A Traffic Anomaly Section Identification System and Method Based on Traffic Flow

Abstract

This invention discloses a traffic anomaly segment identification system and method based on traffic flow, relating to the field of intelligent traffic management and control technology. The invention includes a modeling module for collecting basic data for each segment and establishing a basic model for the corresponding segment based on the basic data; a calibration module for determining the normal traffic flow for the corresponding segment based on historical traffic flow and the basic model; and an update module for setting update conditions and updating the basic model and normal traffic flow for the segment when the update conditions are met. This invention utilizes a small amount of dynamically transferable point cloud data for each lane and combines this with point cloud change information to generate single-point, multi-point, or all fault labels, achieving accurate differentiation of continuous traffic flow sampling and anomaly types with low computational cost. This avoids misjudgment of segments caused by individual vehicles being stationary and provides an operable fault classification basis for traffic control.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic management and control technology, specifically to a traffic anomaly section identification system and method based on traffic flow. Background Technology

[0002] In the field of intelligent traffic management and control, accurate and real-time identification of abnormal road traffic sections (such as congestion, accidents, and temporary traffic control) is a crucial prerequisite for implementing dynamic guidance, signal control, and emergency dispatch. Traditional methods for identifying abnormal sections mainly rely on cross-sectional loop detectors, geomagnetic detectors, or fixed video surveillance, judging road operating status by statistically analyzing traffic flow, average speed, or time occupancy per unit time. However, existing technologies still have the following shortcomings in practical applications: First, the baseline traffic flow is difficult to dynamically adapt to changes in road conditions. Most systems pre-set fixed traffic flow thresholds or build static normal traffic flow models based on historical data. However, when short-term changes occur, such as road construction, temporary traffic control, lane closures, or signal timing adjustments, the baseline model cannot be updated in time. This leads to normal traffic flow being frequently misjudged as abnormal sections, generating a large number of false alarms and reducing the reliability and efficiency of the system. Second, traditional detection methods struggle to distinguish between "individual vehicle malfunctions" and "systemic road congestion." For example, when a vehicle breaks down in a lane but subsequent vehicles can change lanes to bypass it, the cross-section detector may still detect that the cross-section has been empty for an extended period (low traffic flow) and has a zero speed, thus misjudging it as congestion across the entire lane. Conversely, when genuine systemic congestion occurs, the abnormal signals output by the system are indistinguishable from those of a single point of failure. This ambiguous judgment fails to provide traffic management personnel with a clear basis for action—whether to dispatch a tow truck or adjust the traffic light scheme—current technology lacks the ability to provide targeted, tiered output capabilities.

[0003] In view of this, the present invention proposes a traffic anomaly section identification system and method based on traffic flow to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a traffic anomaly section identification system and method based on traffic flow to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a traffic anomaly section identification system based on traffic flow, comprising: A module is established to collect basic data for each segment and build a basic model for the corresponding segment based on the basic data. The calibration module is used to determine the normal traffic flow of the corresponding section based on historical traffic flow and the basic model; The update module is used to set update conditions, and updates the basic model and normal traffic flow of the section when the update conditions are met. The judgment module is used to preset the collection period, collect the real-time traffic flow of each section based on the collection period, compare the real-time traffic flow with the normal traffic flow of the corresponding section, and identify the sections that are not in the normal traffic flow as abnormal sections.

[0006] In a preferred embodiment, the establishment module includes: The segmentation unit is used to divide traffic roads into multiple segments and assign a unique code to each segment. The first collection unit is used to collect road specification parameters and road facilities of each section as basic parameters, and to build a three-dimensional contour model based on the basic parameters, thereby obtaining the basic model of the corresponding section.

[0007] In a preferred embodiment, the calibration module includes: The second collection unit is used to collect the historical traffic flow corresponding to each section at each time period, each weather parameter, and each basic parameter. The historical traffic flow includes the traffic speed and the number of vehicles passing through. The calibration unit is used to determine the normal traffic flow of each section based on each time period, each weather parameter, each basic parameter and the corresponding historical traffic flow. The normal traffic flow includes the predicted speed range and the predicted traffic volume range.

[0008] In a preferred embodiment, the update module includes: The setting unit is used to set multiple update points for the basic model of each section. The update point whose displacement is greater than a preset threshold is used as the update condition for the corresponding basic model. The number of update points corresponding to each update condition is different, and the preset threshold for the displacement corresponding to each update point is different. The update unit is used to update the basic model and normal traffic flow of the corresponding section when the update point meets the update conditions.

