Road network traffic diversion method based on dynamic topology reconstruction

Abstract

The invention discloses a road network traffic diversion method based on dynamic topology reconstruction. According to the method, the position of the abnormal event and the influenced node set are accurately positioned by identifying the abnormal event and determining the influence range. By dynamically updating node representation, event characteristics are propagated to nodes in an affected range, and dynamic attenuation is performed according to the distance between the nodes and a core node and the severity of the event. On the basis, the connection weight of the affected area is reduced, the connection weight of the bypassing path is enhanced, and the dynamic adjacent matrix is reconstructed. And finally, based on the reconstructed adjacent matrix, combining with a path planning algorithm, comprehensively evaluating the passing efficiency and congestion degree of the path, generating an optimal shunting path, and distributing the flow in proportion. According to the invention, through dynamic network reconstruction and intelligent path planning, traffic congestion caused by abnormal events is effectively relieved, and the emergency response capability and the overall traffic efficiency of the traffic network are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent traffic management, and in particular relates to a road network traffic diversion method based on dynamic topology reconstruction. Background Art

[0002] With the acceleration of urbanization and the continuous growth of the number of motor vehicles, the operating pressure of urban transportation systems has increased significantly, especially when unexpected events (such as traffic accidents, road construction and natural disasters) occur. These events often have a serious impact on the capacity and stability of the transportation network. According to research in the literature [1], abnormal events are the main reason for the reduction of transportation network efficiency. Its impact mechanism is mainly manifested in the reduction of local capacity, traffic redistribution and chain congestion effect.

[0003] Under the influence of abnormal events, the traffic network faces two major problems. First, the traffic capacity of local nodes decreases. The traffic capacity of roads is closely related to the speed and density of traffic, and sudden events can cause the traffic speed of local sections to drop significantly or even be completely interrupted [2]. This phenomenon greatly weakens the connectivity of the traffic network, causing the original path planning to fail, and further causing the problem of reduced network operation efficiency. Second, traffic is redistributed and congestion spreads. After an abnormal event occurs, vehicles usually choose alternative routes by spontaneous detours or under the guidance of traffic management departments. However, this traffic redistribution is often unbalanced, which can easily put additional pressure on surrounding sections, form new congestion points, and further aggravate the instability of the traffic system [3]. This type of diffusion effect has nonlinear characteristics and can easily cause local congestion to spread rapidly to the entire traffic network.

[0004] In response to these problems, existing traffic diversion methods have obvious technical limitations, which are manifested in two aspects. First, they lack adaptability to dynamic changes in the network. Most existing methods mainly rely on fixed-weight path planning models, such as the classic Dijkstra algorithm [4] and A* algorithm [5]. These methods assume that the topological structure of the traffic network remains unchanged during the incident, which makes it difficult to cope with complex scenarios where the traffic capacity of the road section changes dynamically, resulting in a lack of flexibility and accuracy in the diversion scheme. Second, they cannot fully capture the accident diffusion effect. Although some methods introduce real-time updates, they usually only make simple adjustments to local sections and fail to systematically re-evaluate the topological structure of the entire traffic network. For example, the shortest path optimization algorithm proposed in the literature [6] only considers the core area of ​​the accident and ignores the diffusion effect of the peripheral sections, resulting in the inability to effectively alleviate the secondary congestion problem.

[0005] In view of the significant deficiencies of the prior art, there is an urgent need for a new traffic diversion method that can flexibly adapt to the dynamic changes of the network and fully capture the accident diffusion effect. In response to the above needs, the present invention proposes a road network traffic diversion method based on dynamic topology reconstruction. The method first scientifically reflects the changes in the connection strength of the affected nodes and sections through dynamic adjacency matrix updates, and at the same time enhances the connection weights between potential detour paths and non-affected nodes, thereby realizing adaptive adjustment of the network topology. Compared with the traditional fixed topology structure method, the present invention can more accurately capture the dynamic changes of the traffic network and significantly improve the flexibility and reliability of the diversion strategy. In addition, the accident-affected area is fully quantified through data-driven, providing data support for topology reconstruction, which not only more accurately reflects the actual impact of the accident, but also lays a scientific foundation for dynamic diversion path planning. In addition, on the updated dynamic topology structure, the optimal path algorithm and the network capacity constraint are combined to generate a scientific and reasonable diversion scheme. Compared with the existing shortest path algorithm, the present invention fully considers the dynamic characteristics of traffic flow and network capacity constraints, can effectively avoid the generation of new congestion points, and maximize the overall operation efficiency of the traffic system. In summary, the present invention innovatively combines dynamic topology reconstruction with data-driven precise evaluation to overcome the shortcomings of existing methods in adaptability and precision. By dynamically updating the adjacency matrix and efficient path planning, the present invention can provide flexible and precise diversion strategies for traffic management departments, effectively deal with congestion problems caused by abnormal events, and improve the adaptive ability of road networks and emergency response efficiency.

