Graph-integrated traffic flow prediction method and traffic control management system

Abstract

Disclosed are a multi-layer attention-based graph fusion traffic flow prediction method, medium and device, which relate to the field of time series data prediction.SOLUTION: In the present disclosure, the historical traffic flow data of all roads in the target road network is first sampled at different sampling intervals to obtain a plurality of historical flow sequences representing different data periods for each road. Then, the road spatial relation graph and the road function similarity relation graph corresponding to each data cycle are constructed for all roads of the target road network. Then, adjacency matrix attention fusion is performed on the road spatial relationship graph and the road functional similarity relationship graph corresponding to each data cycle. Finally, the fused adjacency matrix and the vertex feature matrix under each data period are input into the graph attention network, and weighted fusion is performed on output results to obtain a future flow prediction result of the target road network. The present invention can capture complex traffic flow patterns of different roads, and realize high-precision traffic flow prediction.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention belongs to the field of time-series data prediction, and specifically Graph-integrated traffic flow prediction method and traffic control management system relates to

Background Art

[0002] As the number of automobiles in possession continues to increase, the demand for urban traffic system control has been growing stronger day by day. The construction of an efficient smart traffic system has become an important task to support the scientific management decision-making of transportation agencies. Among them, the realization of accurate traffic flow prediction is considered an important component of the smart traffic system. Accurate traffic flow prediction not only provides real-time data support to traffic managers but also helps in formulating scientific and effective control strategies. By analyzing past traffic data and real-time traffic information, the future traffic flow situation can be predicted timely and accurately, thereby pre-emptively responding to potential traffic jams and changes in traffic conditions. How to make full use of past traffic data and real-time traffic information is an important issue in traffic flow prediction, and the introduction of machine learning algorithms is one of the means to solve this problem.

[0003] Initial machine learning methods mainly focused on a specific road itself, that is, predicting future traffic flow by learning the trend of past traffic data. However, since the traffic flow situations between roads affect each other, these methods cannot effectively extract the dependence relationship between the traffic flows of roads. In recent years, deep learning, especially the adoption of graph neural networks for traffic flow prediction, has gradually become the main means. Graph neural networks can effectively learn the complex dependence relationships between roads due to their powerful spatial modeling capabilities. However, existing graph neural network prediction models mainly consider the relationships between adjacent roads and ignore many other important factors that affect road traffic flow, so it is necessary to improve their prediction accuracy.

Summary of the Invention

[0004] The objective of the present invention is to provide a graph-fused traffic flow prediction method, medium, and apparatus based on multi-layer attention, in order to address the aforementioned problems in graph neural networks for road flow prediction in the prior art. The present invention proposes a new algorithm for calculating road function similarity, constructs graphs of two different relationships, fuses them, aggregates the information, and then simultaneously extracts short-term time-series information and global time-series information to achieve accurate traffic flow prediction.

[0005] The specific technical solution employed in this invention is as follows: In the first embodiment, the present invention provides a graph-fused traffic flow prediction method based on multilayer attention, which includes the following: S1 performs uniform sampling of historical traffic flow data for all roads in the target road network using different sampling intervals to obtain multiple historical flow sequences representing different data periods for each road. The sampling intervals include at least three types: minute level, day level, and week level. S2 constructs a road spatial relationship graph for all roads in the target road network based on the interconnection between roads. The road spatial relationship graph consists of a first vertex set, a first edge set, and a first adjacency matrix. Each vertex in the first vertex set represents a single road, and the vertex attribute is the past flow sequence of the road represented by the vertex. Each edge in the first edge set represents the existence of a spatial interconnection relationship between two roads corresponding to two vertices, and the first adjacency matrix is ​​used to record the edge connection relationships between vertices. S3 constructs a road function similarity relationship graph for each data period based on the functional similarity between roads, using the past flow sequence for each data period for all roads in the target road network. The road function similarity relationship graph consists of a second vertex set, a second edge set, and a second adjacency matrix. The second vertex set is the same as the first vertex set, and each edge in the second edge set represents the existence of a road function similarity relationship between two roads corresponding to two vertices. The second adjacency matrix is ​​used to record the edge connection relationships between vertices. For a road function similarity relationship to exist between two roads corresponding to two vertices, the road geospatial distance must be less than the first threshold and the road past flow sequence distance must be less than the second threshold. The road geospatial distance is the normalized edit distance between the spatial subgraphs centered on the two vertices in the road spatial relationship graph. The road past flow sequence distance is the normalized sequence distance between the past flow sequences for the current data period of the roads corresponding to the two vertices. S4 performs adjacency matrix fusion between the road space relationship graph and the road function similarity relationship graph corresponding to each data period using an attention mechanism, thereby obtaining a fused adjacency matrix corresponding to each data period. S5 obtains a vertex feature matrix by embedding the past flow sequences of all roads in the target road network for each data cycle. Furthermore, it inputs the fused adjacency matrix and vertex feature matrix for each data cycle into a graph attention network to obtain output results corresponding to each data cycle. Weighted fusion is performed on the output results corresponding to all data cycles to finally obtain future flow prediction results for the target road network.

