| Public transportation is an important infrastructure to speed up urban construction.The development of intelligent public transport system can effectively alleviate the current congestion of urban traffic status.The bus arrival time is one of the most important traffic information of travelers.It can improve the accuracy of predicting bus arrival time,and improve the attractiveness of public transport.GPS data records the movement trajectory and status information of the bus at every moment.The data are massive,high-dimensional and valuable.It’s a challenge to the traditional model.How to process big data quickly and extract effective information has become a research hotspot.Hadoop distributed framework makes the data mining algorithm be migrated to the distributed platform.The high reliability and extensibility of MapReduce framework make the mining algorithm can deal with massive data,which greatly improves the complexity and efficiency of the algorithm.According to the above problems,this paper puts forward a prediction model of bus arrival time based on the combination of MapReduce and neural network.This paper includes three parts: 1.The data is the GPS positioning data of the bus,data preprocessing is needed to process the missing data,noise data and abnormal data and calculate the basic data set needed for generation partitioning and prediction.Based on the analysis of the running time of buses and the influencing factors of bus arrival time,this paper chooses the characteristics of station running time,station stay time,week and holiday as the input eigenvector of bus arrival time prediction model.2.Based on the analysis of the influencing factors of bus arrival time,it is found that the time-interval is the main factor affecting the total running time of the bus.Therefore,in order to improve the accuracy of bus arrival time forecasting,this paper puts forward a segmentation prediction model combining clustering and neural network.Combined with the running characteristics of bus,K-means clustering method is used to classify the time-interval of the bus.The bus running data in the same period has high similarity.Then,BP neural network is established for each period to predict the bus arrival time.3.In order to comprehensively analyze the GPS data,to mine more accurate and valuable information,and to deal with huge amounts of data,this paper presents a time-based forecasting model of parallel bus arrival based on MapReduce.The parallelization framework of K-means algorithm and BP algorithm is established on Hadoop platform.Compared with the traditional serial method,the prediction speed of the model is effectively improved. |