[0006] [1]Smith J, et al.Impact of Incidents on Urban TrafficFlow.Transportation Research Part A, 2020.

[0007] [2] Zhao X, et al. Modeling Road Capacity under DisruptiveEvents. Journal of Traffic Engineering, 2021.

[0008] [3]Liu Y, et al.Traffic Flow Redistribution Mechanisms.IEEE Transactions on Intelligent Transportation Systems, 2019.

[0009] [4]Dijkstra EA Note on Two Problems in Connexion withGraphs.Numerische Mathematik, 1959.

[0010] [5]Hart PE, et al. The A*Algorithm for Shortest PathSearch. Communications of the ACM, 1968.

[0011] [6]Wang Y, et al. Optimization of Traffic Routing under Disruption. Journal of Applied Transportation Research, 2022. Summary of the invention

[0012] The purpose of the present invention is to provide a road network traffic diversion method based on dynamic topology reconstruction, which aims to deal with the problems of reduced traffic capacity and traffic redistribution of road network nodes caused by abnormal events. It effectively solves the problems of static diversion strategy, inaccurate impact range assessment and low diversion efficiency in the prior art, provides a flexible and accurate diversion solution for traffic management departments, and greatly improves the adaptive ability and emergency response efficiency of the road network.

[0013] The specific plan is as follows:

[0014] A road network traffic diversion method based on dynamic topology reconstruction first identifies the location and impact range of abnormal events, dynamically updates the feature representation of network nodes, and adjusts the node features based on the geographical distance, topological relationship and event severity between nodes; then, on this basis, dynamically updates the network adjacency matrix, reduces the connection weights between nodes in the affected area, and enhances the connection weights of potential detour paths, reflecting the capacity changes and traffic redistribution requirements caused by the event; finally, a set of candidate diversion paths is generated through a path planning algorithm, and the path capacity utilization, travel time and path cost are comprehensively evaluated to select the optimal diversion path, and the traffic is distributed in proportion. The path weight is dynamically adjusted to form an efficient diversion strategy, so as to achieve scientific redistribution of traffic flow and improvement of network traffic efficiency under the influence of abnormal events. Specifically, the method includes the following steps:

[0015] Step S1: Identification of abnormal events and determination of impact scope

[0016] When an abnormal event occurs in a traffic network, timely and accurate identification of the location and scope of impact of the event is the basis for formulating a diversion strategy. The main goal of this step is to dynamically determine the specific location of the event through analysis of traffic data, identify the affected road sections and their node sets, and quantify the impact of the event on the traffic network. This process provides data support for subsequent topological structure reconstruction and diversion path optimization. Through accurate assessment of the spatial scope of abnormal events and changes in traffic capacity, the adjacency matrix elements in the traffic network can be scientifically adjusted to reflect the disturbance and impact of the event on the overall road network structure. It specifically includes the following sub-steps:

[0017] Sub-step S11: Traffic data input and presentation

[0018] The input data of this method mainly includes three parts: traffic network weighted graph, historical traffic data and abnormal event data.

[0019] The weighted graph structure of the traffic network is defined as G = (V, E, A). V is the node set, representing the key nodes in the road network, E is the edge set, representing the road sections between nodes, and A is the adjacency matrix, describing the static connection relationship between nodes:

[0020]

[0021] Where W ij For edge e ij The weight of represents the traffic capacity.

[0022] Historical traffic data Including each edge e ij Historical traffic (Unit: number of vehicles / time unit), historical speed (Unit: km / h), road section occupancy rate (Reflects vehicle density).

[0023] The abnormal event data is defined as a four-tuple E = {L, S, T, D}, where L represents the event location, S represents the severity of the event, which is reflected as the percentage of the decrease in traffic capacity, T is the time of the event, and D is the duration of the event.

[0024] Sub-step S12: Determine the location of the abnormal event

[0025] Abnormal events are divided into explicit input events and potential abnormal events affected by input events (need to be implicitly identified).

[0026] For the event of display input, directly send the event e k Location k Mapped to network node v i ∈V:

[0027]

[0028] Where d(l k ,v i ) indicates the event location l k To node v i The standardized geographic distance, cos(θ(l k ,v i )) is the cosine value of the angle between the event propagation direction and the road section direction, which is used to measure the direction consistency. d and w θ are the learnable weight coefficients for geographical distance and direction factors respectively.