[0006] As a preferred example of the first embodiment, the data period is divided into three types, and the sampling intervals correspond to the minute level, the day level, and the week level, respectively.

[0007] As a preferred example of the first embodiment described above, the data period is divided into three types, with sampling intervals of 15 minutes, 24 hours, and 7 days, respectively.

[0008] JPEG0007849680000001.jpg58152

[0009]

Number

[0010]

Number

[0011]

Number

[0012]

Number

[0013]

Number

[0014] JPEG0007849680000009.jpg52153

[0015]

Number

[0016]

Number

[0017]

Number

[0018]

Number

[0019] As a preferred example of the first embodiment described above, the graph attention network is constructed by stacking multilayer graph attention layers. The input to the attention layer of each layer graph is the fused adjacency matrix and the vertex feature vector matrix output by the graph attention layer of the previous layer. Within the layer, multi-head self-attention is used to aggregate the feature vectors of adjacent nodes to each central node, and a new vertex feature vector matrix is ​​output to the next layer. The outputs of all attention layers are concatenated and then output by an exponential linear unit activation function, and finally processed by an additional graph attention layer to obtain the output result of the graph attention network.

[0020] As a preferred example of the first embodiment described above, when weighted fusion of output results corresponding to all data cycles, the Hadamard product of the output result corresponding to each data cycle and the learnable parameter matrix is ​​calculated, and then the results of each Hadamard product are summed to obtain a future flow rate prediction result for the target road network.

[0021] In a second embodiment, the present invention provides a computer program product including a computer program / instruction. When the computer program / instruction is executed by a processor, the multilayer attention-based graph-fused traffic volume prediction method described in any of the first embodiments is realized.

[0022] In a third embodiment, the present invention provides a computer-readable storage medium in which a computer program is stored, and when the computer program is executed by a processor, the multilayer attention-based graph-fused traffic volume prediction method described in any of the first embodiments is realized.

[0023] In a fourth embodiment, the present invention provides a computer device including memory and a processor.

[0024] The memory mentioned above is for storing computer programs.

[0025] The above-mentioned processor is for realizing the multilayer attention-based graph-fused traffic volume prediction method described in any of the first embodiments when the computer program is executed.

[0026] The present invention has the following beneficial effects compared to the conventional technology. This invention constructs two types of graphs based on road spatial relationships and road function similarity relationships, and fuses these two graphs to perform flow prediction for each road. By considering not only the traffic volume of neighboring road sections but also the traffic volume of roads with similar functions, it provides more accurate prediction results. When constructing the road function similarity relationship graph, this invention considers both road change laws and past flow sequences to ensure the accuracy of similar function road determination and to learn more accurate information during graph modeling. Based on a graph attention network, this invention can learn different traffic flow relationships in parallel using multi-head self-attention, thus improving the model's expressive power, capturing complex traffic flow modes between different points, and improving prediction accuracy. [Brief explanation of the drawing]

[0027] [Figure 1] This is a step-by-step flowchart of a graph-integrated traffic flow prediction method based on multi-layer attention. [Figure 2] This is a schematic diagram of the process for calculating the geospatial distance between roads corresponding to two vertices. [Figure 3] This is a schematic diagram of the process for calculating the past road flow sequence distance between roads corresponding to two vertices. [Figure 4] This is a schematic diagram of the configuration of the graph-integrated traffic flow prediction model. [Modes for carrying out the invention]

[0028] To make the above-mentioned objectives, features, and advantages of the present invention clearer and easier to understand, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The following description includes many specific details in order to fully understand the present invention. However, the present invention can be carried out in many ways different from those described herein, and those skilled in the art can make similar improvements without contradicting the present invention, so the present invention is not limited to the specific embodiments disclosed below. The technical features of each embodiment of the present invention can be combined, provided that they do not conflict with each other.

