Real-time traffic flow abnormal event tracing method and system based on edge calculation

Abstract

The invention discloses a real-time traffic flow abnormal event tracing method and system based on edge calculation, and the method comprises the steps: obtaining traffic flow data of edge nodes, analyzing the dynamic impedance of a road segment according to the traffic flow data, calculating the path selection probability of each path through employing an abnormal event wave proportion, extracting a traffic flow abnormal road segment, and carrying out the tracing of the traffic flow abnormal event. And performing spatial overlapping analysis on the abnormal road sections according to the path selection probability, matching the traffic event announcement root cause and the meteorological early warning root cause based on the association strength of each abnormal road section, generating an optimal matching root cause through a maximum weight bisection matching algorithm, and tracing the upstream traffic flow by using the optimal matching root cause and the path selection probability. According to the method, dynamic impedance analysis, SIR propagation simulation and Shapley value are utilized to quantify the contribution degree of the root cause to the abnormal road section, full-link traceability of the root cause, the path and the upstream node is formed, relatively complete data support is provided for traffic abnormity diffusion caused by traffic abnormity and meteorological abnormity, and the practicability and decision value of a traceability result are improved.

Description

Technical Field

[0001] This invention relates to the field of traffic flow anomaly event tracing, and in particular to a real-time traffic flow anomaly event tracing method and system based on edge computing. Background Technology

[0002] With the acceleration of urbanization and the continuous growth of motor vehicle ownership, traffic congestion, accidents and other abnormal events occur frequently, affecting road network efficiency and travel safety. Real-time and accurate source tracing of abnormal traffic flow events has become a core requirement of intelligent transportation systems such as ITS. With the popularization of edge computing technology, edge nodes such as roadside units and vehicle terminals can easily obtain multi-source reusable traffic data, providing a rich data foundation for source tracing analysis. At present, traffic flow anomaly source tracing is widely used in urban traffic management, congestion cause analysis, emergency response and other scenarios. By tracing the upstream traffic flow source of abnormal road sections and identifying the impact path, it can provide support for traffic signal optimization, temporary control decisions and travel guidance.

[0003] Current traffic flow tracing relies on manually linking traffic event announcements or meteorological data, leading to discrepancies between path flow allocation and actual road network conditions. Actual tracing primarily focuses on processing vehicle and traffic control information, with limited attention paid to meteorological information and the spread of traffic anomalies caused by abnormal weather. This makes it easy to misjudge or underjudge the contribution of different root causes to abnormal road segments. While it mainly traces the current traffic flow on abnormal road segments, it fails to establish a complete tracing chain from anomaly identification and root cause matching to path selection, thus failing to effectively support refined traffic management decisions. Therefore, how to assess abnormal events and match the complete tracing chain has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for tracing the source of abnormal traffic flow events based on edge computing.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: The first aspect of this invention provides a method for tracing the source of real-time traffic flow anomalies based on edge computing, comprising: Traffic flow data of edge nodes is obtained, and the traffic flow data is preprocessed. The traffic flow data includes origin-destination (OD) data, navigation path, road segment speed, traffic incident announcements, and weather warnings. The dynamic impedance of the road segment is analyzed based on the vehicle speed and origin-destination OD data of the road segment. Based on the dynamic impedance, the abnormal event waves of each navigation path are divided by meteorological warnings. The path selection probability of each path is calculated using the proportion of abnormal event waves. Based on the vehicle speed of the road segment and the corresponding smooth traffic baseline speed, abnormal traffic flow segments are extracted. Spatial overlap analysis is performed on the abnormal road segments according to the path selection probability to obtain the correlation strength of the abnormal road segments. Based on the correlation strength of each abnormal road segment, the root causes of traffic incident announcements and meteorological warnings are matched to obtain the contribution of each abnormal root cause to the road segment. Based on the abnormal road segments and root causes, the optimal matching root cause is generated using the maximum weighted binary matching algorithm according to the contribution. The upstream traffic flow is traced using the optimal matching root cause and the path selection probability, and the influencing path is screened based on the correlation strength to obtain the source tracing results.

[0006] Furthermore, the method for obtaining the path selection probability includes: Based on the road network, the connection relationship between road segments is taken as the road network edge, and the road segment is taken as the road network node to obtain a complex network. According to the complex network, the road segment speed, road segment length and origin-destination (OD) data of the corresponding road segment are obtained. The divisor of the road segment length and road segment speed is taken as the free flow time. According to the flow distribution of the free flow time and origin-destination (OD) data, the dynamic impedance is calculated by the BPR function. The dynamic impedance is taken as the weight of the complex network node. If any two segments in a complex network are directly connected, the product of their dynamic impedances is used as the connection weight. If they are not directly connected, the connection weight is set to zero, resulting in a square adjacency matrix. The sum of the dynamic impedance weights of each segment and its corresponding connected segments is used as the diagonal element of the degree matrix. The Laplace matrix is ​​obtained based on the difference between the degree matrix and the adjacency matrix. The Laplace matrix is ​​then decomposed into eigenvalues. The resulting eigenvectors are then subjected to dimensionality reduction and K-means clustering by row to identify anomalous communities. The dynamic impedance aggregation value of road segments within the abnormal community is used as the influence of the abnormal community. The infection simulation of the abnormal community is carried out through the SIR propagation model to obtain the abnormal event wave. Road segments with dynamic impedance greater than the preset dynamic impedance threshold are marked as abnormal states. Based on the abnormal event wave, the ratio of the length of road segments in abnormal states to the total length of the navigation path is statistically calculated. The ratio, dynamic impedance, and abnormal event wave intensity are used as the influencing factors of path utility, and a path utility function is constructed. Based on the path utility function, the path utility is converted into path selection probability through the Logit model.