[0029] For events that require implicit identification, the potential abnormal event locations are dynamically identified by analyzing the differences between historical traffic data and input traffic, including the traffic change rate ΔQ ij and traffic state degradation rate R i Calculation of (t):

[0030]

[0031]

[0032] Where Q ij (t), V ij (t) is the current flow rate and speed, is the historical flow and speed. If |ΔQ ij | exceeds the threshold and R i (t) is less than the threshold value, and the state duration exceeds a certain value, the road section is determined to be an abnormal road section, and the nodes included in the road section are recorded into the affected node set v affected middle.

[0033] Sub-step S13: Identification of the impact range of abnormal events

[0034] The attention mechanism model is used to quantify the weights of the affected nodes and determine the impact range of the event. For each pair of nodes (v i ,v j ), attention weight α ij Represents node v j For node v i Importance:

[0035]

[0036] Where W a is the trainable weight matrix, F i , F j is the feature representation of the node, and || is the vector concatenation operation.

[0037] According to the attention weight, the affected node set is further filtered:

[0038] v affected = {v i |α ic >T,d(v i ,v c )≤φ(s k )}

[0039] Among them, φ(s k ) is based on the severity of the event k The mapping function, α ic is the calculated attention weight between events and influencing nodes, and T is the weight screening threshold.

[0040] Step S2: Node dynamic representation update

[0041] After an abnormal event occurs, in order to accurately reflect the direct and indirect impact of the event on the traffic network nodes, it is necessary to dynamically update the feature representation of the node. By propagating the event features from the core node to the surrounding nodes, the spatial impact range and time attenuation effect of the abnormal event can be described.

[0042] Based on step S12, the core node v is obtained c , injecting event features directly into its representation to enhance the node’s sensitivity to abnormal events. The node’s representation is updated as follows:

[0043]

[0044] in, For node v c The original feature representation, F e is the event feature, W e Inject weights into learnable event features.

[0045] For nodes affected by core nodes, the node representation is attenuated based on their distance to the core node and the severity of the event, and the node representation is updated as follows:

[0046]

[0047] Among them, η i is the attenuation coefficient, defined as follows:

[0048]

[0049] Among them, d(v i ,v c ) represents node v i To the core node v c Geographic distance, hop(v i,v c ) indicates that in the original road network topology, i To the core node v c The number of hops of the shortest path (i.e., topological distance). λ1 and λ2 are learnable decay rates that control the two distances. k and d k Indicates the time when an event occurs and its duration, reflecting the process of the event's impact decaying over time.

[0050] Step S3: Update of dynamic adjacency matrix

[0051] Based on the updated node representation, the network connection relationship needs to be dynamically adjusted to reflect the reduction in traffic capacity and flow redistribution caused by the event. The specific steps include the following:

[0052] Sub-step S31: Update the weight of the affected road section

[0053] For the node set v affected by the event affected , the connection strength between them needs to be reduced to reflect the weakening of traffic capacity caused by abnormal events. The updated dynamic adjacency matrix Defined as:

[0054]

[0055] in, is the original adjacency matrix, is the deviation rate between the node dynamic representation and the original feature, and α is the connection strength adjustment parameter, which controls the extent of weight weakening.

[0056] Sub-step S32: Potential detour path diversion update

[0057] For nodes other than the affected nodes, it is necessary to strengthen the connection weights between them and non-affected nodes to encourage traffic to be redistributed to detour paths. Updated dynamic adjacency matrix Defined as:

[0058]

[0059] in, is the original adjacency matrix, β is the detour path weight adjustment parameter, d max is the maximum allowable detour distance, which is obtained based on the statistical characteristics of the actual traffic pattern of the network, d(v i ,v k ) is the geographical distance between nodes.

[0060] Sub-step S33: Overall road network connectivity adjustment

[0061] Based on the affected area adjacency matrix obtained in step S31 above The potential detour path adjacency matrix obtained by S32 The weighted fusion is used to obtain the final dynamic adjacency matrix:

[0062]

[0063] Where γ is a learnable weight fusion coefficient. When γ→1, priority is given to weakening the connections in the affected area, emphasizing the direct impact of the event on the network. When γ→0, priority is given to enhancing the weight of the detour path, promoting the diversion of traffic to the unaffected area.