[0029] In this description of the present invention, the terms “first” and “second” are used solely to distinguish between the purposes of the description and are not intended to indicate or imply relative importance or the number of technical features to be described. Accordingly, features designated as “first” or “second” may include at least one such feature, either explicitly or implicitly.

[0030] In a preferred embodiment of the present invention, a graph-fused traffic flow prediction method based on multi-layer attention is provided, the specific steps of which are shown in S1 to S5. In this method, in addition to constructing a road spatial relationship graph that records the spatial connection relationships between roads, a new road function similarity calculation algorithm is introduced to construct a road function similarity relationship graph, these two different relationship graphs are fused, and then richer information is extracted from the fused information based on a graph attention network to achieve accurate traffic flow prediction. As shown in Figure 1, the specific implementation forms of the above steps S1 to S5 will be described in detail below.

[0031] S1, For all roads in the target road network, uniform sampling of the same sample amount is performed using different sampling intervals to obtain multiple historical flow sequences representing different data periods for each road. The sampling intervals include at least three types: minute level, day level, and week level. The specific number and length of the above data periods may be optimized based on the final prediction performance, but it should be noted that they should include at least three types of data periods corresponding to sampling intervals at the minute, day, and week levels.

[0032] In this invention, the minute level refers to a sampling interval of one or ten minutes, representing the flow rate data situation relatively close to the current time. The reason for considering the minute-level data period is that the flow rate situation relatively close to the current time directly influences the subsequent flow rate change trend.

[0033] In this invention, the daily level refers to flow rate data conditions during the same time period as the present time on past dates, with a sampling interval of one day. The reason for considering the daily level data cycle is that there are often similarities in flow rate conditions during the same time period each day, and corresponding information can be extracted from this to assist in flow rate prediction.

[0034] In this invention, the weekly level refers to flow rate data conditions during the same time period as the present time in past weeks, with a week as the sampling interval. The reason for considering the weekly data cycle is that similar patterns of change often exist in flow rate conditions at the same time each week, and corresponding information can be extracted from this to assist in flow rate prediction.

[0035] JPEG0007849680000016.jpg33153

[0036] To facilitate the subsequent explanation of the present invention, regardless of the different sampling intervals, the number of samples taken is always the same; that is, the dimensions of the past flow sequence are the same for different data periods, and the dimensions of the past flow sequence, i.e., the number of samples taken, can be denoted as T. The last past time before the time period to be predicted in the present invention can be denoted as the current time t, and all traffic flow data that constitute the past flow sequence are road traffic flow before the current time t. In actual application scenarios, this road traffic flow data can be acquired by flow sensors installed on the road.

[0037] In practical applications, traffic flow sensors can be placed on roads within the target road network to acquire historical traffic flow data for each road. There are various types of traffic flow sensors, such as camera surveillance systems or vehicle speed detection radars. Camera surveillance systems are typically installed at intersections on main roads, highway entrances and exits, congested road sections, and around large event venues. They capture real-time images of traffic volume via cameras, acquiring information such as the number of vehicles, vehicle type, direction of travel, and speed. Vehicle speed detection radars, on the other hand, are typically installed along the roadside or at stands on highways, expressways, and main roads, using microwave or laser reflection principles to measure the speed and volume of passing vehicles. The specific placement and density of traffic flow sensors can be determined based on the actual conditions of the target road network, ensuring that traffic flow data for all roads within the network can be recorded.

[0038] Furthermore, in practical applications, in addition to placing traffic flow sensors on roads within the target road network, IoT devices with data transmission and processing capabilities can also be placed to form edge nodes. All sensor data acquired by the traffic flow sensors must undergo certain data processing at the edge nodes before being transmitted in real time to a central data processing system via IoT communication equipment with transmission capabilities. Real-time transmission of sensor data will prioritize the use of low-latency, highly stable wireless network technologies (e.g., 4G / 5G or LoRa) to ensure that traffic flow information is collected in the central data processing system in real time. Data will be encrypted during transmission to guarantee data integrity and security. The hardware configuration of the central data processing system can employ a server equipped with a multi-core CPU and GPU, or an FPGA cluster that supports real-time data processing. The central data processing system receives and processes stream data from multiple edge nodes using a big data processing platform (e.g., Apache Kafka and Apache Spark), aggregates and stores the data, and further performs uniform sampling of historical traffic flow data for all roads in the target road network using different sampling intervals to obtain multiple historical traffic flow sequences representing different data periods for each road.

