Big Data-driven Method Of Urban Rail Transit Demand Spatiotemporal Distribution Analysis And Forecasting

As the important foundation of urban rail transit planning,demand analyzing and forecasting has been developed for several decades.The aggregate model,like four-step method and the disaggregate model,like activity-based model have been developed.Because of the improved theoretical and practical development,four-step method has been widely used in practice.However,the traditional aggregated model hardly analyzes the impacts of individual attributes and other external factors on the demand.While,the disaggregate model,which is based on individual travel behavior theory,can consider the impacts of individual heterogeneity and many other factors on travel behavior.Meanwhile,urban big data provide the essential data for activity-based model.Therefore,this study is based on multiple data,such as smart card data(SCD),consumption data,and geographical information data.And then,an individual-based model for urban rail transit demand distribution analyzing and forecasting is built.This study aims to improve the application of individual activity-based demand model in practice.First,a model for parsing coupling relationship among multiple data and integrating information is built based on unsupervised algorithm to analyze the relationship among individual activity,individual attributes and activity location profiles.Based on the relationship,land-usage mixture around a station can be inferred using aggregate activity features,and then,individual trip purpose can be inferred utilizing individual activity features and land-usage mixture of activity location.Individual economic attributes are deduced using individual activity features and economic profiles of activity locations.Finally,some essential data of model are acquired,like individual home-work locations,individual economic attributes,trip chains data with trip purpose,and land-usage mixture around the stations.Besides,in order to analyze the relationship between individual travel characteristics and built environment,a reasonable spatial analysis unit scheme should be chosen.So,a spatial analysis zone delineating model is built based on individual travel behavior similarity and modifiable areal unit problem(MAUP).The object function of this model is minimizing the average trip density and internal trip proportion,and it is constrained by multiple delineating principles.At last,a reasonable scheme is delineated and it is utilized to calculate spatial built environment data.Based on these essential data,a systematic model on individual long-term and short-term travel characteristics analysis and daily mobility spatio-temporal analysis in accordance with individual travel decision framework.At first,the relationship between individual long-term travel characteristics like commuting distance,short-term travel characteristics like travel days with urban rail transit,visiting stations and impact factors like individual attributes,built environment is analyzed.For analyzing the comprehensive impacts of these factors,a recommendation system framework driven by the essential data is built to disengage the relationship between individual attributes,built environment around stations and individual travel characteristics,and then it can be used to forecast individual long-term and short-term travel characteristics.This model is of preferably expansibility,and it provides a model foundation for analyzing policy impacts on the long-term individual travel characteristics.Based on the analyzing results of long-term and short-term travel characteristic,the passengers plan their daily travel features from spatio-temporal dimension.This study clarifies individual mobility mechanism from temporal and spatial dimension respectively based on SCD.For temporal dimension,a classic Markov model is used to analyze individual travel behavior between two days,and a non-homogeneous semi-Markov process is used to describe activity status transition of different time and activity duration distribution.For spatial dimension,an improved Exploration and Preferential Return(EPR)model is used to analyze station choosing mechanism.Meanwhile,station built environment and cluster preference are considered in the model to strengthen its expansibility.Using this model framework,we can analyze and forecast passengers’travel behavior from individual perspective for new passengers and new stations;and finally,the aggregated demand distribution results can be calculated.The models in this framework are all implemented using big data of Beijing to test the validity of models,respectively.The predictive ability for new passengers and new stations are tested respectively.The results show that the temporal demand distribution prediction results for different times of day coincide well with the real data.And for spatial dimension,relative predictions error of station ridership range from 0.25 to 0.38,and model performance can be improved significantly when individual economic attributes are introduced in the model.For OD ridership of weekdays,the absolute prediction error is 30.7 trips,which is better than existing studies.The model in this study performs strong robustness with stable error for multiple simulation.