[0064] Step S4: Generation and optimization of diversion paths

[0065] After the dynamic adjacency matrix is ​​updated, generating diversion paths is a key step to solve traffic congestion caused by abnormal events. This step generates a set of candidate diversion paths based on the path planning algorithm of the dynamic adjacency matrix, and comprehensively considers indicators such as network capacity, detour time and traffic efficiency to select the optimal path and reasonably allocate traffic.

[0066] The dynamic adjacency matrix obtained using S3 As input, combined with the shortest path algorithm (such as Dijkstra algorithm), a candidate diversion path set P = {p1, p2, ..., p k}, where p i ={v1,v2,...,v m}, k is the number of candidate paths. At the same time, to ensure the actual feasibility of the path, the length and detour cost of the candidate path are limited:

[0067] L(p i )≤ψ1·L direct , C(p i )≤ψ2·C direct,

[0068] Among them, L(p i ) and L direct For path p i The length of the direct path, C(p i ) and C direct For path p i The travel time and the direct path travel time, ψ1 and ψ2 are the detour cost coefficients.

[0069] For each candidate path p i ∈P, calculate capacity utilization Where Q ij , C ij are the flow and capacity of each edge in the path respectively. Where T ijis the travel time of each edge in the path. Considering the trade-off between capacity utilization and travel time, the path cost C(p i ), and then sort the candidate paths according to the path cost, and select the path with the minimum cost as the optimal path p opt :

[0070] C(p i )=λ·U(p i )+(1-λ)·T(p i )

[0071]

[0072] Among them, λ∈[0,1] is the weight balance coefficient between capacity and time.

[0073] In order to avoid overloading a single path, the optimal path p opt and other candidate paths p i ∈P distributes traffic proportionally:

[0074]

[0075] Where Q total is the total flow that needs to be diverted.

[0076] The diversion path and traffic distribution ratio are used as the final diversion strategy output, and the optimal path p is finally output opt , path flow allocation ratio Q pi , providing diversion nodes and guidance information for the implementation of the traffic guidance system. In practice, it can be combined with real-time monitoring of the execution effect of the path, recording data such as flow changes, travel time and delays. According to the feedback results, the path weight is dynamically adjusted to optimize the next round of strategies.

[0077] The present invention achieves the following beneficial effects through the above technical solution:

[0078] The purpose of the present invention is to provide a road network traffic diversion method based on dynamic topology reconstruction, which is intended to deal with the problems of decreased capacity and traffic redistribution of road network nodes caused by abnormal events (such as traffic accidents). First, through the dynamic node representation update and the reconstruction of the dynamic adjacency matrix, the present invention can reflect the direct and indirect impact of abnormal events on the traffic network in real time. By injecting event features into core nodes and performing distance and time attenuation processing on indirectly affected nodes, the disturbance impact of events in space and time dimensions is accurately quantified, so that the network structure has stronger dynamic adaptability and expression ability. Secondly, by updating the dynamic adjacency matrix, the connection strength between the affected nodes is reduced, reflecting the weakening of traffic capacity caused by the event, while enhancing the weight of the detour path, and promoting the redistribution of traffic to unaffected areas. This process effectively balances the local congestion and global traffic distribution of the network, alleviates the traffic bottleneck problem caused by abnormal events, and improves the traffic efficiency of the traffic network. Finally, by generating and optimizing the diversion path, combining the dynamic adjacency matrix and the path planning algorithm, a set of candidate diversion paths is generated, and multi-dimensional factors such as network capacity, detour time and travel cost are comprehensively considered to select the optimal path and distribute the traffic in proportion. The present invention provides an efficient and scientific diversion strategy that can dynamically adjust the path weight and feedback optimization to ensure the rationality and real-time performance of traffic distribution. In summary, the present invention can achieve efficient redistribution of traffic flow through accurate impact range identification, dynamic network reconstruction and intelligent diversion strategies when abnormal events occur, significantly improving the emergency response capability, traffic efficiency and stability of the transportation network, and providing technical support with practical application value for intelligent traffic management. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a flow chart of a road network traffic diversion method based on dynamic topology reconstruction. DETAILED DESCRIPTION

[0080] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0081] like Figure 1As shown, the present invention proposes a road network traffic diversion method based on dynamic topology reconstruction. This method dynamically adjusts the connection strength of affected nodes through dynamic adjacency matrix updates, while enhancing the connection weights between detour paths and non-affected nodes, thereby reflecting the dynamic changes in network topology; through data-driven impact range assessment, the accident impact area and traffic capacity changes are accurately quantified to provide a scientific basis for road network topology reconstruction; through efficient traffic redistribution, combined with the optimal path algorithm and network capacity constraints, the optimal diversion path is generated on the dynamic topology structure to achieve scientific distribution of traffic. This method effectively solves the problems of static diversion strategies, inaccurate impact range assessment, and low diversion efficiency in the prior art, and provides traffic management departments with flexible and accurate diversion solutions, greatly improving the adaptive ability of road networks and emergency response efficiency. The process of the road network traffic diversion method based on dynamic topology reconstruction is as follows: Figure 1 As shown, the specific steps include:

