Traffic big data restoration method of graph auto-encoder based on self-attention mechanism

Abstract

The invention discloses a traffic big data restoration method for a graph auto-encoder based on a self-attention mechanism, and the method comprises the steps: determining a region where traffic data restoration needs to be carried out, and collecting the historical traffic data of the region; constructing a mask matrix, and generating an adjacent matrix based on the road network structure of the selected area; constructing a data restoration model based on a self-attention mechanism and a graph convolutional network; training the data restoration model; and for traffic data needing to be repaired, obtaining a data repair result by using the trained data repair model. Structures such as an auto-encoder and a multi-head attention mechanism are introduced on the basis of the GCN, a topological graph structure of an urban road network is effectively learned by using the structure of the GCN, the spatial-temporal correlation of traffic flow data is learned by using the multi-head attention mechanism, and complete traffic data is generated by using the auto-encoder according to missing traffic data; and the accuracy of model data restoration can be effectively improved through the multi-head attention mechanism and the GCN.

Description

technical field [0001] The invention belongs to the technical fields of deep learning and traffic data repair, in particular to a traffic big data repair method based on a graph self-encoder based on a self-attention mechanism. Background technique [0002] With the deployment of a large number of sensors, people have collected massive amounts of traffic data from various channels. The collected traffic data often has the attribute of "coexistence of true and false", which is manifested by abnormal phenomena such as missing, error, and redundancy in the data. Therefore, it is necessary for ITS (Intelligent Transportation System) to be able to repair as accurately as possible from the missing data. The lack of existing in the data improves the completeness of the data. However, the traffic data repair problem has its inherent particularity. The main difference between this problem and other data repair problems is that it needs to consider the spatial road network topology a...

AI Technical Summary

Problems solved by technology

Models based on recurrent neural network (RNN) often assume that the relationship between data is sequential. It cannot be processed in parallel and it is difficult to directly model the interdependence between input data with different timestamps. More importantly, They cannot take advantage of the topological structure of the spatial road network, so it is difficult to further improve the repair accuracy

Images