[0009] In a preferred embodiment, the determination module includes: The data collection unit is used to preset the data collection period and collect real-time traffic flow data for each section based on the data collection period. The comparison unit is used to compare the real-time traffic flow of each section with the normal traffic flow to obtain the comparison results; The judgment unit is used to classify segments that are not in normal traffic flow as abnormal segments and segments that are in normal traffic flow as normal segments. The upload unit is used to upload abnormal sections and codes, as well as real-time traffic flow, to the traffic management center.

[0010] In a preferred embodiment, the acquisition unit includes: The marking unit is used to determine the start and end points of the segment. At the start point of the segment, point cloud marking is performed on n vehicles in multiple lanes. Each point cloud is bound to the outline and color of the marked vehicle. Multiple point clouds in the same lane are associated to obtain multiple association chains. The migration unit is used to set migration conditions, migrate the point cloud based on the migration conditions, and bind the point cloud with the outline and color of the newly marked vehicle after migration. It generates road fault labels based on the change information of the point cloud in multiple association chains. The migration condition is: the vehicle displacement corresponding to the current point cloud is less than a preset displacement threshold. The calculation unit is used to determine the moving speed, moving distance and moving time of multiple point clouds within the segment, and to obtain real-time traffic flow based on the moving speed, moving distance and moving time of multiple point clouds within the segment.

[0011] In a preferred embodiment, the step of generating road fault labels based on the change information of point clouds in multiple association chains includes: Add change information to the migrating point cloud, including lane change information and movement information; Multiple road fault labels are set, and the road fault labels corresponding to the changed information are added to the corresponding abnormal sections. The road fault labels include single-point faults, multi-point faults, and all faults.

[0012] This invention also provides a method for identifying traffic anomaly sections based on traffic flow, comprising the following steps: Step 1: Collect basic data for each segment and build a basic model for the corresponding segment based on the basic data; Step 2: Determine the normal traffic flow for the corresponding section based on historical traffic flow and the basic model; Step 3: Set update conditions, and update the basic model and normal traffic flow of the section when the update conditions are met; Step 4: Collect real-time traffic flow data for each section, compare the real-time traffic flow with the normal traffic flow for the corresponding section, and identify sections that are not in normal traffic flow as abnormal sections.

[0013] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention can update the basic model in a timely manner by changing the update points, and update the normal traffic flow of the corresponding section by updating the basic model; in this way, it can reduce the situation where the corresponding section is judged as an abnormal section due to road construction or temporary control, and reduce the occurrence of false alarms.

[0014] 2. This invention utilizes a small amount of dynamically transferable point cloud data in each lane and combines point cloud change information to generate single-point, multi-point, or all fault labels. This enables accurate differentiation of continuous traffic flow sampling and anomaly types with low computing power cost, avoids misjudgment of sections caused by individual vehicles stopping, and provides an operable fault classification basis for traffic management. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a system block diagram of the present invention.

[0017] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0019] Example 1, please refer to Figure 1 As shown in this embodiment, a traffic anomaly section identification system based on traffic flow includes: A module is established to collect basic data for each segment and build a basic model for the corresponding segment based on the basic data. The calibration module is used to determine the normal traffic flow of the corresponding section based on historical traffic flow and the basic model; The update module is used to set update conditions, and updates the basic model and normal traffic flow of the section when the update conditions are met. The judgment module is used to preset the collection period, collect the real-time traffic flow of each section based on the collection period, compare the real-time traffic flow with the normal traffic flow of the corresponding section, and identify the sections that are not in the normal traffic flow as abnormal sections.

[0020] In one embodiment, the establishment module includes: The segmentation unit is used to divide traffic roads into multiple segments and assign a unique code to each segment. The first collection unit is used to collect road specification parameters and road facilities of each section as basic parameters, and to build a three-dimensional contour model based on the basic parameters, thereby obtaining the basic model of the corresponding section.