[0082] Step S1: Identification of abnormal events and determination of impact scope

[0083] When an abnormal event occurs in a traffic network, timely and accurate identification of the location and scope of impact of the event is the basis for formulating a diversion strategy. The main goal of this step is to dynamically determine the specific location of the event through analysis of traffic data, identify the affected road sections and their node sets, and quantify the impact of the event on the traffic network. This process provides data support for subsequent topological structure reconstruction and diversion path optimization. Through accurate assessment of the spatial scope of abnormal events and changes in traffic capacity, the adjacency matrix elements in the traffic network can be scientifically adjusted to reflect the disturbance and impact of the event on the overall road network structure. It specifically includes the following sub-steps:

[0084] Sub-step S11: Traffic data input and presentation

[0085] The input data of this method mainly includes three parts: traffic network weighted graph, historical traffic data and abnormal event data.

[0086] The weighted graph structure of the traffic network is defined as G = (V, E, A). V is the node set, representing the key nodes in the road network, E is the edge set, representing the road sections between nodes, and A is the adjacency matrix, describing the static connection relationship between nodes:

[0087]

[0088] Where W ij For edge e ij The weight of represents the traffic capacity.

[0089] Historical traffic data Including each edge e ij Historical traffic (Unit: number of vehicles / time unit), historical speed (Unit: km / h), road section occupancy rate (Reflects vehicle density).

[0090] The abnormal event data is defined as a four-tuple E = {L, S, T, D}, where L represents the event location, S represents the severity of the event, which is reflected as the percentage of the decrease in traffic capacity, T is the time of the event, and D is the duration of the event.

[0091] Sub-step S12: Determine the location of the abnormal event

[0092] Abnormal events are divided into explicit input events and potential abnormal events affected by input events (need to be implicitly identified).

[0093] For the event of display input, directly send the event e k Location k Mapped to network node v i ∈V:

[0094]

[0095] Where d(l k ,v i ) indicates the event location l k To node v i The standardized geographic distance, cos(θ(l k ,v i )) is the cosine value of the angle between the event propagation direction and the road section direction, which is used to measure the direction consistency. d and w θ are the learnable weight coefficients for geographical distance and direction factors respectively.

[0096] For events that require implicit identification, the potential abnormal event locations are dynamically identified by analyzing the differences between historical traffic data and input traffic, including the traffic change rate ΔQ ij and traffic state degradation rate R i Calculation of (t):

[0097]

[0098]

[0099] Where Q ij (t), V ij (t) is the current flow rate and speed, is the historical flow and speed. If |ΔQ ij | exceeds the threshold and R i(t) is less than the threshold value, and the state duration exceeds a certain value, the road section is determined to be an abnormal road section, and the nodes included in the road section are recorded into the affected node set v affected middle.

[0100] Sub-step S13: Identification of the impact range of abnormal events

[0101] The attention mechanism model is used to quantify the weights of the affected nodes and determine the impact range of the event. For each pair of nodes (v i ,v j ), attention weight α ij Represents node v j For node v i Importance:

[0102]

[0103] Where W a is the trainable weight matrix, F i , F j is the feature representation of the node, and || is the vector concatenation operation.

[0104] According to the attention weight, the affected node set is further filtered:

[0105] v affected = {v i |α ic >T,d(v i ,v c )≤φ(s k )}

[0106] Among them, φ(s k ) is based on the severity of the event k The mapping function, α ic is the calculated attention weight between events and influencing nodes, and T is the weight screening threshold.

[0107] Step S2: Node dynamic representation update

[0108] After an abnormal event occurs, in order to accurately reflect the direct and indirect impact of the event on the traffic network nodes, it is necessary to dynamically update the feature representation of the node. By propagating the event features from the core node to the surrounding nodes, the spatial impact range and time attenuation effect of the abnormal event can be described.

[0109] Based on step S12, the core node v is obtained c , injecting event features directly into its representation to enhance the node’s sensitivity to abnormal events. The node’s representation is updated as follows:

[0110]

[0111] in, For node v c The original feature representation, F e is the event feature, W e Inject weights into learnable event features.

