Saved in:
Bibliographic Details
Main Authors: Brar, Avalpreet Singh, Su, Rong, Li, Yuling, Zardini, Gioele
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2412.16632
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915422688772096
author Brar, Avalpreet Singh
Su, Rong
Li, Yuling
Zardini, Gioele
author_facet Brar, Avalpreet Singh
Su, Rong
Li, Yuling
Zardini, Gioele
contents Ride-hailing systems often suffer from spatiotemporal supply-demand imbalances, largely due to the independent and uncoordinated actions of drivers. While existing fleet rebalancing methods offer repositioning recommendations to idle drivers to improve service efficiency, they typically assume full driver compliance: an unrealistic premise in practice. We propose an Adherence-Aware Vehicle Rebalancing (AAVR) framework that explicitly models and addresses uncertainties in driver adherence, stemming from individual behavioral preferences and dynamic trust in the recommender system. Our approach integrates (i) region-specific XGBoost models for demand forecasting, (ii) a network-level XGBoost model for inter-region travel time prediction, (iii) driver-specific logit models to capture repositioning preferences, and (iv) driver-specific Beta-Bernoulli Bandit models with Thompson Sampling to track and update each driver's confidence in the system over time.These elements are incorporated into a novel optimization framework that generates adherence-aware repositioning recommendations. To enable real-time implementation, we further develop a linearized version of the AAVR model. Extensive simulations on the NYC taxi dataset demonstrate that AAVR significantly outperforms four state-of-the-art adherence-agnostic baselines, achieving a 28% increase in served demand, a 22.7% reduction in customer wait times, a 28.6% increase in platform earnings and a 26% gain in driver profits on average.
format Preprint
id arxiv_https___arxiv_org_abs_2412_16632
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Vehicle Rebalancing Under Adherence Uncertainty
Brar, Avalpreet Singh
Su, Rong
Li, Yuling
Zardini, Gioele
Systems and Control
Ride-hailing systems often suffer from spatiotemporal supply-demand imbalances, largely due to the independent and uncoordinated actions of drivers. While existing fleet rebalancing methods offer repositioning recommendations to idle drivers to improve service efficiency, they typically assume full driver compliance: an unrealistic premise in practice. We propose an Adherence-Aware Vehicle Rebalancing (AAVR) framework that explicitly models and addresses uncertainties in driver adherence, stemming from individual behavioral preferences and dynamic trust in the recommender system. Our approach integrates (i) region-specific XGBoost models for demand forecasting, (ii) a network-level XGBoost model for inter-region travel time prediction, (iii) driver-specific logit models to capture repositioning preferences, and (iv) driver-specific Beta-Bernoulli Bandit models with Thompson Sampling to track and update each driver's confidence in the system over time.These elements are incorporated into a novel optimization framework that generates adherence-aware repositioning recommendations. To enable real-time implementation, we further develop a linearized version of the AAVR model. Extensive simulations on the NYC taxi dataset demonstrate that AAVR significantly outperforms four state-of-the-art adherence-agnostic baselines, achieving a 28% increase in served demand, a 22.7% reduction in customer wait times, a 28.6% increase in platform earnings and a 26% gain in driver profits on average.
title Vehicle Rebalancing Under Adherence Uncertainty
topic Systems and Control
url https://arxiv.org/abs/2412.16632