[0021] It should be noted that in the division of units, by combining key boundary conditions such as road network planning, intersection distribution, traffic light nodes, traffic flow divergence and merging points, administrative road boundaries, and various functional roads (arterial roads, secondary arterial roads, branch roads, and expressways), the entire area's connecting roads are fragmented and uniformly divided into grids, forming several independent, continuous, non-overlapping, and complete standardized road segments. Simultaneously, each segment is assigned a unique identification code using digital coding, character coding, or a combination of coding methods. This coding information is linked to basic information such as the segment's geographical location, road grade, and affiliated region. It can provide identification for subsequent data collection, traffic flow calibration, and anomaly location. In the first collection unit, it collects basic parameters of each uniquely coded section. Road specification parameters include total road length, road width, number of one-way / two-way lanes, pavement material, slope, curvature, speed limit, and pedestrian crossing distribution. Road facility parameters include the quantity, location, and operational status of ancillary facilities such as traffic lights, surveillance cameras, speed measuring devices, guardrails, bus stops, ramp entrances and exits, parking lot entrances and exits, and emergency lanes. The above-mentioned basic parameters are integrated and entered. Relying on 3D spatial modeling technology, GIS geographic information system, and spatial contour simulation algorithm, the road topography, pavement structure, facility layout, and spatial boundaries are restored to construct a high-precision 3D contour model of the section. Using the 3D contour model as the core carrier, implicit parameters such as road carrying capacity, traffic restrictions, and spatial constraints are superimposed to finally generate a dedicated basic model adapted to the actual traffic conditions of each section, ensuring that the basic model of each section is highly consistent with the actual road scene of the section.

[0022] In one embodiment, the calibration module includes: The second collection unit is used to collect the historical traffic flow corresponding to each section at each time period, each weather parameter, and each basic parameter. The historical traffic flow includes the traffic speed and the number of vehicles passing through. The calibration unit is used to determine the normal traffic flow of each section based on each time period, each weather parameter, each basic parameter and the corresponding historical traffic flow. The normal traffic flow includes the predicted speed range and the predicted traffic volume range.

[0023] It should be noted that in the second collection unit, traffic flow, speed, timestamps, and corresponding weather records for each section for more than three consecutive months are collected and stored in the historical database. In the calibration unit, a multi-input multi-output nonlinear mapping model is constructed. This nonlinear mapping model is used to determine the normal traffic flow of each section under each time period, weather parameter, and basic parameter. The specific construction and training process of the nonlinear mapping model is as follows: First, the model input feature space is designed. The input feature vector includes time period features (hour, minute, day of the week, whether it is a holiday, peak / off-peak / valley labels), weather parameters (weather type, precipitation, visibility, temperature, road surface slip coefficient), and basic parameters extracted from the three-dimensional contour model (number of lanes, road width, intersection density, traffic light cycle and green light ratio, speed limit, presence of ramps or bus stops, etc.). The model output target includes two key indicators: traffic speed (the spatial average speed of all vehicles in the section) and traffic volume (the number of vehicles passing through a specified section per unit time). During the collection of training data for the model, traffic flow, speed, timestamps, and corresponding weather records for each segment for more than three consecutive months are extracted from the historical database. Extreme abnormal data caused by external interference such as accidents, construction, and temporary traffic control are removed. Each sample is divided into training set, validation set, and test set according to time sequence. The preferred model structure is Gradient Boosting Decision Tree (GBDT) or its variants (such as XGBoost, LightGBM), but Random Forest or Multilayer Perceptron (MLP) can also be used. SHAP values ​​can be used for feature contribution analysis. The training process includes Z-score standardization for continuous features and one-hot encoding for discrete features. The model employs code or label encoding, hyperparameter tuning (e.g., using grid search or Bayesian optimization combined with 5-fold cross-validation), selection of a loss function (combining the mean absolute error for speed prediction and the root mean square error for traffic flow prediction to form a loss function), and training termination conditions (i.e., stopping early when the validation set loss no longer decreases after several consecutive rounds). After the model training is completed, for any given input feature vector, the model outputs the predicted traffic speed range and the predicted traffic volume range. Thus, through the construction and training of the above nonlinear mapping model, the predicted traffic speed range and the predicted traffic volume range can be obtained from the input feature vector (features of each time period, each weather parameter, and each basic parameter).

[0024] In one embodiment, the update module includes: The setting unit is used to set multiple update points for the basic model of each section. The update point whose displacement is greater than a preset threshold is used as the update condition for the corresponding basic model. The number of update points corresponding to each update condition is different, and the preset threshold for the displacement corresponding to each update point is different. The update unit is used to update the basic model and normal traffic flow of the corresponding section when the update point meets the update conditions.