[0112] For nodes affected by core nodes, the node representation is attenuated based on their distance to the core node and the severity of the event, and the node representation is updated as follows:

[0113]

[0114] Among them, η i is the attenuation coefficient, defined as follows:

[0115]

[0116] Among them, d(v i ,v c ) represents node v i To the core node v c Geographic distance, hop(v i ,v c ) indicates that in the original road network topology, i To the core node v c The number of hops of the shortest path (i.e., topological distance). λ1 and λ2 are learnable decay rates that control the two distances. k and d k Indicates the time when an event occurs and its duration, reflecting the process of the event's impact decaying over time.

[0117] Step S3: Update of dynamic adjacency matrix

[0118] Based on the updated node representation, the network connection relationship needs to be dynamically adjusted to reflect the reduction in traffic capacity and flow redistribution caused by the event. The specific steps include the following:

[0119] Sub-step S31: Update the weight of the affected road section

[0120] For the node set v affected by the event affected , the connection strength between them needs to be reduced to reflect the weakening of traffic capacity caused by abnormal events. The updated dynamic adjacency matrix Defined as:

[0121]

[0122] in, is the original adjacency matrix, is the deviation rate between the node dynamic representation and the original feature, and α is the connection strength adjustment parameter, which controls the extent of weight weakening.

[0123] Sub-step S32: Potential detour path diversion update

[0124] For nodes other than the affected nodes, it is necessary to strengthen the connection weights between them and non-affected nodes to encourage traffic to be redistributed to detour paths. Updated dynamic adjacency matrix Defined as:

[0125]

[0126] in, is the original adjacency matrix, β is the detour path weight adjustment parameter, d max is the maximum allowable detour distance, which is obtained based on the statistical characteristics of the actual traffic pattern of the network, d(v i ,v k ) is the geographical distance between nodes.

[0127] Sub-step S33: Overall road network connectivity adjustment

[0128] Based on the affected area adjacency matrix obtained in step S31 above The potential detour path adjacency matrix obtained by S32 The weighted fusion is used to obtain the final dynamic adjacency matrix:

[0129]

[0130] Where γ is a learnable weight fusion coefficient. When γ→1, priority is given to weakening the connections in the affected area, emphasizing the direct impact of the event on the network. When γ→0, priority is given to enhancing the weight of the detour path, promoting the diversion of traffic to the unaffected area.

[0131] Step S4: Generation and optimization of diversion paths

[0132] After the dynamic adjacency matrix is ​​updated, generating diversion paths is a key step to solve traffic congestion caused by abnormal events. This step generates a set of candidate diversion paths based on the path planning algorithm of the dynamic adjacency matrix, and comprehensively considers indicators such as network capacity, detour time and traffic efficiency to select the optimal path and reasonably allocate traffic.

[0133] The dynamic adjacency matrix obtained using S3 As input, combined with the shortest path algorithm (such as Dijkstra algorithm), a candidate diversion path set P = {p1, p2, ..., p k}, where p i ={v1,v2,...,v m}, k is the number of candidate paths. At the same time, to ensure the actual feasibility of the path, the length and detour cost of the candidate path are limited:

[0134] L(p i )≤ψ1·L direct , C(p i )≤ψ2·C direct,

[0135] Among them, L(p i ) and L direct For path p i The length of the direct path, C(p i ) and C direct For path p i The travel time and the direct path travel time, ψ1, ψ2 are the detour cost coefficients.

[0136] For each candidate path p i ∈P, calculate capacity utilization Where Q ij , C ij are the flow and capacity of each edge in the path respectively. Where T ij is the travel time of each edge in the path. Considering the trade-off between capacity utilization and travel time, the path cost C(p i ), and then sort the candidate paths according to the path cost, and select the path with the minimum cost as the optimal path p opt :

[0137] C(p i )=λ·U(p i )+(1-λ)·T(p i )

[0138]

[0139] Among them, λ∈[0,1] is the weight balance coefficient between capacity and time.

[0140] In order to avoid overloading a single path, the optimal path p opt and other candidate paths p i ∈P distributes traffic proportionally:

[0141]

[0142] Where Q total is the total flow that needs to be diverted.

[0143] The diversion path and traffic distribution ratio are used as the final diversion strategy output, and the optimal path p is finally output opt , path flow allocation ratio Qpi , providing diversion nodes and guidance information for the implementation of the traffic guidance system. In practice, it can be combined with real-time monitoring of the execution effect of the path, recording data such as flow changes, travel time and delays. According to the feedback results, the path weight is dynamically adjusted to optimize the next round of strategies.