[0007] Furthermore, the method for simulating infection in abnormal communities using the SIR propagation model includes: Based on the navigation path, the path directly affected by the weather warning is taken as the initial infection path. Based on the impact, the number of paths without anomalies is taken as the susceptible state number of the path, the number of paths already affected by the anomaly is taken as the infection state number, the rate at which the anomaly spreads from the susceptible path to the infected path is taken as the infection rate, and the rate at which the infected path recovers to its normal state is taken as the recovery rate. Based on the initial infection path, the number of susceptible states, the number of infected states, the infection rate, and the recovery rate, the SIR propagation model is used to perform differential infection on the anomalous community, obtaining the infection state evolution curves for each navigation path at different time steps. The differential infection formula is: ; in The rate of change of the number of infection routes over time. For a moment path The number of infection statuses For infection rate, For path The number of susceptible states, Recovery rate; Based on the infection state evolution curve, the spatial influence range and temporal evolution state of the infection are treated as an anomalous event wave. The formula for calculating the temporal evolution state is as follows: ; in For a moment path The intensity of the time-evolution state, i.e., the intensity of the anomalous event wave. This is an abnormal community. For indicator functions, if the path Belongs to an abnormal community The value is 1 if it is positive and 0 otherwise. For abnormal communities The influence For path The section of road in the middle, For a moment Time Section The dynamic impedance, The preset dynamic impedance threshold for road segment anomalies is 1.5 times the free-flow time. For path Total length, For path The total number of road segments.

[0008] Furthermore, the method for obtaining the correlation strength includes: Based on traffic flow data, segment vehicle speeds and corresponding free-flow baseline vehicle speeds are obtained. If a segment's vehicle speed accounts for less than or equal to 60% of the free-flow baseline speed, it is considered an abnormal segment in the traffic flow. For any two abnormal segments, the navigation paths they simultaneously traverse are statistically analyzed to obtain a path set. Based on the path selection probability, spatial overlap analysis is performed on the abnormal segments according to the path set, and the correlation strength of the abnormal segments is calculated. The correlation strength calculation formula is as follows: ; ; in For road section Effective capacity, For road section The design capacity of the passageway For road section The dynamic impedance, This is the dynamic impedance correction factor, calibrated from traffic flow data using the least squares method. For a moment Lower section and road section The strength of the association, For road section Effective capacity, For road section Effective capacity, For a set of paths, For a moment path The probability of path selection. In the path Middle section and road sections The distance between them This is the distance attenuation coefficient. Section Inflow of traffic, Section Outflow rate.

[0009] Furthermore, the method for obtaining the contribution includes: The root causes of meteorological warnings in traffic incident announcements are obtained. The timestamp, geographical location, and time difference between the root cause's occurrence and the road segment anomaly are used as root cause attributes. Based on these attributes, the root causes of traffic incident announcements and meteorological warnings are matched according to the correlation strength of each abnormal road segment, maximizing the matching weight to obtain the most relevant set of root causes. The formula for calculating the abnormal root cause matching is as follows: ; in root cause and road section The temporal correlation is calculated based on the overlap ratio between the root cause occurrence time and the road segment anomaly time. As the weight of time correlation, root cause Geographical location and road section Spatial distance, For spatial correlation weight, root cause The response time, i.e., the time difference between the occurrence of the root cause and the road segment anomaly, For response correlation weights; The contribution of each abnormal root cause to the road segment is calculated based on the Shapley value of the root cause set. The contribution calculation formula is as follows: ; in Shapley value representing the contribution of abnormal root causes. For the root cause set, for It does not contain the root cause. a subset of For subset The sum of the correlation strength of the corresponding abnormal road segments For subset The number of root causes included. This represents the total number of root cause sets.

[0010] Furthermore, the method for obtaining the optimal matching root cause includes: An abnormal road segment set is obtained based on abnormal road segments. A candidate root cause set is obtained based on the root causes of traffic incident announcements and weather warnings. The abnormal road segment set and the candidate root cause set are used as two vertex sets of a bipartite graph. The contribution of the root cause is used as the edge weight of the abnormal road segment and the candidate root cause in the bipartite graph. The Hungarian algorithm is used to find the matching maximum edge weight in the bipartite graph. Augmenting paths are iteratively searched until the augmenting paths are empty. The sum of the edge weights of the matching edges is maximized, and the optimal matching root cause is generated. The Hungarian algorithm is applicable to bipartite graph matching where the number of abnormal road segments and the number of candidate root causes are equal or unequal.

[0011] Furthermore, the method for obtaining the source tracing results includes: The optimal matching root cause is used to extract the navigation path and path selection probability under the influence of the corresponding root cause, generating an association table of root cause, path, and probability. Based on the associated paths in the association table, the upstream road segments are traversed backwards from the abnormal road segment to obtain the dynamic impedance and flow distribution of the upstream road segments, and the traffic flow contribution of each upstream road segment is calculated. The formula for calculating the traffic flow contribution is as follows: ; in upstream section Traffic contribution For passing through the upstream section The set of paths For path Middle section Traffic, For path Total flow; Road segments with a traffic flow contribution of ≥30% are designated as core upstream nodes. The correlation strength between core upstream nodes and abnormal road segments is calculated, and a probability density distribution of the correlation strength is generated. Based on the probability density distribution, the core influencing path is obtained by screening using the upper quartile method. The core influencing path and the correlation table are used as the source tracing results.

