Traffic Flow Prediction Research Based On Deep Learning And Spatiotemporal Big Data

With the advent of the era of big data,the scale of cities continues to expand,and a large number of sensors all over the road generate traffic flow data at all times,posing huge challenges to the traffic management system.In addition,the intelligent transportation system(ITS)The emergence also puts forward higher requirements for the problem of traffic flow prediction.Traffic flow data is a kind of classic spatiotemporal data,which not only has time dependence,but also has spatial correlation.Only considering unilateral characteristics will cause problems such as low prediction accuracy.Secondly,the topology of the traffic network changes dynamically,and the traditional convolutional neural network that extracts the features of graph structure data cannot capture such dynamic features.Furthermore,traffic data has long-term dependencies,and the lack of temporal feature capture can easily lead to problems such as poor data prediction effectiveness.In order to improve the accuracy and assist the intelligent transportation system to make accurate decisions quickly,this paper proposes a Spatial-Temporal Network based on LSTM and GAT based on deep learning and spatiotemporal big data,to predict the traffic speed in the traffic road network.The dynamic topology in the traffic network is obtained through GAT,and the LSTM processes the temporal characteristics of the data and captures the periodic characteristics of the traffic flow.In order to capture more temporal dependencies,an attention mechanism is introduced on the basis of STLGAT,and proposed a Spatial-Temporal Network based on LSTM and GAT,introducing Multi-Head Attention,enhance the extraction of long-term dependencies on traffic flow data,process information in parallel,and achieve efficient and accurate prediction.The work of this paper as follows:1.Aiming at the low accuracy of the traditional model that only considers the temporal correlation in the traffic network,a traffic forecasting model that considers both the temporal dependence and the spatial correlation is proposed.The model is mainly divided into spatial modules and The time module can accurately capture the characteristics of traffic data.2.Considering that the topology of the traffic network changes dynamically,this paper uses the graph attention network to apply different weights,reflecting the influence of neighboring nodes on the central node,and successfully capturing the topology in traffic data..3.Due to the strong temporal dependence of traffic flow data,longer-term temporal features should be extracted as much as possible when building models.This paper proposes the MA-STLGAT model,introduces a multi-head attention mechanism to focus on important information,learns the contribution rate of nodes,and further solves the problem of capturing temporal correlations.4.Two models in this paper are compared on the real data set Pe MSD7,and compared with the prediction of the baseline models.The results showed that the errors of the two models are reduced,especially the MA-STLGAT model in the medium and long term.In terms of the prediction of the model,the decline in the prediction error of the model has increased significantly,indicating that MA-STLGAT performs well in long-term prediction,reflecting that the introduction of attention mechanism can indeed effectively obtain more "long-term".