Research On Taxi Travel Demand Forecasting Under The Influence Of Multiple Factor

As a fast and highly mobile mode of travel,the taxi can improve people’s commuting efficiency to a certain extent.However,problems such as difficulty in taking a taxi,waiting for a long time,and an imbalance between the supply and demand of taxis still restrict the development of the taxi industry.To shorten the waiting time and optimize the allocation of travel resources,many researchers have carried out studies of taxi travel demand forecasting.However,taxi travel demand forecasting is not a simple task.On the one hand,high-frequency noise and complex nonlinear patterns exist in taxi travel demand data due to multi-source influencing factors.Although current researches use a variety of algorithms to forecast taxi travel demand,there is little work that comprehensively considers the multisource factors affecting taxi travel demand.This thesis investigates and analyzes multisource influencing factors,and proposes a framework for pre-processing multi-source influencing factor data,which in turn assists taxi travel demand forecasting models to improve accuracy.On the other hand,the study object of the taxi travel demand forecasting problem is often area-based.A large area is usually divided into multiple small areas for study.Modeling the long-term temporal dependence features and nonlinear fluctuation patterns of demand data for each small area is not simple.To solve this problem,a taxi travel demand forecasting model is proposed for a local single area.The small areas have spatial interaction with each other.However,this interaction is not obvious.The nonlinear fluctuation patterns of the demand data of each small area are also difficult to be fitted.Therefore,a taxi travel demand forecasting model is proposed for global interactive areas.The main work of this full thesis is as follows.(1)Influenced by multi-source factors,taxi travel demand has noisy data,complex nonlinear fluctuation patterns,and many local extreme values that are difficult to be forecasted.This thesis investigates and analyzes multi-source influencing factors and proposes a framework for pre-processing multi-source influencing factors.The framework calculates and ranks the importance of different influencing factors by the Random Forest,verifies possible multicollinearity among multi-source influencing factors by the Pearson correlation coefficient,and eliminates multicollinearity by the Principal Component Analysis.This framework is proven to be able to exclude invalid information in multisource influencing factor data,which in turn assists in taxi travel demand forecasting.(2)To address the problem of taxi travel demand forecasting in small localized areas,a taxi travel demand forecasting model is proposed for a local single area.The Multi-source Influencing Factor Component in this model can fuse the pre-processed results of multisource influencing factor data with taxi travel demand data;the Temporal Feature Extraction Component can explore the long-term temporal dependence features of taxi travel demand data;the Attention Pooling Mechanism can improve the attention weight of the model to the influencing factor data,and then increase the forecasting accuracy of the model.(3)To address the problem of taxi travel demand forecasting in the global interactive areas,a taxi travel demand forecasting model is proposed for the global interactive areas.The Multi-source Influencing Factor Component in this model can highlight the importance of influencing factor data;the Spatial-temporal Feature Extraction Component can capture the short-term,medium-term,and long-term spatial-temporal variation patterns of taxi travel demand;the Multi-source Information Fusion Component can fuse the pre-processed results of multi-source influencing factor data with the spatial-temporal information of taxi travel demand,and then improve the forecasting accuracy of the proposed model.