[0012] A second aspect of the present invention provides a real-time traffic flow anomaly event tracing system based on edge computing, comprising: Data acquisition module: used to obtain traffic flow data of edge nodes, and preprocess the traffic flow data, which includes origin-destination (OD) data, navigation path, road segment speed, traffic incident announcements and weather warnings; Path selection module: used to analyze the dynamic impedance of the road segment based on the vehicle speed and origin-destination OD data of the road segment, divide the abnormal event waves of each navigation path based on the dynamic impedance and the meteorological warning, and calculate the path selection probability of each path using the proportion of abnormal event waves. Spatial overlap analysis module: used to extract abnormal traffic flow segments based on the vehicle speed of the road segment and the corresponding smooth baseline vehicle speed, and to perform spatial overlap analysis on the abnormal road segments according to the path selection probability to obtain the correlation strength of the abnormal road segments; Anomaly Root Cause Matching Module: Used to match the root causes of traffic incident announcements and weather warnings based on the correlation strength of each abnormal road segment, and obtain the contribution of each abnormal root cause to the road segment. Source tracing module: Based on abnormal road segments and abnormal root causes, the module generates the optimal matching root cause using the maximum weighted binary matching algorithm according to the contribution of the abnormal root causes. It then uses the optimal matching root cause and path selection probability to trace upstream traffic flow and filters the influencing paths based on the correlation strength to obtain the source tracing results.