Claims

1. A road network traffic diversion method based on dynamic topology reconstruction, characterized in that: First, by identifying the location and impact range of abnormal events, the feature representation of network nodes is dynamically updated, and node features are adjusted based on the geographical distance, topological relationship and severity of the event between nodes. Then, on this basis, the network adjacency matrix is ​​dynamically updated to reduce the connection weights between nodes in the affected area and enhance the connection weights of potential detour paths to reflect the changes in traffic capacity and traffic redistribution requirements caused by the event. Finally, a set of candidate diversion paths is generated through the path planning algorithm, and the path capacity utilization, travel time and path cost are comprehensively evaluated to select the optimal diversion path, distribute the traffic in proportion, and dynamically adjust the path weight to form an efficient diversion strategy, so as to achieve scientific redistribution of traffic flow under the influence of abnormal events and improve network travel efficiency. The method comprises the following steps: Step S1: abnormal event identification and impact scope determination; Step S2: node dynamic representation update; Step S3: Dynamic adjacency matrix update; Step S4: Generation and optimization of diversion paths.

2. A road network traffic diversion method based on dynamic topology reconstruction according to claim 1, characterized in that: The abnormal event identification and impact range determination described in step S1 are as follows: Step S1: Identification of abnormal events and determination of impact scope When an abnormal event occurs in the traffic network, timely and accurate identification of the location and impact range of the event is the basis for formulating a diversion strategy. By analyzing traffic data, the specific location of the event is dynamically determined, the affected road sections and their node sets are identified, and the impact of the event on the traffic network is quantified. This provides data support for subsequent topology reconstruction and diversion path optimization. The specific steps include the following: Sub-step S11: Traffic data input and presentation The input data includes three parts: traffic network weighted graph, historical traffic data and abnormal event data; The weighted graph structure of the traffic network is defined as G = (V, E, A); where V is the node set, representing the key nodes in the road network, E is the edge set, representing the road sections between nodes, and A is the adjacency matrix, describing the static connection relationship between nodes: Among them, W ij For edge e ij The weight of represents the traffic capacity; Historical traffic data Including each edge e ij Historical traffic Historical speed Road section occupancy rate Abnormal event data is defined as a four-tuple E = {L, S, T, D}, where L represents the event location, S represents the severity of the event, which is reflected as the percentage of the decrease in traffic capacity, T is the time of the event, and D is the duration of the event; Sub-step S12: Determine the location of the abnormal event Abnormal events are divided into explicit input events and potential abnormal events affected by input events, among which potential abnormal events need to be implicitly identified; For the event of display input, directly send the event e k Location k Mapped to network node v i ∈V: Among them, v c is the core node; d(l k ,v i ) indicates the event location l k To node v i The standardized geographic distance, cos(θ(l k ,v i )) is the cosine value of the angle between the event propagation direction and the road section direction, which is used to measure the direction consistency; w d and w θ are the learnable weight coefficients for geographical distance and direction factors, respectively; For potential abnormal events that need to be implicitly identified, the potential abnormal event locations are dynamically identified by analyzing the differences between historical traffic data and input traffic, including the traffic change rate ΔQ ij and traffic state degradation rate R i Calculation of (t): Among them, Q ij (t), V ij (t) is the current flow rate and speed, is the historical flow and speed; if |ΔQ ij | exceeds the threshold and R i (t) is less than the threshold value, and the state duration exceeds a certain value, the road section is determined to be an abnormal road section, and the nodes included in the road section are recorded into the affected node set v affected middle; Sub-step S13: Identification of the impact range of abnormal events The attention mechanism model is used to quantify the weights of the affected nodes and determine the impact range of the event; for each pair of nodes (v i ,v j ), attention weight α ij Represents node v j For node v i Importance: Among them, W a is the trainable weight matrix, F i , F j is the feature representation of the node, || is the vector concatenation operation; According to the attention weight, the affected node set v is further filtered affected : v affected ={v i |α ic >T,d(v i ,v c )≤φ(s k )} Among them, φ(s k ) is based on the severity of the event k The mapping function, α ic is the calculated attention weight between events and influencing nodes, and T is the weight screening threshold.