[0039] S2. For all roads in the target road network, a road spatial relationship graph is constructed based on the interconnection between roads. The road spatial relationship graph consists of a first vertex set, a first edge set, and a first adjacency matrix. Each vertex in the first vertex set represents a single road, and the vertex attribute is the past flow sequence of the road represented by the vertex. Each edge in the first edge set represents the existence of a spatial interconnection relationship between two roads corresponding to two vertices, and the first adjacency matrix is ​​used to record the edge connection relationships between vertices.

[0040] JPEG0007849680000017.jpg83155

[0041] S3. For all roads in the target road network, a road function similarity relationship graph is constructed for each data period based on the functional similarity between roads, using the past flow sequence for each data period. The road function similarity relationship graph consists of a second vertex set, a second edge set, and a second adjacency matrix. The second vertex set is the same as the first vertex set, and each edge in the second edge set represents the existence of a road function similarity relationship between two roads corresponding to two vertices. The second adjacency matrix is ​​used to record the edge connection relationships between vertices. For a road function similarity relationship to exist between two roads corresponding to two vertices, the road geospatial distance must be less than the first threshold and the road past flow sequence distance must be less than the second threshold. The road geospatial distance is the normalized edit distance between the spatial subgraphs centered on the two vertices in the road spatial relationship graph. The road past flow sequence distance is the normalized sequence distance between the past flow sequences in the current data period for the roads corresponding to the two vertices.

[0042] Furthermore, the aforementioned road function similarity relationship graphs must be constructed separately for each data period. While the construction method is the same, the historical flow sequences used in the construction process differ. For each data period, the road function similarity relationship graph needs to measure the similarity of past road flow sequences using the historical flow sequences corresponding to that data period.

[0043] JPEG0007849680000018.jpg76153

[0044] JPEG0007849680000019.jpg80153

[0045] In this invention, when determining whether a road function similarity relationship exists between two roads corresponding to two vertices, it is necessary to simultaneously consider two dimensions: the road geospatial distance and the road past flow sequence distance between these two roads. Of these, the smaller the road geospatial distance, the more similar the spatial topological relationship of the local road network is, and the smaller the road past flow sequence distance, the more similar the laws governing changes in traffic flow on the road are. These two dimensions can, to some extent, represent the degree of functional similarity between the roads.

[0046] JPEG0007849680000020.jpg108164

[0047] JPEG0007849680000021.jpg14153

[0048]

number

[0049]

number

[0050]

number

[0051]

number

[0052]

number

[0053] JPEG0007849680000029.jpg23155

[0054] JPEG0007849680000030.jpg69166

[0055]

number

[0056]

number

[0057]

number

[0058]

number

[0059] JPEG0007849680000037.jpg65153

[0060] JPEG0007849680000038.jpg165156

[0061] JPEG0007849680000039.jpg30157

[0062] JPEG0007849680000040.jpg52153

[0063]

number

[0064]

number

[0065] In step S5, the past flow sequences of all roads in the target road network for each data cycle are embedded to obtain a vertex feature matrix. The fused adjacency matrix and vertex feature matrix for each data cycle are then input into a graph attention network to obtain output results corresponding to each data cycle. Weighted fusion is performed on the output results corresponding to all data cycles to finally obtain future flow prediction results for the target road network.

[0066] JPEG0007849680000044.jpg85153

[0067] In embodiments of the present invention, for the sake of facilitating the following explanation, the number of stacked layers is denoted as F, and the GAT is composed of F layers of graph attention layers. The specific value of F may be optimized, and in this embodiment, it is optimized to F=6. Each layer of graph attention layers may have different weights and attention parameters, and the calculation process for the attention weights is expressed by the following formula.

[0068]

number

[0069] Furthermore, when the present invention updates the vertex feature vector matrix, the multi-head self-attention method employed can be expressed as follows.

