Parallel traffic simulation acceleration method for road network load perception

Abstract

The invention relates to the technical field of traffic simulation, and discloses a road network load aware parallel traffic simulation acceleration method, which specifically comprises the following steps of: 1, preprocessing a road network; 2, based on an optimization target, defining an optimization problem of road network division; step 3, carrying out initial division on the road network through graph growth; 4, refining after initial partitioning to further reduce the edge cutting weight; the step 1 of preprocessing the road network comprises the following detailed steps: S100, converting the road network into a weighted graph; s101, selecting an initial weight; s102, obtaining a traffic perception weight; and S103, defining the traffic perception weight of the intersection. According to the method, a traditional method which only depends on the road length is abandoned, weight setting and actual traffic conditions are prone to being disjointed in a traditional mode, and the real status of the road in a traffic system can be accurately captured through the method.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic simulation, and particularly to a parallel traffic simulation acceleration method for road network load perception. Background Art

[0002] In the research and practice of the current traffic field, traffic simulation technology occupies a crucial position. As a powerful tool, it can model and analyze complex traffic systems in a virtual environment, providing important decision-making support for traffic planning, management, and the research and development of intelligent transportation systems. With the acceleration of urbanization and the continuous growth of traffic flow, the requirements for the accuracy and efficiency of traffic simulation are also increasing day by day, which promotes the continuous evolution and development of related technologies.

[0003] In terms of road network division in traditional traffic simulation, most use the basic recursive bisection method for static road network division. When determining the initial weight of a road, usually only the road length is used as the basis. In the actual operation process of this method, due to the lack of full consideration of key factors such as the actual traffic flow, traffic density, and number of lanes of the road, the computational load distribution is extremely uneven during parallel traffic simulation calculations. For example, in some traffic-intensive areas, although the road length is short, the traffic volume is large and the traffic density is high, and insufficient computational resources may be allocated under the traditional division method; while in some long sections with sparse traffic volume, excessive computational resources may be occupied. This not only seriously reduces the computational efficiency during operation and prolongs the simulation time required, but also may cause problems such as data processing delays and error accumulation due to load imbalance, greatly affecting the reliability and practicality of traffic simulation results.

[0004] The load perception-based parallel traffic simulation acceleration method proposed by the present invention can effectively ensure load balance during the reinforcement learning training process through reasonable road network division, thereby significantly improving the efficiency of traffic simulation for reinforcement learning. Summary of the Invention

[0005] The present invention mainly solves the above technical problems and provides a parallel traffic simulation acceleration method for road network load perception.

[0006] To achieve the above object, the present invention adopts the following technical solution. The parallel traffic simulation acceleration method for road network load perception includes the following steps:

[0007] Step 1: Preprocess the road network;

[0008] Step 2: Define the optimization problem of road network division based on the optimization objective;

[0009] Step 3: Perform an initial division of the road network through graph growth;

[0010] Step 4: Refinement after initial partitioning further reduces the edge cut weight;

[0011] Among them, step 1 for preprocessing the road network includes the following detailed steps:

[0012] S100. Convert the road network into a weighted graph;

[0013] S101. Select the initial weight;

[0014] S102. Traffic-aware weight;

[0015] S103. Define the traffic-aware weight of intersections.

[0016] Preferably, in step 1, the road network is converted into a weighted graph G=(E, V), where E is the set of edges and V is the set of vertices. The nodes in the road network are mapped to the vertices of the graph, and the links between nodes are mapped to the edges between vertices. The traffic-aware weight W e = C e represents the road weight, and C e is the access count of road e. Define the traffic-aware weight of intersections:

[0017]

[0018] where E v represents the set of roads connected to vertex v, L e is the length of road e, w' v is the weighted perception weight of vertex v; all vertices are assigned as the base weight, and w v is the traffic-aware weight of the final vertex v.

[0019] Preferably, in step 2, the graph G is cut into m non-overlapping partitions G={G 0 , G 1 ,..., G m-1}, and the optimization objectives are respectively: minimize the workload imbalance minimize the edge cut minimize the number of boundary edges

[0020] Preferably, in step 3, the weighted graph generated from the road network is first initially partitioned by graph growth. The graph growth is based on sorting the vertices in the priority queue. The vertices are sorted according to the partition ID and coordinates marked in their tuples. Starting from the initial vertex, the subgraph grows along the edges of the graph. When the cumulative weight of the vertices in the partition exceeds the average weight, a new vertex partition is started, and all adjacent vertices that have not been enqueued before the current vertex are pushed into the queue and marked as enqueued.

