Real-time Prediction Of Hot Spots Based On Trajectory And POI Data

Nowadays,smart mobile devices are usually equipped with location sensors using global locating technology,and are capable of precisely capturing location information of the mobile device.A great amount of data is generated with the advance in technologies of location acquiring and wireless communication.The data exhibits trajectory logs of different types of moving objects such as human beings,animals and vehicles.And the location information can be transmitted to servers by wireless communication through crowd sourcing,and finally forms volumes of GPS trajectory data.GPS trajectory data contains unprecedented information,which benefits understanding mobile objects and geographical location,and leads to varieties of applications such as location based social network,intelligent transportation system and urban computing.The prevailance of these applications continuously promotes novel systematic study on trajectory big data in turn.In the healthy development loop,mining on GPS trajectory big data becomes a hot research point,attracting unremitting studying of researchers from fields of computer science,sociology and geography information.However,mining semantic information of human activity purely from trajectory data is difficult.From a more abstract perspective,trajectory data is a type of spatio-temporal data,which features spatial attribute and temporal attribute and is tagged with spatial coordinates and timestamp.Accurately predicting the evolvement of spatio-temporal data is of great importance to geographical location recommendation and other related urban computing applications.On the other hand,POI(Point of Interests)of a city and other static geographical information can affect human activity significantly if not essentially,for people live in a geographical environment,and POI is always where people attend their daily social activities.This paper tries to fuse both trajectory spatio-temporal data and geographical static information data,and to predict development of the number of stay point belonging to different regions.Stay point is also one type of spatio-temporal data,referring to the location where taxies wait for a passenger statically or hanging around in low velocity.Prediction for stay points is important for taxies awaiting passengers,passengers taking taxi,interest level of a geographical location,and even traffic dispersion,city security and city construction planning and some other aspects.Prediction for stay points is usually a part of geographical location recommendation,however,in current studies on hot region recommendation and taxi’s pick-up point recommendation some defects occur.One is the lack of abundant data sources,which may causes low accuracy,the other is that mining directly on historical trajectory data brings high computation complexity,which limits extension to larger geographical range and is difficult to have real-time prediction guaranteed.In this paper,we fuse both trajectory data and POI data,and propose a novel prediction framework,which finds similar region according to spatial similarity based on POIs,and then predict the number of stay points of next time interval on target region by its similar region.A real-time processing system based on Apache Storm is also built to simulate the whole process.The experiment results show that the framework has high prediction accuracy and that the stream processing system is capable of processing big data in real time manner with low latency and high throughput.