Estimation And Prediction For Taxi Availability Based On Float Car GPS Data

Urban population,urban traffic demand and the diversity of traffic demand increase continuously with the development of our city and thus raise higher requirement on the urban transportation system as well as the urban taxi system.In order to relieve the taxi hailing difficulty problem in our city,reduce the noload rate and increase the convenience of public trip.Considering both the taxi demand and supply,this dissertation used the taxi GPS data ranged from 2016.8.1 to 2016.8.28 building a model to evaluate the taxi availability based on maximum likelihood estimation method.Based on the result of estimation,this dissertation made a spatial distribution analysis,a hotspot distribution analysis and temporal feature analysis on Beijing taxi availability.It showed that there is an obvious difference between weekday and weekend taxi availability.The taxi availability is poorer and the range of taxi availability is larger on weekday.It distributes widely in our city and more continuously.The worst period of a day is on morning peak.Noon peak,afternoon,evening peak and evening time are also poor in taxi availability.Compared to weekday,taxi availability on weekend is better and its distribution is more scattered.Overall,poor taxi availability region distributes on the main city area among the fourth ring road owing to its high population density and traffic demand.Then,according to the taxi availability and building property data,different areas have various temporal features.For example,housing region is typical region influenced by on duty commute demand.Commercial region is typical region influenced by off duty commute and work travel.And transportation hub is highly influenced by travelers' arriving activity.According to the result and conclusion of analysis,this dissertation used the first three week taxi availability data,building property data,population data,transportation station data and precipitation data as train dataset and use the last week taxi availability data as validation dataset to train a predict model.This dissertation used ARIMA,Linear Regression and Multi-Layer Perceptron model respectively to make a short time prediction and used RMSE and MAPE to evaluate the performance of different model.According to the result,the accuracy of model reduces with predicting interval and among the three model,the Multi-Layer Perceptron model outperformed the others.According to the analysis result and prediction result,taxi driver and passenger can know the taxi hailing difficulty region in advance to drive to/avoid to carry passenger/hail a taxi respectively at a “difficult region” and therefore alleviate the taxi hailing difficulty problem.