[0013] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: This invention characterizes the spatial propagation and temporal evolution of traffic anomalies by combining dynamic impedance analysis with the SIR propagation model, avoiding the lag of static analysis. By analyzing the spread of traffic anomalies caused by meteorological changes, it improves the timeliness and spatial accuracy of anomaly wave identification. By quantifying the contribution of root causes to abnormal road segments using Shapley values, it reduces the limitations of single correlation matching and enhances the interpretability of location for event tracing in scenarios with multiple anomaly root causes. The Hungarian algorithm is used to perform optimal matching between abnormal road segments and root causes, obtaining upstream traffic flow tracing results and correlation strength statistical screening, forming a full-link tracing of root causes, paths, and upstream nodes. This provides relatively complete data support for the rapid handling of traffic anomalies and improves the practicality and decision-making value of the tracing results. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps of the real-time traffic flow anomaly event tracing method based on edge computing in an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0016] Reference Figure 1 As shown, this invention provides a method for tracing the source of real-time traffic flow anomalies based on edge computing, including: Traffic flow data of edge nodes is obtained, and the traffic flow data is preprocessed. The traffic flow data includes origin-destination (OD) data, navigation path, road segment speed, traffic incident announcements, and weather warnings. In the actual assessment, traffic flow data was obtained from edge computing nodes at 12 key traffic nodes in a region. Origin-destination (OD) data was obtained through the park's freight management system, covering the morning peak from 7:30 to 9:00. Major origin-destination pairs included Terminal A to Warehouse Area B and Terminal C to the park's west exit. The average flow rate per OD pair was 45-60 vehicles per 10 minutes. This included 6 core freight navigation routes, with route lengths ranging from 2.3 to 4.8 kilometers. Navigation route data included route ID, sequence of road segments, and free-flow time. Vehicle speeds were obtained using radar equipment. The baseline speed for smooth traffic was 60 km / h. The morning peak congestion period was from 8:00 to 8:40, during which speeds on some sections dropped to 15-25 km / h. During abnormal periods such as accidents or flooding, the minimum speed could reach 7-9 km / h. Traffic incident announcements were simultaneously obtained through the park's traffic command platform, such as "Container vehicle malfunction blocking lane (Location: Main road K2+300 meters, Time: 8:10)." "Temporary road closure for loading and unloading operations at the dock (Location: D exit of the dock, Time: 14:00-15:35)" includes event type, location, scope of impact, and duration. The weather warning area has wind speeds of 0-12 m / s, rainfall of 0-50 mm / h, and visibility of 100-2000 meters. Traffic flow data is preprocessed in the cloud through edge nodes to remove outliers such as vehicle speeds less than 0 km / h and rainfall of equal negative values ​​caused by sensor interference. Linear interpolation is used to supplement missing data, with a single indicator missing rate of ≤3%. Data collected by different devices are unified with timestamps, and spatial information is unified to the WGS-84 coordinate system to ensure the spatiotemporal consistency of road segment vehicle speed, OD flow, and event location. GPS path data is matched to the park's road network topology to generate a road segment ID association table. High-frequency vehicle speed data is smoothed by sliding window, retaining key feature values ​​such as average vehicle speed and vehicle speed variance. The dynamic impedance of the road segment is analyzed based on the vehicle speed and origin-destination OD data of the road segment. Based on the dynamic impedance, the abnormal event waves of each navigation path are divided by meteorological warnings. The path selection probability of each path is calculated using the proportion of abnormal event waves. In the actual assessment, a complex network was constructed using six core road segments (L1-L6) as nodes and the connections between these segments as edges. The L1 main road is 2km long with a designed capacity of 2000 vehicles / hour. The free-flow time at 60km / h is 2 minutes, and the real-time morning peak traffic volume is 1200 vehicles / hour. Dynamic impedance was calculated based on the BPR function. ;,in For free-flow time, for Dynamic impedance at time t, To design traffic capacity, For real-time traffic flow, the dynamic impedance of L1 is 2.19 minutes. The calculated dynamic impedance range for L2-L6 is 1.8-3.2 minutes. L1 and L2 are directly connected road segments, and their connection weight is the product of dynamic impedances, which is 5.04. The weight of non-directly connected road segments is 0, forming a 6×6 adjacency matrix. After decomposing the Laplace matrix eigenvalues, abnormal communities are divided by K-means clustering (K=2). L1, L2, and L3, with high dynamic impedances ≥2.0 minutes, are classified as "congested community C1," while L4-L6 are classified as "normal community C2." Based on meteorological data, communities affected by heavy rain are selected. The path P1 affected by the traffic jam passes through L1 and L2. Taking the 150 potential peak-hour routes as susceptible paths and the 20 already congested paths as initial infected paths, the infection rate is 0.2. Every 10 minutes, one infected path's traffic can infect 0.2 susceptible path traffic. The recovery rate is 0.1, meaning every 10 minutes, one infected path's traffic has a 0.1 probability of returning to normal. Using SIR differentiation, it is calculated that after 30 minutes of simulation, P1 has 85 infected paths and an abnormal event wave intensity of 0.72. Three alternative paths, P1, P2, and P3, are selected. The utility of path P1 is: ;in Given a total dynamic impedance of path P1 of 4.5 minutes and an anomalous event wave proportion of 0.57, the calculated utility of P1 is -2.78. Therefore, the path selection probability obtained through the Logit model is: ;in The probability of path selection for P2 and P3 is 0.22, and the probabilities are 0.35 and 0.43, respectively. Based on the vehicle speed of the road segment and the corresponding smooth traffic baseline speed, abnormal traffic flow segments are extracted. Spatial overlap analysis is performed on the abnormal road segments according to the path selection probability to obtain the correlation strength of the abnormal road segments. In the actual assessment, the real-time vehicle speed of the road segment was 28 km / h, accounting for 46.7% of the baseline vehicle speed. Since it was less than 60%, it was marked as an abnormal road segment. The real-time vehicle speed of road segment L2 was 32 km / h, accounting for 53.3% of the baseline vehicle speed, and it was also marked as an abnormal road segment. The real-time vehicle speed of road segment L3 was 45 km / h, accounting for 75% of the baseline vehicle speed, and it was not marked as abnormal. L1 and L2 were used as the set of abnormal road segments. Navigation paths that passed through both L1 and L2 were selected as three paths: P1, P2, and P3. The effective capacity of abnormal road segment L1 was calculated to be 1230 vehicles / hour, where the dynamic impedance correction coefficient k was calibrated to 0.08 using the least squares method. The effective capacity of L2 was 1008 vehicles / hour. Substituting these values ​​into the correlation strength, where the distance between L1 and L2 in the path is 1.2 km, the distance attenuation coefficient is 1.5, the inflow of L1 is 1000 vehicles / hour, and the outflow of L2 is 900 vehicles / hour, the normalized correlation strength is 0.84. Based on the correlation strength of each abnormal road segment, the root causes of traffic incident announcements and meteorological warnings are matched to obtain the contribution of each abnormal root cause to the road segment. In the actual assessment, a candidate root cause set was extracted, consisting of two core root causes. Root cause b1 was a container truck malfunction blocking the lane, occurring at 8:05 AM, which overlapped with the abnormal morning rush hour period of 8:00-9:00 AM on L1. Geographically, it was located at K2+300 meters on L1, with latitude and longitude of 37.56°N, 121.23°E. At 8:15 AM, 10 minutes after the malfunction, L1 showed significant anomalies, hence the response time was 10 minutes. Root cause b2 was a short-term heavy rain warning, issued at [time stamp missing]. At 7:50, an early warning was issued, coinciding with an abnormal time period for the road segment. The warning's impact center was located in the area between L1 and L2, at 37.55°N, 121.22°E. At 8:05, 15 minutes after the warning was issued, a decrease in vehicle speed occurred at L2. The most relevant root causes were selected based on maximizing the weight of the root cause-road segment matching, with a time correlation weight of 0.5, a spatial correlation weight of 0.3, and a response correlation weight of 0.2. The time correlation was calculated based on the overlap ratio between the root cause's time and the abnormal time of the road segment. If b1 overlaps with L1 by 80%, its weight is 0.8; if it overlaps with L2 by 60%, b2 overlaps with L1 by 70% and with L2 by 90%. Spatial distances are calculated based on actual road network distances. If the starting point of L1 is K2+000 (37.57°N, 121.24°E) and the starting point of L2 is K3+500 (37.54°N, 121.21°E), then the distance weights are 300 meters for b1 and L2, 1800 meters for b2 and L1, and 1200 meters for b2 and L2 respectively. 900 meters, converted to 0.3, 1.8, 1.2, 0.9 and calculated matching weights, where the matching result of root cause b1 with L1 is -1.69, the response time is normalized to "minutes / 60", and corrected to 0.266, root cause b1 with L2 is -0.29, root cause b2 with L1 is -0.06, root cause b2 with L2 is 0.13, and combinations with matching weights ≥ -0.1 are selected to determine root cause set B, which includes b1 and b2. Based on the root cause set, the root cause contribution is calculated using Shapley values, with root cause b1 contributing 56% and root cause b2 contributing 44%. Based on the abnormal road segments and root causes, the optimal matching root cause is generated using the maximum weighted binary matching algorithm according to the contribution. The upstream traffic flow is traced using the optimal matching root cause and the path selection probability, and the influencing path is screened based on the correlation strength to obtain the source tracing results.