[0021] Preferably, in step 4, after the initial partitioning, a benefit function for moving a certain vertex v between partitions is defined where W e(v,u) is the weight of edge e(v,u), and W e(v,p) is the weight of edge e(v,p), U v is the set of adjacent vertices of v in the same partition G i , and is the set of adjacent vertices of v in the adjacent partition G j , and U v is the set of adjacent vertices of v in the same partition G i . After calculating the benefit values of all boundary vertices, it is determined whether a vertex needs to move from its current partition to an adjacent partition, and traversed in descending order of the benefit value.

[0022] Preferably, in step 4, when vertex v moves from its original partition G i to the adjacent partition G j , the conditions to be satisfied include the total weight of the partition after movement satisfies the set threshold range and the number of boundary edges does not increase.

[0023] Preferably, in step 2, when defining the optimization problem of road network partitioning based on the optimization objective, for the calculation of workload imbalance, accurately count the total weight W i of the vertices in each partition G i , and calculate and optimize by minimizing the workload imbalance formula.

[0024] Preferably, in step 2, in the objective calculation of minimizing the edge cut, accurately identify all the edges between partitions G i and G j , and obtain their weights W ij , and sum and optimize according to the formula for minimizing the edge cut.

[0025] Preferably, in step 2, in the objective calculation of minimizing the number of boundary edges, accurately count the number of boundary edges between partitions G i and G j , and sum and optimize according to the formula for minimizing the number of boundary edges.

[0026] Preferably, in step 3, when the graph growth makes an initial partition of the road network, the rule for sorting vertices in the priority queue is to first perform a preliminary sorting according to the partition ID, and for vertices with the same partition ID, perform a secondary sorting according to the coordinates.

[0027] Beneficial effects

[0028] The present invention provides a parallel traffic simulation acceleration method for road network load perception, which has the following beneficial effects:

[0029] (1) In the road network preprocessing stage of this parallel traffic simulation acceleration method for road network load perception, this method converts it into a weighted graph and determines the weights based on various information such as traffic density, flow, lane length, and number, abandoning the traditional approach that solely relies on road length. The traditional method is prone to disconnecting the weight setting from the actual traffic conditions, while the method in the present invention can accurately capture the true status of roads in the traffic system. For example, in areas with high traffic flow but short road lengths, the new weights can reflect their importance, avoiding insufficient allocation of computing resources; for long sections with scarce traffic, weights can also be reasonably allocated to prevent resource waste. In this way, a solid and practical foundation is laid for subsequent road network partitioning, effectively ensuring the accuracy of traffic simulation at the basic data level and enabling the simulation results to more realistically reflect the traffic operation status.

[0030] (2) In the process of road network partitioning of this parallel traffic simulation acceleration method for road network load perception, an optimization objective including minimizing workload imbalance, edge cut, and the number of boundary edges is defined. Traditional static road network partitioning methods often cause uneven computing loads, resulting in excessive or insufficient computing resources in some areas and seriously affecting efficiency. However, this method ensures the reasonable allocation of computing resources in each partition by minimizing workload imbalance; minimizing edge cut effectively reduces a large amount of overhead caused by synchronization between regions. Since the road weights are set based on the number of accesses, reducing the edge cut weight can optimize the synchronization cost; minimizing the number of boundary edges reduces the additional burden brought by shadow edges and shadow nodes. These series of optimization measures work closely together to comprehensively improve the parallel simulation performance, greatly accelerating the operation speed of traffic simulation and significantly improving the efficiency of traffic simulation.

[0031] (3) After the initial road network partitioning is carried out by means of the graph growth algorithm in this parallel traffic simulation acceleration method for road network load perception, on the basis of the initial partitions, the partitions are further refined according to the profit function. In this process, the partition structure is flexibly and dynamically adjusted according to the vertex weights and adjacent relationships. When a vertex moves between partitions, by accurately calculating the profit value and combining strict conditional judgments, such as the profit value being greater than 0, the total weight of the partition after movement being within a reasonable threshold range, and ensuring that the number of boundary edges does not increase, etc., it is ensured that each adjustment develops in a more optimal direction. Description of the Drawings

[0032] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary. For those of ordinary skill in the art, without creative efforts, other implementation drawings can also be obtained based on the provided drawings.