3. A road network traffic diversion method based on dynamic topology reconstruction according to claim 2, characterized in that: The node dynamic representation update described in step S2 is as follows: Step S2: Node dynamic representation update After an abnormal event occurs, in order to accurately reflect the direct and indirect impact of the event on the traffic network nodes, it is necessary to dynamically update the feature representation of the node; by propagating the event features from the core node to the surrounding nodes, the spatial impact range and time attenuation effect of the abnormal event can be described; Based on step S12, the core node v is obtained c , injecting event features directly into its representation to enhance the node’s sensitivity to abnormal events. The node’s representation is updated as follows: in, For node v c The original feature representation, F e is the event feature, W e Inject weights into learnable event features; For nodes affected by core nodes, the node representation is attenuated based on their distance to the core node and the severity of the event, and the node representation is updated as follows: Among them, η i is the attenuation coefficient, defined as follows: Among them, d(v i ,v c ) represents node v i To the core node v c Geographic distance, hop(v i ,v c ) indicates that in the original road network topology, i To the core node v c The number of hops of the shortest path, i.e., the topological distance; λ1 and λ2 are the learnable decay rates that control the two distances; t k and d k Indicates the time when an event occurs and its duration, reflecting the process of the event's impact decaying over time.

4. A road network traffic diversion method based on dynamic topology reconstruction according to claim 3, characterized in that: The dynamic adjacency matrix update described in step S3 is as follows: Step S3: Update of dynamic adjacency matrix Based on the updated node representation, the network connection relationship needs to be dynamically adjusted to reflect the reduction in traffic capacity and flow redistribution caused by the event; specifically, it includes the following sub-steps: Sub-step S31: Update the weight of the affected road section For the node set v affected by the event affected , it is necessary to reduce the connection strength between them to reflect the weakening of traffic capacity caused by abnormal events; the updated dynamic adjacency matrix Defined as: in, is the original adjacency matrix, is the deviation rate between the node dynamic representation and the original feature, α is the connection strength adjustment parameter, which controls the extent of weight weakening; Sub-step S32: Potential detour path diversion update For nodes other than the affected nodes, it is necessary to enhance the connection weights between them and the non-affected nodes to encourage traffic to be redistributed to the detour path; the updated dynamic adjacency matrix Defined as: in, is the original adjacency matrix, β is the detour path weight adjustment parameter, d max is the maximum allowable detour distance, which is obtained based on the statistical characteristics of the actual traffic pattern of the network, d(v i ,v k ) is the geographical distance between nodes; Sub-step S33: Overall road network connectivity adjustment Based on the affected area adjacency matrix obtained in step S31 above The potential detour path adjacency matrix obtained by S32 The weighted fusion is used to obtain the final dynamic adjacency matrix: Where γ is a learnable weight fusion coefficient. When γ→1, priority is given to weakening the connections in the affected area, emphasizing the direct impact of the event on the network. When γ→0, priority is given to enhancing the weight of the detour path, promoting the diversion of traffic to the unaffected area.

5. A road network traffic diversion method based on dynamic topology reconstruction according to claim 4, characterized in that: The generation and optimization of the diversion path described in step S4 is specifically as follows: After the dynamic adjacency matrix is ​​updated, the path planning algorithm based on the dynamic adjacency matrix generates a set of candidate diversion paths, comprehensively considers network capacity, detour time and traffic efficiency indicators, selects the optimal path, and reasonably distributes traffic; The dynamic adjacency matrix obtained using S3 As input, combined with the shortest path algorithm, a candidate diversion path set P = {p1, p2, ..., p k }, where p i ={v1,v2,...,v m }, k is the number of candidate paths; at the same time, in order to ensure the actual feasibility of the path, the length and detour cost of the candidate path are limited: L(p i )≤ψ1·L direct ,C(p i )≤ψ2·C direct, Among them, L(p i ) and L direct For path p i The length of the direct path, C(p i ) and C direct For path p i The travel time is the travel time of the direct path, ψ1, ψ2 are the detour cost coefficients; For each candidate path p i ∈P, calculate capacity utilization Where Q ij , C ij are the flow and capacity of each edge in the path; the travel time Where T ij is the travel time of each edge in the path; taking into account the trade-off between capacity utilization and travel time, the path cost C(p i ), and then sort the candidate paths according to the path cost, and select the path with the minimum cost as the optimal path p opt : C(p i )=λ·U(p i )+(1-λ)·T(p i ) Among them, λ∈[0,1] is the weight balance coefficient between capacity and time; In order to avoid overloading a single path, the optimal path p opt and other candidate paths p i ∈P distributes traffic proportionally: Where Q total is the total flow that needs to be diverted; The diversion path and traffic distribution ratio are used as the final diversion strategy output, and the optimal path p is finally output opt , path flow allocation ratio Q pi , provide diversion nodes and guidance information for the implementation of traffic induction system; it can be combined with real-time monitoring of path execution effect, record traffic changes, travel time and delay data, dynamically adjust path weights according to feedback results, and optimize the next round of strategies.

Images