High-speed Traffic Flow Prediction Method Based on Multimodal Fusion and Graph Attention Mechanism

Abstract

The invention discloses a high-speed traffic flow prediction method based on multi-modal fusion and graph attention mechanism, which includes multi-modal data of expressways, including road speed data at gantry checkpoints, flow data of toll stations, Road section lane information, etc., to make short-term predictions on the future traffic flow of the target checkpoint road section. The method of the present invention comprehensively considers the time and space characteristics in the traffic flow model, integrates various factors through encoding methods, and constructs a speed prediction model using methods such as comprehensive graph attention mechanism and hole convolution, and applies the measured data of a certain expressway section The model is built and verified. The comparison of the predicted results and the measured results shows that the vehicle speed prediction method proposed in this study has a good effect.

Description

technical field [0001] The invention belongs to the technical field of traffic information, and in particular relates to a high-speed traffic flow prediction method based on multi-modal fusion and graph attention mechanism. Background technique [0002] With the help of traffic big data fusion technology to professionally process massive data, it can realize the refined evaluation of the road network, restore the real operating status, dig out more potential value through in-depth processing, realize the value-added of data and provide a basis for decision-making. Provide support; through historical traffic information, real-time monitoring of traffic information and effective prediction of future traffic information, supplemented by reasonable inducing measures, it is expected to alleviate the problem of low traffic efficiency in the current road network. [0003] At present, the road traffic capacity of expressways usually stays in qualitative evaluation and subjective jud...

AI Technical Summary

Problems solved by technology

[0004] At present, there has been some research on traffic flow forecasting: the literature "Yu, Bing, Haoteng Yin, and Zhanxing Zhu. "Spatio-Temporal Graph Convolutional Networks: A Deep Learning Framework for Traffic Forecasting." Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence (2018): n.pag.Crossref.Web" proposed the STGCN model to predict traffic flow. The model includes two spatio-temporal graph convolution blocks (ST-Conv Block) and an output fully connected layer (Output layer), which is used as The core ST-Conv Block is composed of two time-gated convolutions (GLU) and one spatial convolution (GCN); the model only uses the convolutional layer to design the network structure, and uses the bottleneck strategy (bottleneck strategy) to achieve the pass The scaling and feature compression of the compression channel reduces the parameters of the training process, but the GCN module in the model has limited ability to capture the features of complex topological structures, and GLU is only implemented by adding a gating mechanism to the convolutional layer on the basis of CNN. long-term memory capture

[0005] The document "Zhao, Ling et al. "T-GCN: A Temporal Graph Convolutional Network for Traffic Prediction." IEEE Transactions on Intelligent Transportation Systems (2019): 1–11.Crossref.Web" proposed the T-GCN model for traffic speed For prediction, the model combines graph convolutional network (GCN) and gate recurrent neural network (GRU) to capture the spatial and time-dependent characteristics of urban road network traffic respectively, essentially using GCN to learn complex topological structures, and using GRU to learn Dynamic changes in traffic data; but this model has certain limitations: GCN has limited ability to capture complex topology features, GRU has the risk of gradient disappearance in the long sequence training process, and GRU is prone to error accumulation because of iterative training

[0006] The document "Park, Cheonbok&Lee, Chunggi&Bahng, Hyojin&won, Taeyun&Kim, Kihwan&Jin, Seungmin&Ko, Sungahn&Choo, Jaegul.(2019).STGRAT: A Spatio-Temporal Graph Attention Network for Traffic Forecasting" proposed the STGRAT model to predict traffic speed. Changes, distances between nodes, and circulation relationships, use the space-time attention mechanism to capture the space-time dependence in traffic, and use the spatial sentinel vector (spatial sentinel vector) to avoid the calculation of unnecessary attention values, thereby improving the prediction accuracy; but The prediction performance of the model in the isomorphic graph composed of traffic network is good, but the traffic network in the real scene should be a heterogeneous graph containing a variety of nodes and edge weight information. Too many realistic idealized assumptions make the model The ability to predict is limited

Images