Research On Urban Expressway Traffic Flow Forecasting Method Based On Time Convolution Network

With China's urbanization process,while urban transportation facilities are becoming more and more perfect,the number of motor vehicles is rising sharply,and the urban road network is under tremendous pressure,resulting in increasingly serious problems such as urban traffic congestion and environmental pollution.The intelligent traffic system(ITS)has become one of the most effective ways to alleviate traffic congestion and satisfy the wishes of travelers by virtue of its advanced scientific and technological means.Its core function is to realize traffic control and guidance,and real-time and accurate short-term traffic flow prediction is improving ability of traffic control and guidance.By combing the domestic and foreign research status of short-term traffic flow prediction,it is concluded that the existing learning prediction models rarely consider the local prediction effects of different sections,and it is difficult to solve the problem of unstable accuracy caused by violent traffic flow.Therefore,on the basis of studying the overall prediction effect of the road section,this paper further determines that the traffic flow of different sections of the expressway is taken as the research object to better meet the requirements of local prediction accuracy.Firstly,the characteristics of urban expressway traffic flow are analyzed in detail,and the spatiotemporal characteristics of three basic parameters of traffic flow are discussed in detail.On the basis of expounding the technical principles of remote traffic microwave sensor(RTMS)collecting data on expressway traffic flow,the fault data repair model based on statistical correlation analysis provides data support for the short-term traffic flow prediction model established below.Secondly,based on the spatio-temporal characteristics and predictability of traffic flow,on the basis of explaining deep learning theory,a short-term traffic flow prediction model based on time convolution network(TCN)is established.This model combines the advantages of convolutional neural network(CNN)and it avoids the defects of the recurrent neural network(RNN)and avoids the omission of any historical information.Experimental results show that the model can achieve high-precision expressway traffic flow prediction for the entire road segment,and the overall prediction performance is better than the classic prediction model for processing time-series tasks.Finally,for the problem that the prediction accuracy of the model is difficult to maintain due to the change in sample size,on the basis of the above research,the idea of feature engineering and integrated learning is introduced to construct a traffic flow prediction model based on Boosting fusion(ARIMA-TCN-RF).The traditional mathematical algorithm(ARIMA),random forest(RF)model and time convolution network(TCN)model are used as sub-prediction models for integrated learning,and feature extraction and selection of traffic flow data are carried out.The local prediction effect of the fusion model on different sections is emphasized.Experimental results show that compared with other single models,the fusion model not only greatly improves the overall prediction effect,but also achieves high-precision section-level traffic flow prediction.There are 43 pictures,12 tables,and 73 references.