Travel Demand Forecasting: An Evolutionary Learning Approach

摘要

In disaggregate travel demand modeling, we often create an artificial population from a sample of surveyed households and individuals. This synthetic population encompasses the same mobility behaviors as the real one and allows dealing with confidentiality. In this process, Crowdsourcing and Volunteered Geographic Information (VGI) represent very useful data sources. The classical approaches of population synthesis, like synthetic reconstruction and combinatorial optimization, cannot be adapted to manage this huge data. A learning approach is then more suited for the synthesizer to improve the goodness-of-fit of its artificial population as Crowdsourcing data becomes richer. To satisfy such learning requirement, we introduce an evolutionary algorithm for population synthesis. Our results confirm that we can gain incrementality without losing goodness-of-fit.