The Prediction Method Of Bus Fleet Size Based On Relevance Vector Machine

With the rapid development of the city,the problem of urban traffic becomes more and more serious,which compels people to give priority to the development of urban public transport.At the same time,the domestic bus timetable still remains in the manual preparation stage,which is cumbersome and complex,needs a lot of time and is difficult to adjust quickly.In addition,in the beginning of the new bus line planning,the bus company needs to obtain the bus fleet size,which is generally calculated based on the hand-compiled timetable.It's inefficient.In view of the shortcomings of the traditional bus fleet size calculation method,this paper summarizes the feature vector of single-line planning and proposes a method based on Relevance Vector Machine(RVM)to predict the bus fleet size,which is under the premise of the absence of public transport timetable.Compared with classical prediction methods,such as Support Vector Regression(SVR),this method has the advantages of more sparse model and shorter prediction time,and is suitable for small sample size problem.The main contents of this paper are as follows:(1)The existing bus line lacks the complete planning parameters.Based on the characteristics of single-line planning and the bus timetable,this paper extracts the relevant planning parameters and summarizes the feature vector of the line planning.(2)Based on the above-mentioned feature vectors,the single-kernel RVM model is used to predict the bus fleet size.Genetic algorithm(GA)is used to optimize the kernel parameters of the single-kernel RVM.(3)As the single-kernel RVM method of data sensitivity is high,generalization ability and robustness is poor.In this paper,multi-kernel RVM model is proposed based on multi-kernel hybrid idea.This model predicts the fleet size by integrates multiple single-kernel RVM into a combination of multiple linear regression,and uses GA to optimize the linear combination coefficient.Experimental results show that this method has the advantages of higher prediction precision and stronger generalization ability compared with single-kernel RVM method.(4)Since the bus line is gradually planned over time,it is an incremental process.Therefore,this paper builds an online learning model based on Tipping's fast marginal-likelihood algorithm and applies it to the prediction of bus fleet size.(5)In order to compare with the above RVM prediction method,this paper designs a heuristic algorithm for timetable preparation,and then calculates the required fleet size based on the deficit function.In addition,a Web application system based on the SSH framework is developed,and the algorithm of the timetable preparation is integrated into the system.The results show that the algorithm is complex and needs to be re-optimized manually,although its calculation is accurate,it is far less time-dependent than the RVM-based prediction method.This paper aims at the line planning of Xiamen public transportation,and obtains the planning parameters and the timetable of the bus line from the Xiamen City Wireless Public Transport System,calculates and extracts the relevant characteristics,and carries on the experimental analysis to the proposed method.The experimental results show that the method proposed in this paper can predict the size of bus fleet more accurately.