[0025] It should be noted that update points refer to feature parameters used to characterize the key performance state of the basic model, and are divided into two categories: one is geometric structure update points, which come from deformable features in the 3D contour model, such as road surface horizontal displacement monitoring points and lane edge line offsets; the other is traffic parameter update points, which come from key parameters of the calibration model, such as free flow velocity, congestion density, critical density, saturation flow rate, and signal timing period. Displacement refers to the change in parameter values. Update conditions are set in a hierarchical manner, with three preset update conditions of different severity levels, corresponding to different numbers of update points and displacement thresholds. Level 1 is a fine-tuning condition, triggered when the displacement of any update point exceeds the preset small threshold for that point, performing online incremental updates only on the local parameters associated with that update point; Level 2 is a moderate update condition, triggered when any three or more update points belonging to the same... The update module is triggered when each update point in a functional domain exceeds its respective medium threshold. It retrains the local sub-model of that functional domain using valid data from the last seven days and recalculates the normal traffic flow under relevant conditions. Level three is a full reconstruction condition, triggered when more than 50% of the total update points exceed their respective large thresholds. It then recollects historical data from the last thirty days, retrains the complete calibration model from scratch, and regenerates the normal traffic flow baseline table for all conditions. Displacement is calculated as follows: geometric update points use the Euclidean distance between the current measurement location and the recorded location in the basic model; traffic parameter update points use the relative displacement percentage. A background scheduled task (e.g., once per hour) compares the current value of all update points with the historical baseline value. Once an update condition is met, the update module immediately updates the basic model and the normal traffic flow. This allows for timely updates to the basic model based on changes in update points, and updates to the normal traffic flow of the corresponding sections based on the updates to the basic model. This reduces the likelihood of sections being flagged as abnormal due to road construction or temporary traffic control, thus reducing false alarms.

[0026] In one embodiment, the determination module includes: The data collection unit is used to preset the data collection period and collect real-time traffic flow data for each section based on the data collection period. The comparison unit is used to compare the real-time traffic flow of each section with the normal traffic flow to obtain the comparison results; The judgment unit is used to classify segments that are not in normal traffic flow as abnormal segments and segments that are in normal traffic flow as normal segments. The upload unit is used to upload abnormal sections and codes, as well as real-time traffic flow, to the traffic management center.

[0027] It should be noted that each segment is bound to its normal traffic flow under the current time period, weather parameters, and basic parameters. By comparing the real-time traffic flow of each segment with the normal traffic flow of the corresponding segment, if the real-time traffic flow is not within the range corresponding to the predicted travel time and predicted travel speed of the normal traffic flow, the current segment is considered an abnormal segment. Conversely, real-time traffic flow that is not within the range corresponding to the predicted travel time and predicted travel speed is considered a normal segment. The uploading unit then uploads the abnormal segments, their corresponding codes, and real-time traffic flow to the traffic management center, which then sends the abnormal segment codes to the relevant traffic management personnel so that they can promptly handle the abnormal segments.

[0028] In one embodiment, the acquisition unit includes: The marking unit is used to determine the start and end points of the segment. At the start point of the segment, point cloud marking is performed on n vehicles in multiple lanes. Each point cloud is bound to the outline and color of the marked vehicle. Multiple point clouds in the same lane are associated to obtain multiple association chains. The migration unit is used to set migration conditions, migrate the point cloud based on the migration conditions, and bind the point cloud with the outline and color of the newly marked vehicle after migration. It generates road fault labels based on the change information of the point cloud in multiple association chains. The migration condition is: the vehicle displacement corresponding to the current point cloud is less than a preset displacement threshold. The calculation unit is used to determine the moving speed, moving distance and moving time of multiple point clouds within the segment, and to obtain real-time traffic flow based on the moving speed, moving distance and moving time of multiple point clouds within the segment.

[0029] In one embodiment, the step of generating road fault labels based on the change information of point clouds in multiple association chains includes: Add change information to the migrating point cloud, including lane change information and movement information; Multiple road fault labels are set, and the road fault labels corresponding to the changed information are added to the corresponding abnormal sections. The road fault labels include single-point faults, multi-point faults, and all faults.

