Traffic prediction method based on adaptive spatial self-attention map convolution

Abstract

The invention discloses a traffic prediction method based on adaptive spatial self-attention map convolution, belongs to the traffic field and the deep learning field, and provides an adaptive spatial self-attention graph convolution network (ASSAGCN) for traffic prediction. The ASSAGCN is formed by stacking two residual error blocks. Each residual block is composed of a graph convolution module (GCN), a multi-head spatial self-attention module (MHSSA), a gating fusion module (GF) and a multi-receptive-field cavity causal convolution module (MRDCC), Wherein the GCN performs modeling on local spatial correlation of a road network based on connectivity; the MHSSA is used for capturing implicit spatial correlation of a road network and aggregating information of each node globally; the GF fuses the output of the GCN and the output of the MHSSA; and the MRDCC is used to model temporal correlation. An input layer adopts a simple full-connection layer to map input to a high-dimensional space to improve the expression ability of the model, and an output layer adopts two 1 * 1 convolutional layers. The method can capture the potential spatial correlation in the road network, and adapts to the dynamic change of the road network structure.

Description

Technical field [0001] The invention belongs to the field of transportation and deep learning, and specifically relates to traffic condition prediction. Background technique [0002] With the rapid development of intelligent transportation systems, urban traffic prediction has attracted more and more attention. Accurate and timely traffic forecasts can help travelers plan their travel routes reasonably, alleviate traffic congestion, and improve traffic operation efficiency. They are of great significance to urban traffic planning, traffic management, and traffic control. However, since traffic data exhibit complex spatiotemporal correlations, the traffic prediction problem has always been a challenging research topic in the transportation field. [0003] As a typical spatiotemporal prediction problem, traffic prediction has been studied for decades. Early traffic prediction methods were mainly based on statistical models or simple machine learning models. The most represe...

AI Technical Summary

Problems solved by technology

However, the traffic forecasting problem has been a challenging research topic in the field of transportation due to the complex spatio-temporal correlations exhibited by traffic data.

First of all, the topological structure of the road network may be affected by certain factors and change (for example, a marathon is held in a certain place, or the road is icy, or a traffic accident on a certain road section causes these road sections to be temporarily blocked), and the existing methods no matter Neither the predefined adjacency matrix nor the learnable adjacency matrix can capture the dynamic changes of the road network structure.

Second, since graph convolution can only stack limited layers, it can only aggregate neighbor node information within a limited range, but cannot model global spatial correlation

For large-scale graphs, only using graph convolution is not ideal

Images