Ghost Kitchen Location Problem for Meal Delivery Services

Abstract

Meal delivery are a key part of on-demand logistics in cities worldwide. Ghost kitchens (“delivery-only kitchens”) have emerged as facilities for preparing and distributing meals to meet online demand. This thesis develops an actor-classification framework covering ghost kitchens, platforms, couriers, and customers. Despite their rise, research on ghost kitchens remains limited, particularly regarding courier behaviour, on-demand delivery and location optimisation. Couriers on e-bikes, bikes, or scooters collect from ghost kitchens and deliver to one or more customers. Since delivery speed depends on kitchen location, site selection is strategically important. It affects private-sector profitability through delivery efficiency and demand fulfillment and influences public planning through zoning, infrastructure, and urban policy. Demand randomness and order variability are key features. Entropy maximisation leads to a Markov model that reproduces pick up, delivery frequencies and mean delivery time. Key indicators such as mean and variance of delivery times are derived from model parameters. The model is irreducible, ensuring a unique steady state. Two parameter estimation methods are proposed: one uses an urgency input to produce delivery time. The model is calibrated using a public Grubhub dataset and validated with a likelihood ratio test. The model also supports street network trip assignment and when combined with a route choice model, estimates demand for bike lanes or signals. This thesis examines optimal kitchen location using an entropy-based derivation. Grid search shows how relocation affects delivery time and demand. Two solution methods are tested: a modified Weiszfeld algorithm and an Adaptive Step Size Gradient (ASG) method. ASG yields better delivery times and demand outcomes. The thesis offers practical guidance for kitchen siting and represents a novel contribution by optimising dynamic pickup and delivery using a Markov model of courier behaviour.