[0033] The structures, ratios, sizes, etc. disclosed in this specification are only used to cooperate with the content disclosed in the specification for those who are familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention. Therefore, they do not have technical substantive significance. Any modification of the structure, change of the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention.

[0034] Figure 1 It is a schematic diagram of the preprocessing process for road network load perception;

[0035] Figure 2 It is a schematic diagram of the optimization process for road network division;

[0036] Figure 3 It is a schematic diagram of the initial division process for the growth of the road network map;

[0037] Figure 4 It is a schematic diagram of the zoning algorithm process for the growth of the urban road network diagram. Specific embodiments

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0039] Refer to Figures 1-4 , for the parallel traffic simulation acceleration method for road network load perception, by reasonably dividing the road network to ensure load balance during the reinforcement learning training process, thereby improving the efficiency of traffic simulation for reinforcement learning. The specific steps are as follows:

[0040] Step 1: Preprocess the road network;

[0041] Step 2: Define the optimization problem of road network division based on the optimization goal;

[0042] Step 3: Initially divide the road network through graph growth;

[0043] Step 4: Further refine after the initial zoning to reduce the edge cutting weight;

[0044] Further, Step 1: The preprocessing of the road network includes the following detailed steps:

[0045] S100. Convert the road network into a weighted graph. When using the graph partitioning algorithm, the road network needs to be converted into a weighted graph G=(E, V) first, where E is the set of edges, V is the set of vertices, the nodes in the road network are mapped to the vertices of the graph, and the links between nodes are mapped to the edges between vertices. The partitions after graph decomposition will be mapped back to the road network. Each edge in the road network contains workload information and data dependency information, such as traffic density, flow, lane length and number, etc. Based on this, weights are added to the vertices and edges of the graph.

[0046] S101. Initial weight selection. Most parallel traffic simulations use the basic recursive bisection method for static road network partitioning. The difference lies in the selection of the initial weight of the road. The traditional partitioning method using road length as the initial weight will cause uneven computational load and reduce the computational efficiency during operation.

[0047] S102. Traffic-aware weight: To solve the problem mentioned in S101, the road weight is represented by the traffic-aware weight. Specifically, first traverse all the static routing paths of the vehicles to be simulated, and count the access to each road. For each road e, its traffic-aware weight W e =C e , where C e is the access count of road e.

[0048] S103. Define the traffic-aware weight of intersections: For each vertex v, define its traffic-aware weight w v as follows:

[0049]

[0050] where E v represents the set of roads connected to vertex v, L e is the length of road e, w′ v is the weighted perception weight of vertex v; all vertices are assigned as the base weight. Even in the case of no expected traffic, they also participate in the partitioning correctly. w v is the final traffic-aware weight of vertex v.

[0051] Further, Step 2: Based on the optimization objective, define the optimization problem of road network partitioning. Specifically,

[0052] The graph partitioning algorithm divides the road network G into m non-overlapping partitions G={G 0 ,G1 ,..., G m-1}. Suppose the partition G i (0 ≤ i < m) has a vertex set V i Then And To improve the performance of parallel simulation, the following three optimization goals are considered during the partitioning process: The first goal is to minimize the workload imbalance. Suppose W i is the total weight of the vertices in V i .

[0053] The first goal is expressed as:

[0054] where is the average weight of all partitions.

[0055] The second goal is to minimize the edge cut, which is the total weight of the edges between all partitions. Since we set the weight of each road to the number of times it is visited, minimizing the edge cut can reduce the overhead of synchronization between regions. Suppose the weight of all edges between partition G i and G j is W i,j (i ≠ j), then the second goal is expressed as:

[0056]

[0057] The third goal is to minimize the number of boundary edges. Considering the introduced shadow edges and shadow nodes, by minimizing the number of boundary edges, the additional overhead of shadow edges and shadow nodes can be reduced. Suppose the number of boundary edges between partition G i and G j is n i,j (i ≠ j), the third goal is expressed as:

[0058]

[0059] Furthermore, Step 3: The road network is initially partitioned by graph growth. The weighted graph G generated from the road network is first initially partitioned by graph growth. The essence of graph growth lies in sorting the vertices in the priority queue. Each item in the priority queue is actually a tuple (i, v), which consists of the partition ID assigned to the vertex and the vertex itself. The vertices are mainly sorted according to the partition ID marked in their tuples, and secondly according to their coordinates. These two rules have a direct impact on the number of adjacent partitions and the number of boundary edges. In this paper, the above method is used to achieve Goal 1 and Goal 3. Starting from the initial vertex, the subgraph grows along the edges of the graph one by one.