[0030] It should be noted that the point cloud is not permanently generated for each vehicle, but rather serves as a moving probe to mark different vehicles in the traffic flow based on migration conditions. In the marking unit, a starting point cross-section (such as the exit lane of an upstream intersection) and an ending point cross-section (such as the entrance lane of a downstream intersection) are pre-defined in the road section. High-resolution LiDAR and cameras are deployed at the starting point to cover all lanes. When a vehicle passes the starting point, the system generates a unique point cloud set for the vehicle, which includes a three-dimensional contour (used to distinguish vehicle types) and surface color and texture features. The data structure corresponding to the point cloud records the point cloud ID, generation timestamp, initial lane number, and vehicle contour features, and binds the vehicle's contour and color to the point cloud. Each lane maintains only a preset number of n point clouds at the same time, where n ranges from 2 to 5 (preferably 3). That is, a new point cloud is not generated for every vehicle that enters, but only the initial marking is completed for the first n vehicles to enter. After marking is completed, the system links all the point clouds currently held in the same lane into a doubly linked list according to the actual front-to-back order of the vehicles, forming an association chain. Each point cloud points to the preceding and following vehicles through pointers. At the same time, each point cloud is associated with its current lane number. The migration unit periodically detects the real-time displacement of vehicles bound to each point cloud (e.g., once every 3 seconds). If the displacement of the vehicles associated with the point cloud is less than a preset displacement threshold (e.g., 0.5 meters) within a preset time window (e.g., 1 second), the migration condition is considered met. At this time, the system searches for the migration target in the association chain of the same lane, preferably marking the vehicle following the current point cloud (if the following vehicle has already been marked, it continues to move backward until a vehicle not marked by the point cloud is marked). If the following vehicle is also stationary, it continues to search backward, skipping a maximum of a preset number of vehicles (the preset number of vehicles is generally set to 5). After the target is found in the point cloud, the system unbinds the data of the original point cloud (including point cloud ID, movement history, and cumulative mileage) from the old vehicle and rebinds it to the target vehicle; the association chain is updated accordingly. In addition, when the system passes through the point cloud... When a vehicle changes lanes laterally, its point cloud is inserted into the association chain of the lane that has changed lanes, and a lane change marker (including lane change time, original lane, and target lane) is added to the point cloud change information. Based on the change information of point clouds in multiple association chains, the migration unit continuously generates road fault labels: if only a single point cloud is stagnant for a long time and other point clouds in the same or adjacent lanes can migrate normally, it is determined to be a single-point fault; if multiple discontinuous point clouds are stagnant at the same time but there are still point clouds migrating normally, it is determined to be a multi-point fault; if all or more than 80% of the point clouds in a section are stagnant at the same time or their speed drops suddenly and the duration exceeds the threshold (e.g., 30 seconds), it is determined to be a full fault (corresponding to systemic road congestion). Finally, the upload unit outputs these fault labels, along with the abnormal section code, the involved point cloud ID, and the road fault label to the upload unit. In the computing unit, based on the continuously migrating point cloud, the motion parameters of each point cloud are calculated. The moving speed can be obtained directly from the instantaneous speed of the vehicle bound to the point cloud (e.g., from V2X or radar speed measurement), or calculated based on the ratio of the longitudinal displacement difference between two consecutive position updates to the time interval. The moving distance accumulates along the lane centerline from the moment the point cloud enters the starting point until it moves out of the end point section. The moving time records the difference between the timestamp of the point cloud entering the section and the timestamp of its last appearance in the section, reflecting the travel time of the virtual probe represented by the point cloud traversing the section. For real-time traffic flow calculation, the system sets a sliding time window (e.g., 1 minute) and counts the number of point cloud moving out events that pass through the end point section of the section within the window. This number is proportional to the number of real vehicles. The proportionality coefficient is obtained through offline calibration, and then the absolute value is calculated. Traffic flow rate; or directly using the number of points removed from the point cloud per unit time as a relative traffic flow index, which can be compared with the historical normal traffic flow range to determine anomalies; when the average speed is low but the traffic flow is also low, the system combines the fault labels given by the migration unit to distinguish between low traffic flow caused by single-point faults and low traffic flow caused by system congestion; the calculation unit outputs the real-time speed, mileage, travel time, section traffic flow, and congestion level of each point cloud to the judgment module at a fixed collection period (e.g., 30 seconds) for the final abnormal section identification and alarm; furthermore, by using a small number of dynamically migrated point clouds in each lane and combining the point cloud change information to generate single-point, multi-point, or all fault labels, it achieves continuous sampling of traffic flow and accurate differentiation of abnormal types with low computing power cost, avoids misjudgment of sections caused by individual vehicles stopping, and provides an operable fault classification basis for traffic management.

[0031] Example 2, please refer to Figure 2 As shown in this embodiment, a method for identifying abnormal traffic sections based on traffic flow includes the following steps: Step 1: Collect basic data for each segment and build a basic model for the corresponding segment based on the basic data; Step 2: Determine the normal traffic flow for the corresponding section based on historical traffic flow and the basic model; Step 3: Set update conditions, and update the basic model and normal traffic flow of the section when the update conditions are met; Step 4: Collect real-time traffic flow data for each section, compare the real-time traffic flow with the normal traffic flow for the corresponding section, and identify sections that are not in normal traffic flow as abnormal sections.