[0070]

number

[0071] JPEG0007849680000049.jpg27154

[0072] JPEG0007849680000050.jpg15153

[0073]

number

[0074] JPEG0007849680000053.jpg14153

[0075]

number

[0076] JPEG0007849680000056.jpg31153

[0077]

number

[0078] As can be seen from the above, the S1-S5 steps of the present invention enable accurate prediction of future traffic flow of the target road network by modeling on past traffic flow data. However, it should be noted that the entirety of S1-S5 corresponds to a single graph-fused traffic flow prediction model, and its model structure is as shown in Figure 4, and that it contains many learnable parameters. Before using the above graph-fused traffic flow prediction model for actual inference, it is necessary to perform supervised learning on a training dataset in advance to obtain the optimal parameters necessary for accurate inference. Supervised learning of a model belongs to the prior art and will not be discussed in this invention.

[0079] The data processing in steps S2 to S5 described above can all be performed on a central data processing system. Furthermore, after obtaining the future traffic flow prediction results for the target road network through steps S1 to S5, the central data processing system can apply this to actual traffic management. For example, it can transmit the future traffic flow prediction results for the target road network to a traffic control management system, thus facilitating traffic management and control by the traffic management department.

[0080] In embodiments of the present invention, the traffic control management system can set appropriate control functions according to the demand, and the selectable control function modules are as follows. This traffic flow visualization module visualizes the historical, current, and future traffic flow conditions of a target road network within the GUI interface of a traffic control management system. A real-time traffic signal control module allows a traffic control management system to adjust the duration of traffic signals at different intersections in real time using built-in control logic based on the received future traffic flow forecast results for the target road network. For example, for an intersection with traffic signals, the traffic control management system optimizes the duration of the signal for each direction based on the current vehicle flow, vehicle speed, congestion level, and future vehicle flow forecast results for the four directions of the intersection, reducing the vehicle flow that will soon enter a section where congestion is expected. If traffic volume increases significantly in a particular direction at the intersection, the system increases the green light duration for that direction to allow more vehicles to pass through smoothly. Because an intersection has both a sidewalk and a roadway, a sidewalk that does not partially align with the direction of traffic on the roadway (e.g., a sidewalk that intersects perpendicularly with the roadway) directly reduces the efficiency of traffic on the roadway. If the system predicts that traffic volume will reach its peak (e.g., during rush hour or off-peak hours), it may adaptively extend the green light duration for the roadway and reduce the green light duration for the sidewalk. This early warning module for vehicle flow displays, when a vehicle flow display device is installed in the road network, will display the current and future vehicle flow of the road section ahead where the display device is located, allowing drivers to understand the vehicle flow situation on the road ahead in real time. In road networks where smart network vehicles are relatively widespread, if it is predicted that the future vehicle flow of a road section on the vehicle's route may cause congestion, the system will transmit early warning information to the in-vehicle terminal or navigation software, informing the vehicle owner of the possibility of congestion ahead and advising them to plan a congestion avoidance route in advance.

[0081] To more clearly demonstrate the technical effects of the embodiments of the present invention, the methods described in S1 to S5 of the above embodiments were trained on a training dataset processed with actual measured data as a single graph-fused traffic flow prediction model, and its performance was verified on a test set. At the same time, in order to show the difference in effectiveness compared with conventional models, the embodiments of the present invention were further compared with five types of conventional models, and the mean absolute error (MAE) and root mean square error (RMSE) were used as evaluation indicators. The five types of conventional models are a historical average (HA) model, a vector autoregression (VAR) model, a long short-term memory (LSTM) model, and a gated recurrent unit (GRU) model. The results of the comparative experiment are shown in Table 1.

[0082] [Table 1] Table 1 shows that the prediction accuracy of the graph-integrated traffic flow prediction model of the present invention is superior to that of the four conventional models.

[0083] Furthermore, to verify the impact of introducing the road space relationship graph and the road function similarity relationship graph in this invention on prediction performance, two sets of ablation experiments were performed for each. One set used only the road space relationship graph, and the other set used only the road function similarity relationship graph. The removal of a particular graph in the ablation experiment can be adjusted by the calculation process of the fused adjacency matrix A. Finally, the results of the ablation experiments in this embodiment are shown in Table 2.

[0084] [Table 2] Table 1 shows that in the graph-integrated traffic flow prediction model of the present invention, the introduction of both the road space relationship graph and the road function similarity relationship graph improves the final prediction accuracy to some extent. When only the road space relationship graph is used, the RMSE index is slightly higher than that of the present invention, but the improvement is not significant, and the MAE index is clearly inferior to that of the present invention. Therefore, overall, by introducing both the road space relationship graph and the road function similarity relationship graph simultaneously, it is possible to ensure that the overall prediction performance is optimally achieved by considering not only the traffic flow of neighboring sections but also the traffic flow of roads with similar functions during the prediction process.

