Research On Urban Road Travel Time Prediction Model Based On Gaussian Process Regression

The intelligent transportation system(ITS),which plays an important role in urban traffic dispersion,road monitoring,vehicle management and intelligence analysis of traffic flow,is an intelligent solution to solving the traffic congestion,improving traffic efficiency,reducing accidents and environmental pollution.The key to construction of ITS is to study how to change from collecting and distributing the traffic information passively,to analyzing,predicting,managing and servicing initiatively.Vehicle travel time prediction is one of the important issues in traffic flow prediction.At present,the theory of traffic flow prediction is mainly about the road travel time of the urban overall vehicle from the perspective of urban traffic and traffic flow,which plays a positive role in urban planning and development and construction of road network.However,these theories are seldom studied the issues of vehicle travel time prediction from the point of view of regular travel,which is difficult to provide personalized short-term travel time prediction service for convenience of people’s travel.This paper studies the prediction of short-term travel time in urban road network based on the characteristics of regular travel.This paper including:This paper presents an algorithm for short-term travel time prediction based on Gauss Process Regression(GPR).In terms of high noise and nonlinear,time-varying of traffic data,the travel time prediction model based on GPR is established,with the use of good adaptability of GPR which can deal with complex problems like small sample,high dimension and nonlinear problem.The model separates the working days and rest days depending on the regular travel of urban residents.It shows high accuracy compared with the support vector regression model and autoregressive moving average model.The experimental results indicate that the prediction accuracy of GPR algorithm is improved by 13%-35%compared with other two algorithms.An algorithm for vehicle short-term travel time prediction of GPR based on AdaBoost is presented.The prediction accuracy of GPR in stationary period is very high,but the prediction results fluctuate greatly in the peak period,for GPR in the process of obtaining the hyper parameters may still fall into the local optimal solution,and its prediction effect is directly dependent on the choice of kernel function.Therefore,a strong predicting machine is got by means of AdaBoost integrated with GPR weak classifier,which improves learning ability and generalization ability of GPR.The experimental results show that the accuracy of GPR algorithm based on AdaBoost the improved 34%compared with GPR algorithm in rush hours.