Research And Application Of Urban Traffic Flow Prediction Based On Deep Learning

With the continuous improvement of the degree of urbanization in my country,the vehicle inventory in the city is gradually increasing,and the problem of traffic congestion is becoming more and more obvious.The Intelligent Traffic System(ITS)has emerged as the times require.The intelligent transportation system effectively applies advanced information technology to the transportation management system to establish a real-time,accurate,and efficient transportation management system,to achieve the purpose of ensuring traffic safety,improving transportation efficiency and reducing energy consumption.Traffic flow prediction is an important part of the intelligent transportation system.The real-time and accurate flow prediction results are of great significance to traffic signal optimization and travel route planning.Traffic flow data has complex spatiotemporal correlation and nonlinear characteristics,and traditional statistical methods alone cannot achieve sufficient prediction accuracy.This paper improves on the previous research,and the main work is as follows:(1)This paper proposes a combined time series forecasting model based on LSTM and ARIMA,and applies it to traffic flow forecasting.The nonlinear characteristics of the time series are mainly captured by the LSTM model,and the linear characteristics of the time series are captured by the ARIMA model.In addition,this paper adds a skip unit based on LSTM to extract periodic features in traffic flow.This paper is tested on multiple time series datasets and compared with the baseline methods,and the results show that the proposed method has the best predictive performance.(2)This paper proposes a traffic flow prediction method based on graph attention network.Aiming at the non-Euclidean spatial association of road networks and the time series characteristics of traffic flow data,a traffic flow prediction model that integrates graph attention network and gated recurrent unit is designed.The model extracts the correlation between roads at different times through a graph attention network,uses a gated recurrent unit to extract the temporal features of the traffic flow on each road,and fuses the temporal features to predict the traffic flow.Experiments are successfully validated on two traffic datasets.(3)This paper designs a traffic flow prediction system based on deep learning.Based on the above two prediction algorithms,combined with practical application requirements,this paper designs a traffic flow prediction system for ordinary residents,traffic control departments and traffic operation enterprises.The system adopts a distributed architecture,which has strong stability and is easy to expand later.In addition,the system interface is concise,the operation is simple,and it has a good user experience.