Taxi Trajectory And Regional Semantics-based Residents' Travel Flow Prediction Algorithm

With the advancement of informatization and urbanization,people's life have improved,but it have as well brought about challenges like environment pollution,backwards in planning,uneven resource distributions,traffic congestion,and so on.Traffic problems have always been one of the most important factors hindering urban development.The unbalanced spatial and temporal distribution of public transportation and the complexity of the surrounding environment of urban road networks have made it difficult to solve urban traffic problems.With the developing of share economy and mobile Internet applications,a series of new urban data is gradually emerging,such as vehicle trajectory data,crowd mobility,etc.Vehicle trajectory data is a kind of data in urban public transport.It reflects the status of urban traffic,and records the trend of human traffic.Finding potential knowledge from massive trajectory data has become an important research topic for traffic problems.And it is also the important research content of combining traffic and artificial intelligence.This paper uses vehicle trajectory data and regional semantic information which is mainly the POI(Point of Interest)in this paper to extract the spatio-temporal characteristics of massive trajectory data from two angles,the movement of taxi and residents,and then proposes the method that adds the information point to the convolution neural network for prediction.The main innovations of the thesis are as follows:1.Analyze the characteristics of traditional traffic data,and find the characteristics of both taxis and residents from a large number of trajectory data,so as to dig out residents travel patterns and taxi hotspots.Specificly analyze the characteristics of taxi movement;find the unloaded and hovered?the starting and arriving points ?loaded and unloaded areas,and visualize through multiple kinds of traffic information.2.Use crawler technology.We crawl a large number of information points in the study area,and process that through coordinate transformation and pattern classification.We divide the study area into geographical grids.Because the characters of traditional city routes data is insufficient,we add a lot of information point data to that.So we consider the effect not only from the traditional data,but also from the distribution of information point data on the traffic flow.Semantically analyze the point information data,and add that to the traffic flow prediction model.3.The traditional neural network can't well fit the defects of multi-regression problem,so we in this paper transfer the traditional flow data to graph or tensor forms,and propose the convolutional neural network model to predict regional traffic flow.In summary,this paper uses the Chengdu taxi trajectory data and information point data and proposes a convolutional neural network algorithm that superimposes regional semantic.What else,we design and use a deeper convolutional neural network structure that performs better than the BP network and the RBF network do.