01763991

Component

https://annualmeeting.myt...nlineProgram/Details/15830

As the complexity of transportation interventions and the availability of data increases, demand models must become more representative of thought processes and intentions behind those choices. This study presents an approach to generating daily activity schedules, or individual sequences of activities for a day, based on dynamic Bayesian networks. Following the framework of Latent Plan Models, the study leverages Model-Based Machine Learning (MBML) to estimate a temporal Bayesian network with a latent layer. In the framework, an observable action at any time period is based on the plan at the previous time period, which is in turn conditioned on the previously performed action. The proposed model structure includes a latent layer associated with a plan and an observed one associated with the performed activity. The MBML approach offers a coherent framework, based on probability theory and Bayesian Inference (BI), which formally accounts for the uncertainty and dynamic issues that are inherent to travel decision making. The generative performance of the presented model is tested by comparing it to two baseline models without latent layers: one that uses BI for the estimation and one that is based on Maximum Likelihood Estimation (MLE). The results suggest that the authors' model outperforms both baselines at generating activity sequences that resemble the true sequence distribution. The authors demonstrate that the proposed model produces meaningful patterns in the latent space, showing different regimes for activity-making that evolve throughout the day. These patterns provide insights regarding the structural dynamics that condition the observed travelling behavior.

This paper was sponsored by TRB committee AEP50 Standing Committee on Transportation Demand Forecasting.

TRBAM-21-03933

English

20p

2021

Transportation Research Board 100th Annual Meeting

Digital/other

Figures; References; Tables

Activity choices; Bayes' theorem; Demand; Distributions (Statistics); Estimating; Forecasting; Machine learning; Networks; Travel behavior

Planning and Forecasting; Transportation (General)

Transportation Research Board Annual Meeting 2021 Paper #TRBAM-21-03933

TRIS, TRB, ATRI

Dec 23 2020 11:17AM