Bus Arrival Time Prediction Based On Static And Dynamic Algorithms

With the rapid development of urban modernization,people gradually improve requirements for the quality of life.It is obviously to notice that the number of cars have risen sharply especially in metropolis.How to solve the limited road resources and to enhance the public image of the traffic has become an urgent problem that need to been study.To address these questions,our study suggests the ways may ease traffic pressure in the public transport system,such as increasing accuracy of forecast for the bus arrival moments,and reducing the ownership of private cars or traffic control likewise,etc.First of all,this paper introduces the forecast research background and related research status of the bus arrival time prediction,and introduces some models that already exist and used for forecasting.Secondly,we analyze the possible factors which may affect the bus arrival time and puzzle GPS devices.Therefore,we report the process of preprocessing the original data and how to correct the error data.Afterwards,we described the static algorithm combined with dynamic algorithm in detail,including the static algorithm for the support vector machine(SVM)algorithm and commonly used kernel function and the parameters of the kernel function.After we introduce the basic principle of the dynamic modified Kalman filter,and five basic state update recursive equation continuously.According to the last state and the current state variables of the input to predict the next state.Due to the Kalman filter uses only the last state of multi-innovation information,therefore,in consideration of the improvement of Kalman filtering algorithm based on multi-innovation theory,that‘s not just rely on the last state but more than one useful data in the past.Then,in order to verify the effectiveness of the proposed algorithm which are proposed in this thesis,we empirically choose the historical data of Shaoxin BRT line 1 in 2013 as our research samples which is mainly consist of four different parts.After our preprocessing including 4 input vectors,the peak or not peak,weekday or weekend,holidays or not holiday,considering weather factors.Then dynamically modify the results which are supposed to expose the traffic status such as bus arrival time.Our forecast is experimented by using static SVM regression results,the SVM regression combined with standard Kalman Filter forecast results,the SVM regression combined with multi-innovation Kalman Filter forecast.The results output from the experiment in this thesis showed our result present the lowest error rate.At last,we made some summary about our work,contributions,deficiencies,and prospects as well.