[0017] In the actual evaluation, the abnormal road segment set (L1, L2) and the abnormal root cause set (b1, b2) are used. The abnormal road segment set is taken as the left vertex of the bipartite graph, and the abnormal root cause set is taken as the right vertex. The contribution is taken as the edge weight of the bipartite graph. The maximum weight matching is solved by the Hungarian algorithm. Since the number of abnormal road segments is 2, which is equal to the number of candidate root factors, the standard Hungarian algorithm is used for iteration. The first step is to subtract the minimum value of each row of the weight matrix to eliminate negative bias. The minimum value of the L1 row is subtracted by 120000, and the minimum value of the L2 row is subtracted by 95000, resulting in a simplified matrix. The second step is to find independent zero elements so that each row and each column has only one zero. In the simplified matrix, 115123 in L1-b1 and 90123 in L2-b2 are independent non-zero elements. The element has a maximum total weight of 0.84. When no augmenting path can be found, the optimal matching root cause is output as truck malfunction (L1→b1) and rainstorm warning (L2→b2). If the number of abnormal road segments is not equal to the number of root causes (e.g., 3 road segments and 2 root causes), the navigation path corresponding to the optimal matching root cause is extracted through a virtual root cause completion matrix: Root cause b1 is associated with L1, affecting path P1 (L1 to warehouse area A) and path P2 (L1 to dock B), with path selection probabilities of 0.25 and 0.3 respectively. Root cause b2 affects paths P3 (L2 to dock C) and P4 (L2 to park exit), with path selection probabilities of 0.2 and 0.25 respectively. A root cause, path, and probability association table is generated. The association table is as follows:

[0018] Starting from abnormal road segments L1 and L2, traverse the upstream road segments in reverse along the associated path. U1 is a 1.2km upstream segment of L1, U2 is a 0.8km upstream segment of L1, U3 is a 1.5km upstream segment of L2, and U4 is a 1.0km upstream segment of L2. The dynamic impedance and flow distribution of the upstream road segments are as follows:

[0019] Calculate the traffic flow contribution of each upstream road segment: upstream segment U1 has a contribution of 6.25%, upstream segment U2 has a contribution of 9%, and upstream segment U3 has a contribution of 12%. Add another path P5 to upstream segment U4, with a path selection probability of 0.1. The traffic flow of U4 in P5 is 450, and the total traffic flow of P5 is 1500. Therefore, the traffic flow contribution of upstream segment U4 is 9.25%. Road segments with a traffic flow contribution ≥ 30% are designated as core upstream nodes, with the key upstream segment U5 being the upstream of L1, specifically associated with path P1. In P1, the traffic flow of U5 is 900, and in P2, the traffic flow of U5 is 840. The calculated traffic contribution is 39.75%, and U5 is identified as the core upstream node. The correlation strength between the core upstream node U5 and the abnormal road segments L1 and L2 is calculated. The effective capacity of U5 and L1 is 1800 vehicles / hour, the corresponding design capacity is 2000 vehicles / hour, the dynamic impedance is 1.8 minutes, the effective capacity of L1 is 1230 vehicles / hour, the path overlap probability is 0.85, and the distance attenuation is... The term is 0.61, the minimum inflow and outflow is 1200 vehicles / hour, the correlation strength between U5 and L1 is calculated to be 0.94, the correlation strength between U5 and L2, and other upstream nodes and abnormal road segments are obtained, resulting in a correlation strength set (0.94, 0.38, 0.27, 0.23, 0.02). After sorting the correlation strength set, the upper quartile Q3 is calculated to be 0.45. Paths with a correlation strength greater than or equal to Q3 are retained, namely P1, specifically U5→L1→Warehouse Area A, and P2 is U5→L1→Barracks. Head B is identified as the core impact path. The structured results are integrated and a source tracing report is output. The optimal matching root cause is: L1 anomaly is caused by "truck malfunction (b1)" and L2 anomaly is caused by "rainstorm warning (b2)". The core upstream node is U5 (the main road upstream of L1, with a traffic flow contribution of 39.75%). The core impact paths are P1 (selection probability 0.25) and P2 (selection probability 0.3). The control recommendations are to prioritize diverting traffic on the U5 section and temporarily open the emergency lane next to L1 to alleviate the impact of the malfunction blocking the road.

[0020] In this embodiment, the method for obtaining the path selection probability includes: Based on the road network, the connection relationship between road segments is taken as the road network edge, and the road segment is taken as the road network node to obtain a complex network. According to the complex network, the road segment speed, road segment length and origin-destination (OD) data of the corresponding road segment are obtained. The divisor of the road segment length and road segment speed is taken as the free flow time. According to the flow distribution of the free flow time and origin-destination (OD) data, the dynamic impedance is calculated by the BPR function. The dynamic impedance is taken as the weight of the complex network node. If any two segments in a complex network are directly connected, the product of their dynamic impedances is used as the connection weight. If they are not directly connected, the connection weight is set to zero, resulting in a square adjacency matrix. The sum of the dynamic impedance weights of each segment and its corresponding connected segments is used as the diagonal element of the degree matrix. The Laplace matrix is ​​obtained based on the difference between the degree matrix and the adjacency matrix. The Laplace matrix is ​​then decomposed into eigenvalues. The resulting eigenvectors are then subjected to dimensionality reduction and K-means clustering by row to identify anomalous communities. The dynamic impedance aggregation value of road segments within the abnormal community is used as the influence of the abnormal community. The infection simulation of the abnormal community is carried out through the SIR propagation model to obtain the abnormal event wave. Road segments with dynamic impedance greater than the preset dynamic impedance threshold are marked as abnormal states. Based on the abnormal event wave, the ratio of the length of road segments in abnormal states to the total length of the navigation path is statistically calculated. The ratio, dynamic impedance, and abnormal event wave intensity are used as the influencing factors of path utility, and a path utility function is constructed. Based on the path utility function, the path utility is converted into path selection probability through the Logit model.

