Bus Travel Time Prediction Based On Deep Belief Network With Back Propagation

With the rapid development of the social economy and the rapid increase in the number of motor vehicles,the contradiction between supply and demand in the urban transportation system has become more and more serious,so that the construction measures of urban roads are far from meeting the growing travel demand,and various Traffic problems.The application of intelligent transportation system plays a key role in the dynamic traffic management of urban roads.In the intelligent transportation system,accurate public transportation information is crucial for passengers to arrange departure time and reasonable route selection.Therefore,the development of a high-precision bus travel time prediction model has received more and more attention,Therefore,scientifie and accurate bus travel time prediction has academie value and practical significance.Aiming at the current problems in this research field,this paper proposes a bus travel time prediction model based on the concept of deep belief network(DBN).In general,deep confidence networks have good functional learning capabilities.However,the classic deep confidence network was developed to learn features from binary data,but traffic data is continuous.Therefore,this paper uses an RBM variant in the deep confidence network that can process continuous data,called Gaussian-Bemoulli RBM(GBRBM).First,in this model,several Gaussian Bernoulli Restricted Boltzmann machines(GBRBM)are used to learn features in an unsupervised manner.A back propagation(BP)neural network is then introduced to predict travel time in supervised learning.In order to accurately describe the existing traffic conditions,several variables are considered in the model,including bus running time and dwell time(passenger load).Finally,based on the actual traffic data of Shenyang,China,several experiments were carried out to verify the technology.The results show that the model is more accurate than the traditional methods,including k nearest neighbor algorithm(k.NN),artificial neural network(ANN),support vector machine(SVM)and random forest(RFs).