Research On Data Mining Based On Hadoop For Massive Trajectories

Nowadays,“Big Data”is growing faster and faster,and it solve many problems and take conveniences into our lives.As an important branch of Data Mining,The mining of trajectories are gaining more and more attentions benefiting from the mature mining technologies and the easy access to the source data,most importantly,the trajectories is rich with potential information of the moving objects.At other hand,many big cities in developing country are facing a more and more serious traffic problem,how to mining the trajectories so as to solve or release the traffic problem became a significant subject.Here in the paper,we take the massive taxi technologies to construct a recommendation model for passengers and a recommendation model for taxi drivers,while recommending a good strategy to passengers to find an available taxi quickly and recommending a good strategy to taxi drivers to find a passengers as soon as passable.The main works of our research lie in:(1)A storage structure of road networks is designed to satisfy the demand of quick searching of road networks,which it utilize both linked lists and arrays.Additionally,,a map matching algorithm is improved and optimized in order to speed up the map matching work,it use a dynamic boundary instead of a static one to narrow the searching area so as to lift the efficiency of map matching,on the base of this algorithm,a map matching algorithm based on SVM is proposed,which take the input includes the speed of the vehicle,the degree of the speed,the euclidean distance between the vehicle and the road and the last state of the matching and forecast the current positions,and it is realized using MapReduce in Hadoop;at the meanwhile,an extended grid hybrid clustering algorithm(EGHC)is proposed to extract the parking places which is based on OPTICS,and it is realized using MapReduce.Experiments are implemented to verify the efficiency and accuracy of Hadoop on deal with massive trajectories data,the map matching algorithms and EGHC.(2)To release the problem of “Hard to take a taxi for passengers in the big cities”,a recommendation model for passengers is present which push the waiting time for a free taxi and some of the nearest parking places around to passengers to guide their decisions.We dig into the trajectories data and give a meticulous analysis,and it reveals that people by the road have more taxi demands than the people at the parking places,and this is what the recommendation model focuses on.In the recommendation model,we use a method named two section fitting to fit the curve of the arrival available taxi on a road,in consideration of influence from the weathers,TMI(Travel Meteorological Index)is included into model to quantize people's feelings to the weathers so as to promote the applicability and generality of the model.At last,two experiments is designed to testify the accuracy of the recommendation model and the high efficiency of Hadoop,respectively.(3)To release the problem of “Hard to find passengers for taxi drivers in some case”,we construct a recommendation model for taxi drivers,and it consist of two parts: offline and online,the offline part takes the job of calculate the parking place,the demand curve of the parking place and so on,while the online part is responsible to provide a real-time recommendation service by calculating the minimum expecting waiting time;In order to adjust the model to the weather,human body comfort meteorological index(BCMI)is included to measure the weather conditions.To the effectiveness and the accuracy of the recommendation model and clustering algorithm,experiments are implemented in the field in different conditions.