Taxi Travel Demand Forecasting In Pick-up Hotspots Areas Based On GPS Trajectory Data

As one of the main transport in the urban transportation system,taxis not only provide convenient“point to point”service for residents,but also an important supplement to conventional bus transit.On the one hand,along with the rapid development of China's social economy,the scale of the city is expanding and the urban population continues to grow up,which leads to the increasing travel demand for residents.On the other hand,for the blindness of vacant taxis when looking for the passengers and the randonness and flexibility of passengers' travel,there is a contradiction between the residents1 travel demand and the taxi service supply.Based on the GPS trajectory data of Beijing taxis,this paper analyzes the spatial-temporal distribution of residents' travel demand,excavates the taxi pick-up hotspots by the taxi GPS trajectory and predicts the travel demand of residents in the pick-up hotspots areas by considering the multiple characteristics of residents' travel needs.The main tasks of this article are:(1)By the taxi GPS trajectory data preprocessing,the hidden Markov map-matching algorithm is selected to correct the taxi trajectory points which deviate from the actual road network and to enable accurate visualization in ArcGIS tools.On this basis,this paper extracts the pick-up and pick-off track data,the total taxi travel demand and passenger time of taxi for different period of time,besides,visualizes and analyzes the travel demand spatial-temporal distribution and spatial distribution in area of taxi operating with customers of different period of time on working day and rest day.(2)Based on the spatial-temporal distribution research of taxi GPS trajectory data,a combination model of K-means and Systematic clustering is used to cluster the taxi pick-up points and excavate pick-up hotspots areas on working day and rest day,providing support for the analysis of travel demand forecasting model features.(3)Based on the analysis of spatial-temporal distribution of residents' travel demand in the pick-up hotspots areas,the multi-feature travel demand forecasting model is established according to the historical residents' travel demand in the hotspots areas by considering the multiple characteristics of residents' travel needs,and the travel demand forecasting model is realized by means of DNN regression algorithm,which is applied to the pick-up hotspots areas' demand forecasting.Finally,the support vector regression model is established to conduct comparative experiment,and the model evaluation indicators are selected to evaluate the accuracy and applicability of the forecasting model.Analyzing the spatial-temporal distribution of residents' travel demand and forecasting the travel demand of the pick-up hotspots areas can provide data support for the rational taxi dispatching and guide the empty taxis to find passengers in the pick-up hotspots areas.In addition,it will reduce taxi empty-loading ratio and the waiting time for residents,and improve the efficiency and service level of taxis,which,to a certain extent,effectively alleviate the imbalance between residents' travel demand and taxi supply.