Distributionally Robust and Data-Driven Solutions to Commercial Vehicle Routing Problems

Abstract

In this paper, we study the routing of commercial electric trucks through an application of distributionally robust optimization (DRO) for route planning and dispatch. This approach aims to minimize total cost of operation for the fleet, and considers the variability in energy consumption due to uncertain road conditions, traffic, weather and driving behavior. Furthermore, we augment the distributionally robust energy minimizing vehicle routing problem by learning the energy efficiency distribution over a horizon. We show that convergence to the true distribution is achieved while learning from samples taken from vehicles in operation on the network. With DRO, it is possible to reduce the number of failures due to insufficient battery energy along the route. This stands in contrast to deterministic optimization, which assumes constant energy consumption and cannot learn from data, resulting in occasional failures. Numerical experiments are conducted to validate this method and to compare with the deterministic model.