[0032] 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 traffic anomaly section identification system based on traffic flow, characterized in that, include: A module is established to collect basic data for each segment and build a basic model for the corresponding segment based on the basic data. The calibration module is used to determine the normal traffic flow of the corresponding section based on historical traffic flow and the basic model; The update module is used to set update conditions, and when the update conditions are met, the basic model and normal traffic flow of the section are updated. The judgment module is used to preset the collection period, collect the real-time traffic flow of each section based on the collection period, compare the real-time traffic flow with the normal traffic flow of the corresponding section, and identify the sections that are not in the normal traffic flow as abnormal sections.

2. The traffic anomaly section identification system based on traffic flow as described in claim 1, characterized in that, The establishment module includes: The segmentation unit is used to divide traffic roads into multiple segments and assign a unique code to each segment. The first collection unit is used to collect road specification parameters and road facilities of each section as basic parameters, and to build a three-dimensional contour model based on the basic parameters, thereby obtaining the basic model of the corresponding section.

3. The traffic anomaly section identification system based on traffic flow according to claim 1, characterized in that, The calibration module includes: The second collection unit is used to collect the historical traffic flow corresponding to each section at each time period, each weather parameter, and each basic parameter. The historical traffic flow includes the traffic speed and the number of vehicles passing through. The calibration unit is used to determine the normal traffic flow of each section based on each time period, each weather parameter, each basic parameter and the corresponding historical traffic flow. The normal traffic flow includes the predicted speed range and the predicted traffic volume range.

4. The traffic anomaly section identification system based on traffic flow as described in claim 1, characterized in that, The update module includes: The setting unit is used to set multiple update points for the basic model of each section. The update point whose displacement is greater than a preset threshold is used as the update condition for the corresponding basic model. The number of update points corresponding to each update condition is different, and the preset threshold for the displacement corresponding to each update point is different. The update unit is used to update the basic model and normal traffic flow of the corresponding section when the update point meets the update conditions.

5. A traffic anomaly section identification system based on traffic flow as described in claim 1, characterized in that, The determination module includes: The data collection unit is used to preset the data collection period and collect real-time traffic flow data for each section based on the data collection period. The comparison unit is used to compare the real-time traffic flow of each section with the normal traffic flow to obtain the comparison results; The judgment unit is used to classify segments that are not in normal traffic flow as abnormal segments and segments that are in normal traffic flow as normal segments. The upload unit is used to upload abnormal sections and codes, as well as real-time traffic flow, to the traffic management center.

6. A traffic anomaly section identification system based on traffic flow according to claim 5, characterized in that, The acquisition unit includes: The marking unit is used to determine the start and end points of the segment. At the start point of the segment, point cloud marking is performed on n vehicles in multiple lanes. Each point cloud is bound to the outline and color of the marked vehicle. Multiple point clouds in the same lane are associated to obtain multiple association chains. The migration unit is used to set migration conditions, migrate the point cloud based on the migration conditions, and bind the point cloud with the outline and color of the newly marked vehicle after migration. It generates road fault labels based on the change information of the point cloud in multiple association chains. The migration condition is: the vehicle displacement corresponding to the current point cloud is less than a preset displacement threshold. The calculation unit is used to determine the moving speed, moving distance and moving time of multiple point clouds within the segment, and to obtain real-time traffic flow based on the moving speed, moving distance and moving time of multiple point clouds within the segment.

7. A traffic anomaly section identification system based on traffic flow as described in claim 6, characterized in that, The step of generating road fault labels based on the change information of point clouds in multiple association chains includes: Add change information to the migrating point cloud, including lane change information and movement information; Multiple road fault labels are set, and the road fault labels corresponding to the changed information are added to the corresponding abnormal sections. The road fault labels include single-point faults, multi-point faults, and all faults.

8. A method for identifying abnormal traffic sections based on traffic flow, used to implement the traffic abnormality section identification system based on traffic flow as described in any one of claims 1-7, characterized in that, Includes the following steps: Step 1: Collect basic data for each segment and build a basic model for the corresponding segment based on the basic data; Step 2: Determine the normal traffic flow for the corresponding section based on historical traffic flow and the basic model; Step 3: Set update conditions, and update the basic model and normal traffic flow of the section when the update conditions are met; Step 4: Collect real-time traffic flow data for each section, compare the real-time traffic flow with the normal traffic flow for the corresponding section, and identify sections that are not in normal traffic flow as abnormal sections.

Images