TY - JOUR
T1 - Online Framework for Demand-Responsive Stochastic Route Optimization
AU - Peled, Inon
AU - Lee, Kelvin
AU - Jiang, Yu
AU - Dauwels, Justin
AU - Pereira, Francisco Camara
PY - 2020
Y1 - 2020
N2 - This study develops an online predictive optimization framework for operatinga fleet of autonomous vehicles to enhance mobility in an area, where thereexists a latent spatio-temporal distribution of demand for commuting betweenlocations. The proposed framework integrates demand prediction and supplyoptimization in the network design problem. For demand prediction, ourframework estimates a marginal demand distribution for each Origin-Destinationpair of locations through Quantile Regression, using counts of crowd movementsas a proxy for demand. The framework then combines these marginals into a jointdemand distribution by constructing a Gaussian copula, which captures thestructure of correlation between different Origin-Destination pairs. For supplyoptimization, we devise a demand-responsive service, based on linearprogramming, in which route structure and frequency vary according to thepredicted demand. We evaluate our framework using a dataset of movement counts,aggregated from WiFi records of a university campus in Denmark, and the resultsshow that our framework outperforms conventional methods for routeoptimization, which do not utilize the full predictive distribution.
AB - This study develops an online predictive optimization framework for operatinga fleet of autonomous vehicles to enhance mobility in an area, where thereexists a latent spatio-temporal distribution of demand for commuting betweenlocations. The proposed framework integrates demand prediction and supplyoptimization in the network design problem. For demand prediction, ourframework estimates a marginal demand distribution for each Origin-Destinationpair of locations through Quantile Regression, using counts of crowd movementsas a proxy for demand. The framework then combines these marginals into a jointdemand distribution by constructing a Gaussian copula, which captures thestructure of correlation between different Origin-Destination pairs. For supplyoptimization, we devise a demand-responsive service, based on linearprogramming, in which route structure and frequency vary according to thepredicted demand. We evaluate our framework using a dataset of movement counts,aggregated from WiFi records of a university campus in Denmark, and the resultsshow that our framework outperforms conventional methods for routeoptimization, which do not utilize the full predictive distribution.
M3 - Journal article
JO - ArXiv
JF - ArXiv
ER -