[0021] In this embodiment, the method for simulating infection in abnormal communities using the SIR propagation model includes: Based on the navigation path, the path directly affected by the weather warning is taken as the initial infection path. Based on the impact, the number of paths without anomalies is taken as the susceptible state number of the path, the number of paths already affected by the anomaly is taken as the infection state number, the rate at which the anomaly spreads from the susceptible path to the infected path is taken as the infection rate, and the rate at which the infected path recovers to its normal state is taken as the recovery rate. Based on the initial infection path, the number of susceptible states, the number of infected states, the infection rate, and the recovery rate, the SIR propagation model is used to perform differential infection on the anomalous community, obtaining the infection state evolution curves for each navigation path at different time steps. The differential infection formula is: ; in The rate of change of the number of infection routes over time. For a moment path The number of infection statuses For infection rate, For path The number of susceptible states, Recovery rate; Based on the infection state evolution curve, the spatial influence range and temporal evolution state of the infection are treated as an anomalous event wave. The formula for calculating the temporal evolution state is as follows: ; in For a moment path The intensity of the time-evolution state, i.e., the intensity of the anomalous event wave. This is an abnormal community. For indicator functions, if the path Belongs to an abnormal community The value is 1 if it is positive and 0 otherwise. For abnormal communities The influence For path The section of road in the middle, For a moment Time Section The dynamic impedance, The preset dynamic impedance threshold for road segment anomalies is 1.5 times the free-flow time. For path Total length, For path The total number of road segments.

[0022] In this embodiment, the method for obtaining the correlation strength includes: Based on traffic flow data, segment vehicle speeds and corresponding free-flow baseline vehicle speeds are obtained. If a segment's vehicle speed accounts for less than or equal to 60% of the free-flow baseline speed, it is considered an abnormal segment in the traffic flow. For any two abnormal segments, the navigation paths they simultaneously traverse are statistically analyzed to obtain a path set. Based on the path selection probability, spatial overlap analysis is performed on the abnormal segments according to the path set, and the correlation strength of the abnormal segments is calculated. The correlation strength calculation formula is as follows: ; ; in For road section Effective capacity, For road section The design capacity of the passageway For road section The dynamic impedance, This is the dynamic impedance correction factor, calibrated from traffic flow data using the least squares method. For a moment Lower section and road section The strength of the association, For road section Effective capacity, For road section Effective capacity, For a set of paths, For a moment path The probability of path selection. In the path Middle section and road sections The distance between them This is the distance attenuation coefficient. Section Inflow of traffic, Section Outflow rate.

[0023] In this embodiment, the method for obtaining the contribution level includes: The root causes of meteorological warnings in traffic incident announcements are obtained. The timestamp, geographical location, and time difference between the root cause's occurrence and the road segment anomaly are used as root cause attributes. Based on these attributes, the root causes of traffic incident announcements and meteorological warnings are matched according to the correlation strength of each abnormal road segment, maximizing the matching weight to obtain the most relevant set of root causes. The formula for calculating the abnormal root cause matching is as follows: ; in root cause and road section The temporal correlation is calculated based on the overlap ratio between the root cause occurrence time and the road segment anomaly time. As the weight of time correlation, root cause Geographical location and road section Spatial distance, For spatial correlation weight, root cause The response time, i.e., the time difference between the occurrence of the root cause and the road segment anomaly, For response correlation weights; The contribution of each abnormal root cause to the road segment is calculated based on the Shapley value of the root cause set. The contribution calculation formula is as follows: ; in Shapley value representing the contribution of abnormal root causes. For the root cause set, for It does not contain the root cause. a subset of For subset The sum of the correlation strength of the corresponding abnormal road segments For subset The number of root causes included. This represents the total number of root cause sets.

[0024] In this embodiment, the method for obtaining the optimal matching root cause includes: An abnormal road segment set is obtained based on abnormal road segments. A candidate root cause set is obtained based on the root causes of traffic incident announcements and weather warnings. The abnormal road segment set and the candidate root cause set are used as two vertex sets of a bipartite graph. The contribution of the root cause is used as the edge weight of the abnormal road segment and the candidate root cause in the bipartite graph. The Hungarian algorithm is used to find the matching maximum edge weight in the bipartite graph. Augmenting paths are iteratively searched until the augmenting paths are empty. The sum of the edge weights of the matching edges is maximized, and the optimal matching root cause is generated. The Hungarian algorithm is applicable to bipartite graph matching where the number of abnormal road segments and the number of candidate root causes are equal or unequal.

[0025] In this embodiment, the method for obtaining the tracing result includes: The optimal matching root cause is used to extract the navigation path and path selection probability under the influence of the corresponding root cause, generating an association table of root cause, path, and probability. Based on the associated paths in the association table, the upstream road segments are traversed backwards from the abnormal road segment to obtain the dynamic impedance and flow distribution of the upstream road segments, and the traffic flow contribution of each upstream road segment is calculated. The formula for calculating the traffic flow contribution is as follows: ; in upstream section Traffic contribution For passing through the upstream section The set of paths For path Middle section Traffic, For path Total flow; Road segments with a traffic flow contribution of ≥30% are designated as core upstream nodes. The correlation strength between core upstream nodes and abnormal road segments is calculated, and a probability density distribution of the correlation strength is generated. Based on the probability density distribution, the core influencing path is obtained by screening using the upper quartile method. The core influencing path and the correlation table are used as the source tracing results.

