction of the segment. When the segment is located downstream of the congestion boundary, if the boundary displacement is positive, the congestion boundary is moving downstream, and it is determined that the congestion boundary is moving towards the segment. If the boundary displacement is away from the segment or is zero, the congestion boundary is determined not to be moving towards the segment and no warning is added. For each pair of adjacent segments with upstream and downstream connections in the road network, the same congestion boundary detection, calibration and tracking are performed in each analysis time window to obtain the calibration position and movement speed of each congestion boundary.

[0079] Based on the pre-configured upstream and downstream connections of each monitored road segment in the road network, the spatial gradient of the fusion speed is calculated for each pair of adjacent road segments. When the absolute value of the gradient exceeds a set threshold, a congestion boundary is determined to exist between the two road segments. Then, the specific location of the congestion boundary between the centers of the two road segments is linearly interpolated using the proportional relationship of the fusion speed of the upstream and downstream road segments. The offset of the boundary position is tracked within a continuous time window to calculate the direction and speed of the boundary's movement, and the congestion boundary information is fed back to the traffic flow status determination in step four. When a road segment is adjacent to the non-congested side of a congestion expansion boundary and the boundary is moving towards that road segment, an impending congestion warning mark is added to that road segment. The beneficial effect of this step is that it achieves precise positioning and dynamic tracking of the congestion boundary in the road network space, enabling traffic management departments not only to know which road segments are congested but also to know in which direction the congestion is spreading and how fast it is spreading. At the same time, through the early warning feedback mechanism, spatial propagation information is integrated into the status determination of each road segment to form a closed loop, so that road segments that are about to be affected by congestion can be identified and warned in advance.

[0080] Step Six: Based on the traffic flow status labels, fusion speed, free-flow speed, and congestion boundary information of each road segment, dynamically classify the congestion severity of each road segment to obtain and output the congestion level label for each road segment. The specific method is as follows:

[0081] First, divide the fusion speed of each monitored road segment in the current analysis time window by the free flow speed of each monitored road segment to obtain the speed ratio of each monitored road segment. Then, subtract the speed ratio of each monitored road segment from 1 to obtain the speed deviation of each monitored road segment in the current analysis time window. The speed deviation is used to quantify the degree to which the current fusion speed of each road segment deviates from the free flow speed. The speed deviation ranges from 0 to 1. 0 means that the current speed is equal to or greater than the free flow speed, i.e., completely unobstructed. 1 means that the current speed is 0, i.e., completely blocked. The larger the speed deviation, the more severe the congestion.

[0082] If the traffic flow status label of a road segment is "free flow", then the congestion level label of that road segment is directly determined to be "smooth", because the vehicle density is low enough in the free flow state and there is no risk of congestion regardless of the speed deviation value.

[0083] If the traffic flow status label of a road segment is "transitional flow" and the speed deviation is less than the first judgment threshold of 0.5, then the congestion level label of the road segment is judged as "basically smooth," meaning that although vehicles have begun to affect each other, the speed reduction is still within an acceptable range. If the traffic flow status label is "transitional flow" and the speed deviation is greater than or equal to the first judgment threshold of 0.5, then the congestion level label of the road segment is judged as "mild congestion," meaning that the vehicle speed is significantly lower than the free flow level.

[0084] If the traffic flow status label of a road segment is synchronous congestion and the speed deviation is less than the first judgment threshold of 0.5, then the congestion level label of the road segment is judged as mild congestion; if the traffic flow status label of a road segment is synchronous congestion and the speed deviation is greater than or equal to the first judgment threshold of 0.5 and less than the second judgment threshold of 0.7, then the congestion level label of the road segment is judged as moderate congestion, that is, although the vehicles are moving slowly, their speed has decreased significantly.

[0085] If the traffic flow status label of a road segment is synchronous congestion and the speed deviation is greater than or equal to the second judgment threshold of 0.7, then the congestion level label of the road segment is judged as severe congestion, that is, the vehicle speed has dropped to an extremely low level and is close to stagnation.