[0085] Furthermore, the method steps described in S1 to S5 above can essentially be implemented in the form of a computer program.

[0086] Therefore, based on the same inventive concept, the present invention corresponds to the multilayer attention-based graph fusion traffic flow prediction method provided in the above embodiment and provides a computer electronic device including memory and a processor. The aforementioned memory is used to store computer programs. The processor is for realizing the graph-fused traffic flow prediction method based on multi-layer attention when the computer program is executed. The logical instructions in the memory may be implemented in the form of a software function unit and, if sold or used as an independent product, may be stored on a single computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be represented in the form of a computer program product, which is stored on a storage medium and contains several computer programs / instructions that cause a processor of a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention.

[0087] Based on the same inventive concept, the present invention provides a computer-readable storage medium corresponding to a graph-fused traffic flow prediction method based on multilayer attention, the storage medium storing a computer program for realizing the graph-fused traffic flow prediction method based on multilayer attention when the computer program is executed by a processor.

[0088] Based on the same inventive concept, the present invention provides a computer program product that includes a computer program / instruction, and when the computer program / instruction is executed by a processor, it is possible to realize the graph-fused traffic flow prediction method based on multi-layer attention as described above.

[0089] Specifically, the computer-readable storage medium according to the three embodiments can execute steps S1 to S5 when the stored computer program is executed by the processor.

[0090] It can be understood that the storage medium may include random access memory (RAM) and at least one non-volatile memory (NVM), such as a magnetic disk memory. Thus, the storage medium may be various media capable of storing program code, such as a USB memory stick, a removable hard disk, a magnetic disk, or an optical disk.

[0091] The processor may be a general-purpose processor including a Central Processing Unit (CPU), a Network Processor (NP), etc. It may also be a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component.

[0092] It should be noted that, for the sake of convenience and brevity of explanation, the specific operating procedures of the system described above are not described repeatedly here, as they may be described by referring to the corresponding procedures in the embodiments of the method described above, as will be clearly understood by those skilled in the art. In each embodiment provided in this application, the distinction between steps or modules in the system and method described above is merely a logical functional distinction, and in actual implementation, multiple modules or steps may be combined or integrated, or a single module or step may be divided.

[0093] The embodiments described above are merely preferred embodiments of the present invention and do not limit it. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Claims obtained using the above equivalent substitutions or equivalent transformations are within the scope of protection of the present invention.

Claims

1. A graph-fused traffic flow prediction method based on a multilayer graph attention layer, which is executed by the processor of a computer device as a computer program / instruction, Step S1 (Step 1) involves performing uniform sampling of the same sample amount from past traffic flow data for all roads in the target road network using different sampling intervals, thereby obtaining multiple past flow sequences representing different data periods for each road, wherein the sampling intervals include at least three types: minute level, day level, and week level. Step S2 (Step 2) constructs a spatial connectivity relationship graph between roads based on the connectivity between roads for all roads in the target road network. The spatial connectivity relationship graph between the roads consists of a first vertex set, a first edge set which is a set of first edges representing spatial connectivity between the first vertices, and a first adjacency matrix. Each vertex in the first vertex set represents a single road, the vertex attribute is the past flow sequence of the road represented by the vertex, each first edge in the first edge set represents the existence of spatial connectivity between two roads corresponding to two vertices, and the first adjacency matrix is ​​used to record the edge connection relationships between vertices. Step S3 constructs a road function similarity relationship graph corresponding to each data period for all roads in the target road network, based on the past flow rate sequence for each data period and the functional similarity between roads. The aforementioned road function similarity relationship graph consists of a second vertex set, a second edge set, and a second adjacency matrix, wherein the second vertex set is the same as the first vertex set, each second edge in the second edge set represents the existence of a road function similarity relationship between two roads corresponding to two vertices, and the second adjacency matrix is ​​used to record the edge connection relationships between vertices. Here, for a road function similarity relationship to exist between the roads corresponding to the two vertices, the road geospatial distance must be less than the first threshold and the road past flow sequence distance must be less than the second threshold. The road geospatial distance is the normalized graph edit distance (GED) between spatial subgraphs that each contain the two vertices in the spatial connection relationship graph between roads, and the road past flow sequence distance is the normalized sequence distance representing the similarity between past flow sequences for the same data period on the roads corresponding to the two vertices. Step S4 (Step 4) is performed by weighted summation by distributing appropriate weights to obtain a new fused adjacency matrix. This is done by an attention mechanism that fuses the adjacency matrices of the spatial communication relationship graph between the roads and the road function similarity relationship graphs corresponding to each data period, thereby obtaining a fused adjacency matrix corresponding to each data period. Step S5 (Step 5) involves embedding the past flow sequences of all roads in the target road network for each data period to obtain a matrix representing vertex features. Furthermore, the fused adjacency matrix and vertex feature matrix for each data cycle are input into a Graph Attention Network (GAT), output results corresponding to each data cycle are obtained, and weighted fusion is performed by applying weights to the output results corresponding to all data cycles, along with the outputs of traffic flow prediction results for different data cycles (minute level, daily level, weekly level, etc.), to finally obtain future traffic flow prediction results for the target road network. Including each of the above steps, The graph-fused traffic flow prediction method based on a multilayer graph attention layer is characterized in that the aforementioned past traffic flow data is past traffic flow data for each road obtained using traffic flow sensors placed on roads within the target road network.