[0026] A second aspect of the present invention also provides a real-time traffic flow anomaly event tracing system based on edge computing, comprising: Data acquisition module: used to obtain traffic flow data of edge nodes, and preprocess the traffic flow data, which includes origin-destination (OD) data, navigation path, road segment speed, traffic incident announcements and weather warnings; Path selection module: used to analyze the dynamic impedance of the road segment based on the vehicle speed and origin-destination OD data of the road segment, divide the abnormal event waves of each navigation path based on the dynamic impedance and the meteorological warning, and calculate the path selection probability of each path using the proportion of abnormal event waves. Spatial overlap analysis module: used to extract abnormal traffic flow segments based on the vehicle speed of the road segment and the corresponding smooth baseline vehicle speed, and to perform spatial overlap analysis on the abnormal road segments according to the path selection probability to obtain the correlation strength of the abnormal road segments; Anomaly Root Cause Matching Module: Used to match the root causes of traffic incident announcements and weather warnings based on the correlation strength of each abnormal road segment, and obtain the contribution of each abnormal root cause to the road segment. Source tracing module: Based on abnormal road segments and abnormal root causes, the module generates the optimal matching root cause using the maximum weighted binary matching algorithm according to the contribution of the abnormal root causes. It then uses the optimal matching root cause and path selection probability to trace upstream traffic flow and filters the influencing paths based on the correlation strength to obtain the source tracing results.

[0027] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for tracing the source of real-time traffic flow anomalies based on edge computing, characterized in that, Includes the following steps: Traffic flow data of edge nodes is obtained, and the traffic flow data is preprocessed. The traffic flow data includes origin-destination (OD) data, navigation path, road segment speed, traffic incident announcements, and weather warnings. The dynamic impedance of the road segment is analyzed based on the vehicle speed and origin-destination OD data of the road segment. Based on the dynamic impedance, the abnormal event waves of each navigation path are divided by meteorological warnings. The path selection probability of each path is calculated using the proportion of abnormal event waves. Based on the vehicle speed of the road segment and the corresponding smooth traffic baseline speed, abnormal traffic flow segments are extracted. Spatial overlap analysis is performed on the abnormal road segments according to the path selection probability to obtain the correlation strength of the abnormal road segments. Based on the correlation strength of each abnormal road segment, the root causes of traffic incident announcements and meteorological warnings are matched to obtain the contribution of each abnormal root cause to the road segment. Based on the abnormal road segments and root causes, the optimal matching root cause is generated using the maximum weighted binary matching algorithm according to the contribution. The upstream traffic flow is traced using the optimal matching root cause and the path selection probability, and the influencing path is screened based on the correlation strength to obtain the source tracing results.

2. The method for tracing the source of real-time traffic flow anomalies based on edge computing according to claim 1, characterized in that, The method for obtaining the path selection probability includes: Based on the road network, the connection relationship between road segments is taken as the road network edge, and the road segment is taken as the road network node to obtain a complex network. According to the complex network, the road segment speed, road segment length and origin-destination (OD) data of the corresponding road segment are obtained. The divisor of the road segment length and road segment speed is taken as the free flow time. According to the flow distribution of the free flow time and origin-destination (OD) data, the dynamic impedance is calculated by the BPR function. The dynamic impedance is taken as the weight of the complex network node. If any two segments in a complex network are directly connected, the product of their dynamic impedances is used as the connection weight. If they are not directly connected, the connection weight is set to zero, resulting in a square adjacency matrix. The sum of the dynamic impedance weights of each segment and its corresponding connected segments is used as the diagonal element of the degree matrix. The Laplace matrix is ​​obtained based on the difference between the degree matrix and the adjacency matrix. The Laplace matrix is ​​then decomposed into eigenvalues. The resulting eigenvectors are then subjected to dimensionality reduction and K-means clustering by row to identify anomalous communities. The dynamic impedance aggregation value of road segments within the abnormal community is used as the influence of the abnormal community. The infection simulation of the abnormal community is carried out through the SIR propagation model to obtain the abnormal event wave. Road segments with dynamic impedance greater than the preset dynamic impedance threshold are marked as abnormal states. Based on the abnormal event wave, the ratio of the length of road segments in abnormal states to the total length of the navigation path is statistically calculated. The ratio, dynamic impedance, and abnormal event wave intensity are used as the influencing factors of path utility, and a path utility function is constructed. Based on the path utility function, the path utility is converted into path selection probability through the Logit model.

3. The method for tracing the source of real-time traffic flow anomalies based on edge computing according to claim 2, characterized in that, The method for simulating infection in abnormal communities using the SIR propagation model includes: Based on the navigation path, the path directly affected by the weather warning is taken as the initial infection path. Based on the impact, the number of paths without anomalies is taken as the susceptible state number of the path, the number of paths already affected by the anomaly is taken as the infection state number, the rate at which the anomaly spreads from the susceptible path to the infected path is taken as the infection rate, and the rate at which the infected path recovers to its normal state is taken as the recovery rate. Based on the initial infection path, the number of susceptible states, the number of infected states, the infection rate, and the recovery rate, the SIR propagation model is used to perform differential infection on the anomalous community, obtaining the infection state evolution curves for each navigation path at different time steps. The differential infection formula is: ; in The rate of change of the number of infection routes over time. For a moment path The number of infection statuses For infection rate, For path The number of susceptible states, Recovery rate; Based on the infection state evolution curve, the spatial influence range and temporal evolution state of the infection are treated as an anomalous event wave. The formula for calculating the temporal evolution state is as follows: ; in For a moment path The intensity of the time-evolution state, i.e., the intensity of the anomalous event wave. This is an abnormal community. For indicator functions, if the path Belongs to an abnormal community The value is 1 if it is positive and 0 otherwise. For abnormal communities The influence For path The section of road in the middle, For a moment Time Section The dynamic impedance, The preset dynamic impedance threshold for road segment anomalies is 1.5 times the free-flow time. For path Total length, For path The total number of road segments.