[0086] If the traffic flow status label of a road segment is "severe congestion", then the congestion level label of that road segment will be directly determined as "severe congestion", because in the state of severe congestion, vehicles are basically at a standstill, and regardless of the specific value of the speed deviation, it belongs to the most severe congestion level; the first judgment threshold for speed deviation is set to 0.5 and the second judgment threshold is set to 0.7.

[0087] For road segments where the fusion density marker is missing in step three, since it is impossible to determine the traffic flow status based on the speed-density two-dimensional plane, a simplified congestion assessment is directly performed based on the fusion speed: the fusion speed of the road segment is divided by the free-flow speed to obtain the speed ratio. If the speed ratio is greater than or equal to 0.8, the congestion level label is determined to be smooth; if the speed ratio is greater than or equal to 0.5 and less than 0.8, the congestion level label is determined to be basically smooth; if the speed ratio is greater than or equal to 0.3 and less than 0.5, the congestion level label is determined to be slightly congested; if the speed ratio is greater than or equal to 0.15 and less than 0.3, the congestion level label is determined to be moderately congested; if the speed ratio is less than 0.15, the congestion level label is determined to be severely congested.

[0088] Then, the congestion level labels of each road segment are adjusted for warning. When a road segment is marked with an impending congestion warning label from step five, if the current congestion level label of the road segment is smooth, it is adjusted to basically smooth. If the current congestion level label of the road segment is basically smooth, it is adjusted to light congestion. If the current congestion level label is light congestion or higher, it remains unchanged, so as to reflect the warning effect that the congestion boundary is approaching the road segment.

[0089] Then, video data is used to assist in verifying the congestion level labels of each road segment. Specifically, for road segments with congestion level labels of smooth or basically smooth, if the video queue length in the current window of the road segment exceeds 0.5 times the length of the road segment, the congestion level label of the road segment is upgraded by one level, i.e., smooth is changed to basically smooth, and basically smooth is changed to light congestion, because a longer queue length indicates that there is vehicle backlog at the intersection, which does not match the smooth state. For road segments with congestion level labels of moderate or severe congestion, if the video queue length in the current window of the road segment is less than 20 meters and the average number of vehicles in the video is less than 5, the congestion level label of the road segment is downgraded by one level, i.e., severe congestion is changed to moderate congestion, and moderate congestion is changed to light congestion, because a shorter queue length and fewer vehicles indicate that the actual congestion level at the intersection may be overestimated.

[0090] Finally, the real-time monitoring results of each monitored road segment are integrated and output. The output includes: the road segment number, the timestamp of the current analysis time window, the fusion speed, the fusion density, the traffic flow status label, the congestion level label, and the speed deviation value. For adjacent road segments with congestion boundaries, the coordinates of the congestion boundary's location, the congestion boundary type, and the congestion boundary's moving speed are also output. For road segments with an upcoming congestion warning mark, the warning mark and the estimated congestion arrival time are also output. The estimated congestion arrival time is equal to the distance from the center point of the road segment to the congestion boundary's location divided by the congestion boundary's moving speed. The above output results are continuously updated in cycles based on the analysis time window. After each analysis time window ends, the next round of data collection and processing is automatically triggered, realizing real-time and continuous monitoring of the traffic congestion status of the urban road network.

[0091] The ratio of fused velocity to free-flow velocity is converted into a speed deviation. Then, traffic flow status labels are cross-combined with the speed deviation to determine congestion levels, categorized into five levels: smooth, mostly smooth, lightly congested, moderately congested, and severely congested. For road segments lacking fused density, a simplified assessment is performed directly based on the speed ratio. Following this, the congestion level of road segments about to be affected by congestion is adjusted upwards based on the warning markers from step five. Finally, video queue length and vehicle count are used to assist in verifying and correcting the determined congestion levels. All monitoring results are integrated and output, continuously updated within an analysis time window. The beneficial effects of this step are: establishing a multi-layered congestion classification mechanism based on traffic flow status label acceleration deviation, warning markers, and video verification; this is more refined and reliable than simply classifying levels based on speed thresholds; independent verification of video data can identify and correct overestimation or underestimation of levels caused by systematic deviations in fused velocity; and the adjustment of warning marker levels reflects the potential impact of congestion spatial propagation on surrounding road segments, making the final output congestion level more closely reflect actual road conditions.

