Research On Deep Learning Based Short-Term Urban Traffic Flow Prediction Models

Transportation system is the foundation and lifeline of urban prosperity and development.In recent years,with the continuous development of urbanization and the increase of urban population in our country,urban traffic congestion has become more and more serious,which has aggravated the occurrence of environmental pollution,traffic accidents and waste of resources,and hindered the healthy development of cities and the travel quality of residents.As an effective way to alleviate traffic congestion,intelligent transportation systems have received more and more attention.Short-term traffic flow forecasting is an important part of ITS,it can help the drivers plan their routes,shorten travel time and improve traffic conditions.Therefore,it has become the key points of researches.In recent years,the rapidly developing deep learning methods have attracted more and more attention.Aiming at the shortcomings of existing models in multi-section and multi-step short-term traffic flow forecasting,this paper first proposes a prediction model based on generative adversarial network,which makes full use of the spatial correlation between multi-section traffic flow and achieves network level short-term traffic flow forecasting.In addition,this paper also studies the problem of multi-step traffic flow forecasting,and proposes a short-term traffic flow forecasting model based on seq2 seq framework.This model solves the time dependence of traffic flow and achieves effective multi-step traffic flow forecasting to reflect variation trend in future traffic conditions.The work done in this paper is following aspects:(1)Aiming at the problem that the existing models can not make full use of the spatial correlation of traffic flow and the blurry prediction of the depth model,proposing a multi-section short-term traffic flow prediction model based on generative adversarial network.The model uses the adapted deformable convolution network to learn the spatial characteristics of traffic flow data,and solves the problem of fuzzy prediction through adversarial learning.(2)Aiming at the problem that existing models can not achieve stable multi-step prediction,proposing a multi-step short-term traffic flow prediction model using seq2 seq structure and attention mechanism based on the framework of generative adversarial network.The model captures the spatial characteristics of traffic flow data using bidirectional random walk graph convolution network,and achieves multi-step prediction on the basis of guaranteeing the prediction accuracy.