4. The method for tracing the source of real-time traffic flow anomalies based on edge computing according to claim 1, characterized in that, The method for obtaining the correlation strength includes: Based on traffic flow data, segment vehicle speeds and corresponding free-flow baseline vehicle speeds are obtained. If a segment's vehicle speed accounts for less than or equal to 60% of the free-flow baseline speed, it is considered an abnormal segment in the traffic flow. For any two abnormal segments, the navigation paths they simultaneously traverse are statistically analyzed to obtain a path set. Based on the path selection probability, spatial overlap analysis is performed on the abnormal segments according to the path set, and the correlation strength of the abnormal segments is calculated. The correlation strength calculation formula is as follows: ； ; in For road section Effective capacity, For road section The design capacity of the passageway For road section The dynamic impedance, This is the dynamic impedance correction factor, calibrated from traffic flow data using the least squares method. For a moment Lower section and road section The strength of the association, For road section Effective capacity, For road section Effective capacity, For a set of paths, For a moment path The probability of path selection. In the path Middle section and road sections The distance between them This is the distance attenuation coefficient. Section Inflow of traffic, Section Outflow rate.

5. The method for tracing the source of real-time traffic flow anomalies based on edge computing according to claim 1, characterized in that, The method for obtaining the contribution includes: The root causes of meteorological warnings in traffic incident announcements are obtained. The timestamp, geographical location, and time difference between the root cause's occurrence and the road segment anomaly are used as root cause attributes. Based on these attributes, the root causes of traffic incident announcements and meteorological warnings are matched according to the correlation strength of each abnormal road segment, maximizing the matching weight to obtain the most relevant set of root causes. The formula for calculating the abnormal root cause matching is as follows: ; in root cause and road section The temporal correlation is calculated based on the overlap ratio between the root cause occurrence time and the road segment anomaly time. As the weight of time correlation, root cause Geographical location and road section Spatial distance, For spatial correlation weight, root cause The response time, i.e., the time difference between the occurrence of the root cause and the road segment anomaly, For response correlation weights; The contribution of each abnormal root cause to the road segment is calculated based on the Shapley value of the root cause set. The contribution calculation formula is as follows: ; in Shapley value representing the contribution of abnormal root causes. For the root cause set, for It does not contain the root cause. a subset of For subset The sum of the correlation strength of the corresponding abnormal road segments For subset The number of root causes included. This represents the total number of root cause sets.

6. The method for tracing the source of real-time traffic flow anomalies based on edge computing according to claim 1, characterized in that, The method for obtaining the optimal matching root cause includes: An abnormal road segment set is obtained based on abnormal road segments. A candidate root cause set is obtained based on the root causes of traffic incident announcements and weather warnings. The abnormal road segment set and the candidate root cause set are used as two vertex sets of a bipartite graph. The contribution of the root cause is used as the edge weight of the abnormal road segment and the candidate root cause in the bipartite graph. The Hungarian algorithm is used to find the matching maximum edge weight in the bipartite graph. Augmenting paths are iteratively searched until the augmenting paths are empty. The sum of the edge weights of the matching edges is maximized, and the optimal matching root cause is generated. The Hungarian algorithm is applicable to bipartite graph matching where the number of abnormal road segments and the number of candidate root causes are equal or unequal.

7. The method for tracing the source of real-time traffic flow anomalies based on edge computing according to claim 1, characterized in that, The method for obtaining the source tracing results includes: The optimal matching root cause is used to extract the navigation path and path selection probability under the influence of the corresponding root cause, generating an association table of root cause, path, and probability. Based on the associated paths in the association table, the upstream road segments are traversed backwards from the abnormal road segment to obtain the dynamic impedance and flow distribution of the upstream road segments, and the traffic flow contribution of each upstream road segment is calculated. The formula for calculating the traffic flow contribution is as follows: ； in upstream section Traffic contribution For passing through the upstream section The set of paths For path Middle section Traffic, For path Total flow; Road segments with a traffic flow contribution of ≥30% are designated as core upstream nodes. The correlation strength between core upstream nodes and abnormal road segments is calculated, and a probability density distribution of the correlation strength is generated. Based on the probability density distribution, the core influencing path is obtained by screening using the upper quartile method. The core influencing path and the correlation table are used as the source tracing results.

8. A real-time traffic flow anomaly event tracing system based on edge computing, used to execute the real-time traffic flow anomaly event tracing method based on edge computing as described in any one of claims 1 to 7, characterized in that, The system includes: Data acquisition module: used to obtain traffic flow data of edge nodes, and preprocess the traffic flow data, which includes origin-destination (OD) data, navigation path, road segment speed, traffic incident announcements and weather warnings; Path selection module: used to analyze the dynamic impedance of the road segment based on the vehicle speed and origin-destination OD data of the road segment, divide the abnormal event waves of each navigation path based on the dynamic impedance and the meteorological warning, and calculate the path selection probability of each path using the proportion of abnormal event waves. Spatial overlap analysis module: used to extract abnormal traffic flow segments based on the vehicle speed of the road segment and the corresponding smooth baseline vehicle speed, and to perform spatial overlap analysis on the abnormal road segments according to the path selection probability to obtain the correlation strength of the abnormal road segments; Anomaly Root Cause Matching Module: Used to match the root causes of traffic incident announcements and weather warnings based on the correlation strength of each abnormal road segment, and obtain the contribution of each abnormal root cause to the road segment. Source tracing module: Based on abnormal road segments and abnormal root causes, the module generates the optimal matching root cause using the maximum weighted binary matching algorithm according to the contribution of the abnormal root causes. It then uses the optimal matching root cause and path selection probability to trace upstream traffic flow and filters the influencing paths based on the correlation strength to obtain the source tracing results.

Images