[0092] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0093] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for real-time monitoring of traffic congestion based on multi-source data fusion from the Internet of Things, characterized in that, include: Step 1: Set up geomagnetic coil detectors, roadside microwave radars, floating car GPS data acquisition terminals, and intersection video surveillance cameras on each monitoring section of the urban road network. Collect traffic status data for each monitoring section according to their respective preset collection cycles. Unify all types of data into the same analysis time window for analysis to obtain standard data records for each monitoring section. Each standard data record includes coil speed, coil flow, coil occupancy rate, equivalent density, radar speed, radar flow, GPS speed, GPS sample size, video queue length, and number of vehicles in the video. Step 2: Pair the coil speed, radar speed and GPS speed in the standard data records of each monitoring section to construct a cross-verification matrix. Based on the cross-verification matrix, calculate the credibility score of each data source for coil speed, radar speed and GPS speed in different analysis time windows for each monitoring section. Step 3: Based on the credibility scores of each data source within different analysis time windows for each monitored road segment, the speed values ​​of each data source are weighted and fused to obtain the fusion speed and fusion density of each monitored road segment within each analysis time window; Step 4: Map the fusion speed and fusion density onto a two-dimensional plane with density as the horizontal axis and speed as the vertical axis to obtain the trajectory. Determine the trajectory movement direction based on the fusion speed and fusion density of the current analysis time window and the previous analysis time window. Distinguish the current traffic flow status of each monitored road segment and obtain the traffic flow status label of each monitored road segment. Step 5: Obtain the upstream and downstream connection relationships of each monitored road segment in the road network, calculate the speed spatial gradient of each pair of adjacent road segments with upstream and downstream connection relationships, detect and track the congestion boundary between adjacent road segments, and obtain the calibrated position and moving speed of the congestion boundary. Step Six: Determine the congestion level label based on the traffic flow status label and speed deviation of each road segment. At the same time, use the video queue length and the number of vehicles in the video to assist in the verification of the congestion level label, and integrate and output the real-time monitoring results of each monitored road segment.

2. The method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, as described in claim 1, is characterized in that... A uniform analysis time window length of 5 minutes is set, and data with a collection timestamp greater than or equal to the start time of the window and less than the end time of the window are included in the corresponding analysis time window; for data from geomagnetic coil detectors, the cross-sectional traffic flow, cross-sectional average speed, and cross-sectional time occupancy rate are directly used as coil flow, coil speed, and coil occupancy rate, respectively. For roadside microwave radar data, the average radar detection speed of each data point within the window is used as the radar speed, and the sum of radar detection flow rate is used as the radar flow rate. For floating car GPS data, the average speed of each operating vehicle within the window is calculated first, and then the average speed of all operating vehicles is averaged to obtain the GPS speed. The percentage value of coil occupancy is divided by 100, multiplied by 1000, and then divided by the preset average effective vehicle length to obtain the equivalent density.

3. The method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, as described in claim 1, is characterized in that... The specific method for obtaining the reliability scores of each data source—coil velocity, radar velocity, and GPS velocity—within different analysis time windows for each monitored road segment is as follows: For each monitored road segment within each analysis time window, a 3x3 cross-validation matrix is ​​constructed using three data sources: coil speed, radar speed, and GPS speed from the standard data records. The first data source is coil speed, the second is radar speed, and the third is GPS speed. The element in the i-th row and j-th column of the matrix represents the cross-validation consistency score between the i-th and j-th data sources. After the cross-validation matrix is ​​constructed, the sum of all off-diagonal elements in each data category's row is divided by the number of off-diagonal elements to obtain the consistency mean for the corresponding data. This yields the consistency mean Ai for coil speed, radar speed, and GPS speed. For coil speed and radar speed, a basic integrity score of 1 is given if the data is normal, and a score of 0 is given if the data is missing or abnormal. For GPS speed... The basic integrity score is equal to the minimum of the quotient obtained by dividing the GPS sample size by the sample size threshold and 1. The sample size threshold is equal to the number of lanes in the corresponding road segment multiplied by 3. Thus, the basic integrity scores Bi for coil speed, radar speed and GPS speed are obtained. The row consistency mean Ai of each data source is multiplied by 0.7 to obtain the first product. Then, the basic integrity scores Bi of each data source are multiplied by 0.3 to obtain the second product. Finally, the first product and the second product are added together to obtain the credibility score Qi of each data source. For each monitored road segment, the same analysis is performed on the three data sources of coil speed, radar speed and GPS speed in the standard data records in each analysis time window. Thus, the credibility scores of each data source of coil speed, radar speed and GPS speed in different analysis time windows of each monitored road segment are obtained.

