Traffic Flow Analysis And Prediction Based On Res2Net And Gated Circulation Unit

With the construction and development of today’s cities,the transportation problem has developed into a real problem that cannot be ignored.And the effective prediction of the future passenger flow of traffic can solve a series of problems including serious contradiction between road supply and demand,urban traffic road congestion,and frequent accidents.The role of traffic flow forecasting in the development of urban transportation is crucial.In recent years,it has become a research hotspot of experts and scholars in the field of transportation.However,due to the influence of time dependence,spatial dependence,city topology information,various events and weather conditions,and the high nonlinearity and complexity of traffic flow data,it is difficult to accurately predict the traffic flow data.The traffic forecasting model does not have accuracy and reliability.Therefore,accurately predicting traffic flow becomes a difficult task.Based on the massive data of urban traffic,this thesis first analyzes the correlation between the traffic data after sampling and preprocessing,and analyzes that the traffic flow has the characteristics of time and space dependence.The LVW(Las Vegas Wrapper)feature selection algorithm is used to select the traffic flow with the best input time delay and the amount of road space for the task of traffic flow prediction.In traffic flow prediction,this thesis proposes a deep learning model(Res2GRU)that combines a improved residual network(Res2Net)and a temporal gated recurrent neural network(GRU)to extract features from urban traffic flows,where Res2 Net is represented by a more fine-grained The characteristics of multi-scale features extract the deep abstract spatial correlation of traffic flow,and GRU is used to learn the temporal correlation of urban road traffic flow.In addition,because the local proximity operation of the convolution operation in the model ignores the problem of global information capture,this paper optimizes and enhances the convolution operator by means of the fusion of the self-attention mechanism.The model was experimentally verified on the Shenzhen taxi speed data set,and different models were used to predict the future traffic flow of 15 minutes,30 minutes,and 60 minutes,respectively.Experimental results show that the model has more accurate prediction capabilities.This thesis uses the Res2 GRU prediction model proposed by the integrated idea to solve the problem that most existing models only excavate the features of the traffic flow at the time level,while ignoring the feature extraction of the traffic space.In addition,the traffic flow data that is highly relevant to the future is selected through feature selection,and these traffic data are divided into three time period components of proximity,period and trend.This approach can fully excavate the proximity,period and Trending characteristics.Under such data processing and model training,the effect of traffic velocity prediction has been effectively improved...