Research On Short-Term Traffic Congestion State Prediction Model Based On Deep Learning

A series of phenomena,such as the increase of urban population,the increase of the number of motor vehicles,the diversified development of travel modes,and the increase of vehicle exhaust pollution,appear more and more in today’s increasingly developing cities.Urban traffic congestion is the most direct problem brought by these phenomena.Traffic congestion poses a serious threat to the economic development of cities.To predict the traffic state,it is necessary to identify and judge the traffic state.Firstly,this thesis studies traffic status identification and compares several common indicators,including INRIX,RCI,DC and so on.However,all of them have their own shortcomings,such as lack of regional information,lack of timeliness,and too complicated calculation.The TTI congestion coefficient solves these problems effectively.Therefore,TTI congestion coefficient can be used to judge the traffic state.In order to quickly calculate the TTI congestion coefficient through massive GPS data,this thesis builds and uses a distributed data server cluster based on Hadoop.Then,the parallelogram method is used to process the matching work of GPS data and road network data,and then the TTI coefficient of the city region is obtained,and the congestion discrimination of the city region is made.Secondly,on the traffic congestion forecast,compared with the traditional traffic forecast model,this article does not adopt the traditional prediction methods based on vehicle speed,but use,is TTI congestion coefficient,the coefficient is calculated based on the road of their own free flow speed,can let the streets are unified with different speed level to TTI coefficient as evaluation of congestion,Therefore,compared with the prediction method based on the traditional vehicle speed,it is more accurate.Finally,this thesis compared the traditional neural network prediction model with the CBILSTM prediction model,and verified that the CNN layer in C-BILSTM can indeed extract the spatial feature information of the TTI coefficient,but the prediction accuracy was still not high.Therefore,a CS-BILSTM network prediction framework was proposed after improvement.Softmax layer was added between the CNN processing spatial information layer and the LSTM processing time information layer to strengthen the spatial feature information extracted by the CNN layer and thus improve the accuracy of the prediction model.