4. The method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, as described in claim 3, is characterized in that... The specific method for obtaining the cross-validation consistency score between the i-th type of data source and the j-th type of data source in the matrix is ​​as follows: For the velocity value Vi of the i-th data source and the velocity value Vj of the j-th data source in the matrix, first subtract Vj from Vi and take the absolute value to obtain the deviation value Dij. Then divide Dij by the maximum value between Vi and Vj to obtain the relative deviation Rij. Finally, subtract Rij from 1 to obtain the mutual verification consistency score Mij. If the result of the Mij calculation is less than 0, it is taken as 0. The diagonal elements of the matrix are always taken as 1.

5. The method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, as described in claim 3, is characterized in that... The specific method for obtaining the fusion speed and fusion density of each monitored road segment within each analysis time window is as follows: For each monitoring segment within a single analysis time window, the quotient obtained by dividing the reliability score Qi of that data source by the sum of the reliability scores of all valid data sources participating in the fusion within the current window is used as the fusion weight Wi of that data source within that analysis time window for that monitoring segment. The product of the coil speed multiplied by the coil fusion weight, the product of the radar speed multiplied by the radar fusion weight, and the product of the GPS speed multiplied by the GPS fusion weight within that window is used as the fusion speed Vf within that analysis time window. When the coil data is valid within that window, the equivalent density is directly used as the fusion density Kf within that window. When the coil data is missing, the quotient obtained by multiplying the radar flow rate within that window by 12 and then dividing it by the fusion speed is used as the fusion density Kf within that window. When both coil and radar data are missing, the fusion density within that window is marked as missing. The reliability scores of each data source for coil speed, radar speed, and GPS speed within different analysis time windows for each monitoring segment are analyzed in the same way to obtain the fusion speed and fusion density of each monitoring segment within each analysis time window.

6. The method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, as described in claim 5, is characterized in that... The specific method for obtaining traffic flow status labels for each monitored road segment is as follows: When the fusion density is less than 0.5 times the critical density, it is determined to be in a free-flow state; when the fusion density is greater than or equal to 0.5 times the critical density but less than the critical density, if the trajectory movement direction is in the direction of deterioration, stability, abnormal decay, or free growth, it is determined to be in a transitional flow state; if the trajectory movement direction is in the direction of improvement, it is determined to be in a free-flow state; when the fusion density is greater than or equal to the critical density and the fusion speed is greater than 0.2 times the free flow speed, it is determined to be in a synchronous congestion state. A state of severe congestion is defined as when the fusion density is greater than or equal to the critical density and the fusion rate is less than or equal to 0.2 times the free flow rate.

7. The method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, as described in claim 6, is characterized in that... The specific method for determining the direction of trajectory movement is as follows: When the absolute value of the density change is less than 2 vehicles per kilometer and the absolute value of the speed change is less than 1 kilometer per hour, it is considered a stable state; when the density change is greater than 0 and the speed change is less than 0, it is considered a deteriorating trend; when the density change is less than 0 and the speed change is greater than 0, it is considered an improving trend; when the density change is greater than 0 and the speed change is greater than 0, it is considered a free growth trend; when the density change is less than 0 and the speed change is less than 0, it is considered an abnormal decay trend.