[0060] The graph growth algorithm plays a crucial role in the road network partitioning process. Among them, the priority queue is an important tool for controlling the vertex access order. During the algorithm execution, when the cumulative weight of vertices in a certain partition exceeds the average weight, a new vertex partition operation will be considered. At the same time, for all adjacent vertices that have not been queued before the current vertex, they are all pushed into the queue and marked as queued, thus ensuring the orderliness and rationality of the graph growth process and laying a foundation for the subsequent road network partitioning. The specific algorithm process is as follows:

[0061] Input: Urban road network G, number of partitions m.

[0062] Output: Set of road network vertex partitions {V i}

[0063] 1), Initialize Q ← empty set

[0064] 2), Determine the initial vertex V of graph growth init ,

[0065] 3), Add a tuple (i, v init ) to the priority queue Q and mark V init as queued (enqueue)

[0066] 4), When Q is not empty, perform the following operations

[0067] 5), Pop a tuple from Q, assume its vertex is v

[0068] 6), Judge whether a new partition needs to be started

[0069] 7), Add vertex v to the current partition node set V i

[0070] 8), foreach vertex u, e(u, v) ∈ E do

[0071] 9), if u has not been queued before then

[0072] 10), Add a tuple (i, u) to the priority queue Q and mark u as queued (enqueue)

[0073] 11), Go back to {V i}

[0074] Through the above steps, the graph growth algorithm is executed orderly, continuously partitioning the vertices of the urban road network. During this process, the partitioning situation is dynamically adjusted according to vertex weights and adjacency relationships to ensure that the construction of each partition complies with the rules set by the algorithm. Finally, a set of road network vertex partitions is successfully generated, laying a preliminary network structure foundation for subsequent traffic simulation acceleration, enabling the entire traffic simulation to be carried out more efficiently based on a reasonable road network division, achieving the goal of effectively initially partitioning the road network through the graph growth algorithm to a certain extent, and ensuring the coherence and feasibility of the entire traffic simulation process.

[0075] Further, Step 4: Refinement after initial partitioning further reduces the edge cut weight. After the initial partitioning, the partitions are refined to reduce edge cuts to achieve Goal 2. First, a benefit function for moving a certain vertex v between partitions is defined. Assume moving vertex v from partition G i to G j . At this time, it is equivalent to converting the edge weight between vertex v and region Gj into an internal weight. Similarly, the internal weight of vertex v will be converted into an edge weight. Therefore, the benefit value of this move is obtained:

[0076]

[0077] where W e(v,u) is the weight of edge e(v,u), W e(v,p) is the weight of edge e(v,p), U v is the set of adjacent vertices of v in the same partition G i , is the set of adjacent vertices of v in the adjacent partition G j , U v is the set of adjacent vertices of v in the same partition G i . After calculating the benefit values of all edge vertices, then determine whether the vertex needs to move from its current partition to an adjacent partition, traversing in descending order of the benefit value. For v to move from its original partition G i to the adjacent partition G j , the following conditions need to be met:

[0078] (1) That is, the benefit value is greater than 0;

[0079] (2) When v moves from partition G i to partition G j , the total weight W i in partition G i needs to satisfy

[0080] where α and β are the upper and lower bound threshold parameters of the weight respectively, to ensure the load balance of each partition during the subdivision process.

[0081] (3) After moving v, it is necessary to ensure that the number of boundary edges does not increase. Whenever a vertex is moved, the profit values of its adjacent vertices are updated, and the boundary vertex set is updated. When all the moving situations of the boundary vertices have been traversed, one refinement is completed. Repeat this process until no vertex moves in one traversal or the number of traversals reaches a predefined maximum value.

[0082] In the present invention, by preprocessing the road network, defining the optimization objective, initial partitioning of graph growth to partition refinement, the key problems in traffic simulation are comprehensively and deeply solved, thus showing great potential in improving parallel simulation performance and providing an innovative and effective technical approach for the research and practice in the traffic field.