2.

3. The aforementioned Graph Attention Network (GAT) is composed of a stack of multilayer graph attention layers. The input to each layer of the graph attention layer is the fused adjacency matrix and a vector matrix of vertex features output by the previous layer's graph attention layer. Within each layer, multi-head self-attention is used to aggregate the feature vectors of adjacent nodes to each central node, and a new vector matrix of vertex features is output to the next layer. The graph-fused traffic flow prediction method based on a multilayer graph attention layer according to claim 1, characterized in that the outputs of all graph attention layers are output by an exponential linear unit activation function (ELU) after concatenation, and finally processed by an additional graph attention layer to obtain the output result of the graph attention network (GAT).

4. The method for performing the weighted fusion on output results corresponding to all the aforementioned data periods is to calculate the Hadamard product of the output result corresponding to each data period and the parameter matrix that can be learned by the steps S1 (step 1) to S5 (step 5), and then sum up the results of each Hadamard product calculation to obtain a future flow prediction result for the target road network, characterized in that the graph fusion traffic flow prediction method based on a multilayer graph attention layer according to claim 1.

5. The traffic flow sensor is a camera monitoring system or a vehicle speed detection radar, The aforementioned camera monitoring system is installed around intersections on main roads, highway entrances and exits, road sections prone to congestion, and large event venues. It captures real-time images of traffic volume via cameras and can acquire information on at least the number of vehicles, vehicle type, direction of travel, and speed. The aforementioned vehicle speed detection radar is installed on the roadside or at a gas station on a highway, expressway, or main road, and can measure the speed and traffic volume of passing vehicles using the reflection principle of microwaves or lasers. A graph-fused traffic flow prediction method based on a multilayer graph attention layer as described in feature 1.

6. In addition to placing traffic flow sensors on roads within the aforementioned road network, IoT devices with data transmission and processing capabilities are also placed, and these IoT devices form edge nodes. The sensor data acquired by all traffic flow sensors is processed at the edge node and then transmitted in real time to the central data processing system via the communication equipment of the IoT device having transmission capabilities. The central data processing system receives and processes data from multiple edge nodes through a big data processing platform, aggregates and stores the data, and then performs uniform sampling of historical traffic flow data for all roads in the target road network, using different sampling intervals to obtain multiple historical traffic flow sequences, each representing a different data period. A graph-fused traffic flow prediction method based on a multilayer graph attention layer as described in feature 1.

7. A traffic control management system, After the central data processing system obtains future traffic flow prediction results for the target road network using the graph-fused traffic flow prediction method based on a multilayer graph attention layer as described in any one of claims 1 to 6, it transmits them to the traffic control management system. The aforementioned traffic control management system includes, as selectable control function modules, A traffic flow visualization module used in the GUI interface to visualize the historical, current, and future flow conditions of the target road network. A real-time traffic light control module used to adjust the time duration of traffic lights at different intersections in real time using built-in control logic based on the received future traffic flow prediction results for the target road network, and A traffic control management system characterized by including a vehicle flow display early warning module that installs vehicle flow display devices in the target road network and displays the current and future vehicle flow of the road section ahead where the vehicle flow display device is located, so that drivers can grasp the vehicle flow situation on the road ahead in real time.

Images