Research On Short Term Taxi Region Distribution Predict Based-on Trajectory Data Mining

Taxi is one of the most important parts of urban public transportation system,which provides a convenient and fast way to travel.In recent years,with the development of mobile Internet,net car service platform like Di Di rise,on the one hand,provides a basis for the selection of a more efficient and convenient way to travel to the public,on the other hand the traditional taxi operation mode caused a certain degree of impact,the traditional taxi industry faces greater competitive pressure.Taxi operators urgently need to improve operational efficiency and enhance competitiveness,to compete with the impact of the net car.The taxi dispatch load is effective to reduce taxi empty rate,the key to enhance the operational efficiency,while the empty taxi scheduling based on reasonable for a period of time in the future in different areas of taxi demand and for some time in the future's taxi distribution in different regions in the forecast of taxi demand is far greater than taxi supply area is the destination of the taxi dispatch,will appear in the supply area for no-load taxi dispatch to the future for smaller than for the area in the future,so as to reach the balance of supply and demand state of no-load taxi taxi scheduling goal.This paper focuses on the forecast of regional distribution of taxi in the future.Unlike some random taxi request with the next period of time the taxi distribution in different regions and the current taxi distribution in each region are highly correlated,so to predict the regional distribution of taxi mainly use the regional distribution information of the current taxi.Through historical taxi trajectory data,this paper proposes a short taxi regional distribution of different types of prediction algorithm,and verified by the experiment compared the prediction effect of each algorithm,three algorithms are the prediction algorithm of Markov process belongs to probability and statistics,which belongs to the matrix decomposition algorithm for supervised learning,unsupervised learning belongs to the prediction of GBRT algorithm.The Markov process is a stochastic process,the most important is the nature of Markov prediction problem in no aftereffect,short taxi area distribution,prediction algorithm based on Markov process by real time taxi regional distribution abstract vector form,and then describe a day in this period by taxi transtion-probablity matrix transfer area matrix multiplication between the various regions,so as to obtain a period of time after the taxi distribution in each region prediction.Matrix decomposition is a key step in latent semantic model algorithm,mainly used in the field of recommender system,this paper will introduce the taxi distribution matrix decomposition algorithm in prediction problems,and the spatial and temporal characteristics of regional distribution of taxi based on matrix decomposition algorithm on the basis of the application of transformation,so that it can effectively solve the problem.GBRT algorithm is a typical supervised machine learning algorithm has good generalization ability,performance prediction precision,this paper through for each regional training alone is a regression method to predict the number of taxis within each region will appear in future.In order to mass trajectory data of taxi operators has accumulated over the years the effective utilization of mining technology to preprocess the original data using the trajectory trajectory data in this paper,extract the related information from the temporal and spatial distribution of taxi trajectory data in the original,and the organization is in the form of Tensor.In the subsequent algorithm training phase,the relevant information stored in Tensor is converted into the form of training and simulation prediction.