Research On Bus Arrival Time Prediction Method Based On SVR And LSTM

The rapid development of Chinese socialist economy not only accelerates the process of urbanization,but also stimulates the growth of people's travel demand.With the continuous growth of per capital private cars,the problems of automobile exhaust pollution,traffic jams and frequent traffic accidents are becoming more and more serious.In order to solve the above problems,urban public transport has entered the public vision because of its advantages of large capacity,low energy consumption,high efficiency and safety.Therefore,it is an important research direction to manage the urban transportation system through the construction of information platform to promote the benign communication among passengers,vehicles,station facilities and traffic conditions,accelerate the construction of intelligent public transportation system,and support the majority of passengers to reasonably arrange their travel plans.And accurate real-time bus arrival time prediction is one of the important content of the construction of the intelligent public traffic system,and also the important embodiment of the service quality's promotion.Based on the above background,the construction of urban intelligent public transport system in China has achieved rapid development in recent years.In the aspect of bus arrival time prediction,on the one hand,scholars inside and outside have carried out a lot of research,forming a relatively rich theory and time application results.However,from the practical application of the current situation analysis.On the other hand,some cities have tried the bus arrival time forecast on some lines and achieved good social benefits.But,for the overall research and practical application effect,because the bus driving is like the weather,road network structure,site distribution,road traffic flow,passenger flow and vehicle state,driver and passenger driving habits,as well as standing crowded degree,and special events such as interference in a variety of factors,most inaccurate forecast time problems still exist,there are still many problems to be further in-depth study.To solve this problem,based on the bus journey data of a certain city,this paper carried out relevant research on improving the prediction accuracy of bus arrival time.The main research contents and research results include:(1)On the basic of arranging documents of the research status of the bus arrival time prediction,analyzing the process of bus stopping at stations and travelling between stations and their major influencing factors.On the basic of arranging documents of the research status of the bus arrival time prediction,analyzing the process of bus stopping at stations and travelling between stations and their major influencing factors.The bus running time distribution at different time period and same time period in a day and historical same time period was obtained using quantitative method.On the basis,the overall distribution of the bus arrival time is obtained.(2)The paper proposes a general framework based on external characteristics for bus arrival time prediction,and the prediction of bus arrival time is divided into two parts,namely,the prediction of bus dwell time and the prediction of bus travel time.In term of bus dwell time prediction model building,an improved particle swarm optimization(pso)algorithm was proposed for parameter optimization of SVR model based on grid search algorithm and stochastic inertia weight strategy.In terms of bus travel time prediction model building,proposes a LSTM method based on the external characteristics,taking the major influencing factors into the scope of feature selection considerations,building the characteristics of data sets.On that basis,the model was trained and optimized through the function selection and the network structure.(3)The combined prediction model based on the above research results is verified by using the real bus data and combining with several control experiments.The results show the prediction model for bus arrival time prediction performs well on the accuracy of bus arrival time prediction.