8. The method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, as described in claim 6, is characterized in that... The specific method for obtaining the location and speed of movement of the congestion boundary is as follows: For adjacent road segments with upstream and downstream connections, the speed difference is obtained by subtracting the fusion speed of the downstream road segment from the fusion speed of the upstream road segment. This speed difference is then divided by the distance between the center points of the two road segments, and the quotient is used as the speed spatial gradient of the pair of adjacent road segments. When the absolute value of the speed spatial gradient exceeds a gradient threshold, a congestion boundary is determined to exist between the pair of adjacent road segments. If the traffic flow status label of one side of the pair of adjacent road segments is synchronous congestion or severe congestion, while the traffic flow status label of the other side is free flow or transitional flow, then the congestion boundary is determined to be a congestion expansion type boundary. The fusion speed of the upstream road segment is labeled Vu, the fusion speed of the downstream road segment is labeled Vd, and the distance between the center points of the two road segments is labeled L. The distance is calculated using D = Vu ÷ (Vu + Vd). The offset distance D from the center point of the upstream road segment to the congestion boundary is calculated using ×L. The location reached by measuring this offset distance from the geographical coordinates of the upstream road segment's center point downstream is the designated position of the congestion boundary. The offset distance from the congestion boundary to the center point of the upstream road segment in the current analysis time window is subtracted from the offset distance in the previous analysis time window to obtain the boundary displacement. The absolute value of the boundary displacement is divided by 1000 to convert it to kilometers, then divided by the analysis time window length, and multiplied by 3600 to convert it to kilometers per hour to obtain the moving speed of the congestion boundary. When a road segment is adjacent to the non-congested side of a congestion expansion boundary, and the current traffic flow status label for that road segment is transitional, and the congestion boundary is moving towards that road segment, an impending congestion warning is added to that road segment.

9. A method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, as described in claim 8, is characterized in that... The specific method for determining congestion level labels and performing auxiliary verification is as follows: The speed ratio is obtained by dividing the fused speed of each monitored road segment in the current analysis time window by the free-flow speed. The speed deviation is then calculated by subtracting the speed ratio from 1. When the speed ratio is greater than 1, the speed deviation is set to 0. If the traffic flow status label is "free-flowing," the congestion level label is determined to be "smooth." If the traffic flow status label is "transitional flowing" and the speed deviation is less than 0.5, it is determined to be "basically smooth." If the speed deviation is greater than or equal to 0.5, it is determined to be "mildly congested." If the traffic flow status label is "synchronous congestion" and the speed deviation is less than 0.5, it is determined to be "mildly congested." If the speed deviation is greater than or equal to 0.5 and less than 0.7, it is determined to be "moderately congested." If the speed deviation is greater than or equal to 0.7, it is determined to be "severely congested." If the traffic flow status label is "severely congested," it is directly determined to be "severely congested." Regarding the fusion density... For road segments marked as missing, if the speed ratio is greater than or equal to 0.8, it is considered smooth; if it is greater than or equal to 0.5 and less than 0.8, it is considered basically smooth; if it is greater than or equal to 0.3 and less than 0.5, it is considered lightly congested; if it is greater than or equal to 0.15 and less than 0.3, it is considered moderately congested; and if it is less than 0.15, it is considered severely congested. When a road segment is marked with an impending congestion warning, if the congestion level label is smooth, it is adjusted to basically smooth; if it is basically smooth, it is adjusted to lightly congested. For road segments with a congestion level label of smooth or basically smooth, if the video queue length exceeds 0.5 times the length of the road segment, the congestion level label is raised by one level. For road segments with a congestion level label of moderately or severely congested, if the video queue length is less than 20 meters and the average number of vehicles in the video is less than 5, the congestion level label is lowered by one level.

10. A method for real-time monitoring of traffic congestion based on multi-source data fusion of the Internet of Things, as described in claim 9, is characterized in that... The specific method for integrating and outputting the real-time monitoring results of each monitored road segment is as follows: The output includes the segment number of each monitored road segment, the timestamp of the current analysis time window, the fusion speed, the fusion density, the traffic flow status label, the congestion level label, and the speed deviation value. For adjacent road segments with congestion boundaries, the output also includes the coordinates of the congestion boundary's location, the congestion boundary type, and the congestion boundary's moving speed. For road segments with an upcoming congestion warning sign, the output also includes the warning sign and the estimated congestion arrival time. The estimated congestion arrival time is equal to the distance from the center point of the road segment to the congestion boundary's location divided by the congestion boundary's moving speed.

Images