[0083] The above has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A road network load-aware parallel traffic simulation acceleration method, characterized in that: The following steps are involved: Step 1: Preprocess the road network; Step 2: Based on the optimization goal, define the optimization problem of road network division; Step 3: Initially divide the road network through graph growth; Step 4: After the initial partition, refinement is performed to further reduce the edge cut weights; Wherein, step 1 specifically includes the following detailed steps: S100, converting the road network into a weighted graph; S101, initial weight selection; S102, traffic perception weight; S103: Define intersection traffic perception weights.

2. The road network load-aware parallel traffic simulation acceleration method according to claim 1, characterized in that: In step 1, the road network is converted into a weighted graph G = (E, V), where E is a set of edges, V is a set of vertices, the nodes in the road network are mapped to the vertices of the graph, and the links between the nodes are mapped to the edges between the vertices. The traffic-aware weight W is used. e =C e represents the road weight, C e is the visit count of road e, and defines the intersection traffic awareness weight: Where E v represents the set of paths connected to vertex v, L e is the length of road e, w′ v is the weighted perception weight of vertex v; all vertices are assigned As the basic weight, w v is the traffic-aware weight of the final vertex v.

3. The road network load-aware parallel traffic simulation acceleration method according to claim 1, characterized in that: In step 2, the graph G is divided into m non-overlapping partitions G = {G0, G1, ..., G m-1 }, the optimization objectives are: minimize workload imbalance Where W i Yes V i The total weight of the vertices in V i It is partition G i The vertex set in is the average weight of all partitions; Minimize edge cutting where W i,j is the weight of all edges between partitions G i and G j for 0 ≤ j < m; minimize the number of boundary edges where n i,j is the number of boundary edges between partitions G i and G j ​ 4. The road network load-aware parallel traffic simulation acceleration method according to claim 1, characterized in that: In step 3, the weighted graph generated for the road network is first initially divided by graph growth. The graph growth is based on sorting the vertices in the priority queue. The vertices are sorted according to the partition ID and coordinates marked in their tuples. Starting from the initial vertex, the subgraph grows along the edge of the graph. When the cumulative weight of the vertices in the partition exceeds the average weight, a new vertex partition is started, and all adjacent vertices that have not been queued before the current vertex are pushed into the queue and marked as queued.

5. The road network load-aware parallel traffic simulation acceleration method according to claim 3, characterized in that: In step 4, after the initial partitioning, the benefit function of moving a vertex v between partitions is defined as Where W e(v,u) is the weight of edge e(v,u), W e(v,p) is the weight of edge e(v,p), U v For v in the same partition G i The set of adjacent vertices in , For adjacent partition G j The set of adjacent vertices of v in , U v Is v in the same partition G i After calculating the benefit values ​​of all edge vertices, determine whether the vertex needs to be moved from its current partition to the adjacent partition, and traverse from high to low according to the benefit value.

6. The road network load-aware parallel traffic simulation acceleration method according to claim 5, characterized in that: In step 4, vertex v is removed from its original partition G i Move to adjacent partition G j , the conditions that need to be met include After the move, the total weight of the partition meets the set threshold range and the number of boundary edges does not increase.

7. The road network load-aware parallel traffic simulation acceleration method according to claim 1, characterized in that: In step 2, when defining the optimization problem of road network partitioning based on the optimization objective, for the calculation of workload imbalance, accurately count the G of each partition. i The total weight W of the vertices in i , and minimize the workload imbalance formula for calculation and optimization.

8. The road network load-aware parallel traffic simulation acceleration method according to claim 3, characterized in that: In step 2, in the objective calculation of minimizing edge cutting, the partition G is accurately identified. i and G j All the edges between and get their weights W ij , sum and optimize according to the minimum edge cutting formula.

9. The road network load-aware parallel traffic simulation acceleration method according to claim 3, characterized in that: In step 2, in the objective calculation of minimizing the number of boundary edges, the partition G is accurately counted. i and G j The number of boundary edges between them is summed and optimized according to the formula for minimizing the number of boundary edges.

10. The road network load-aware parallel traffic simulation acceleration method according to claim 4, characterized in that: In step 3, when graph growth performs initial partitioning on the road network, the rule for sorting vertices in the priority queue is to first perform preliminary sorting based on the partition ID, and then perform secondary sorting based on the coordinates for vertices with the same partition ID.

Images