A Research Of Coach Bus Arrival Time Prediction Using Multi-source Data

Coach bus transportation,as an important way to travel,is the basic industry of socio-economic development.Promoting the informatization construction of coach transportation and building a sound intelligent transport system are effective ways to improve public transportation service level.The arrival time of the vehicle is one of the most concerned information.Coach bus arrival time prediction is not only conducive to passengers' reasonable travel arrangements and vehicles' real-time scheduling,but also to achieve a reasonable allocation of social resources,ease city traffic congestion,and energy saving.With the popularity of vehicle GPS,maturation of GIS technology,the rise of "Internet +Traffic" concept,it makes great significance to use new technology to predict coach bus arrival time.Firstly,this paper designed a lossless method to clean data,which is based on azkaban+Hadoop+nagios.It is used to clean many types of GPS data including coach bus,bus,taxi and lorry.Azkaban is used to schedule cleaning tasks,then the qualified and unqualified data stored in HDFS isolated rather than abandoning noisy data directly as traditional way.Nagios is used to monitor data's cleaning quality.When cleaning quality is of substandard quality,it will initiate an alarm automatically,and send a message to the relevant person in charge until the problem is solved before lifting the alarm.Secondly,after getting the impact of factors including type of road,long holidays and other factors on the coach buses' long running,we discuss the running characteristics of bus and coach bus,then support vector regression is used to predict coach bus arrival time.The paper gives full consideration to model's characteristic factors,at last,factors are chosen including the type of road,long holidays,weather,traffic,distance,time,scheduling information.In this way,we can avoid the occurred problem in bus field,such as excessive reliance on experience,too less factors,and unreasonable model factors.Genetic algorithm is selected to achieve the best parameters of support vector regression machine,experimental results show that the genetic algorithm can save about 50% of the time compared to the traditional K-CV algorithm.Finally,to get the required model characteristic factors,we use Hadoop platform to do abnormal GPS records processing,determine departure time,and use storm platform to do map matching,one dimensional transformation of latitude and longitude,traffic calculation.In experiments,training data is a total of 300 sets of coach bus GPS data,test data is a total of 124 sets,results show the coach bus arrival time prediction model proposed in this paper is useful.Based on the research contents of this paper,a coach bus arrival time prediction software system can be achieved,which has entered the development stage of 6 months.It will produce practical application value,and serve the people.