Reconfiguration of last-mile supply chain for parcel delivery using machine learning and routing optimization

Abstract

Last-mile delivery has several negative environmental impacts in urban areas because of its high levels of greenhouse gas emissions and air pollution, as well as traffic congestion. These issues motivate decision-makers to redesign the delivery networks and make them more sustainable and efficient. A well-planned territory design can reduce total travel times and distances in urban distribution systems, in addition to balancing the workload between drivers. In this study, a two-echelon parcel distribution network modeled as the two-echelon vehicle routing problem with territory design and satellite location decisions is considered. A three-stage decomposition algorithm is proposed to solve this problem. In the first stage, a non-supervised machine learning clustering method is applied, followed by an algorithm based on the nearest-neighbor routing procedure, to find a set of routes for the second and first echelons. An improvement heuristic was also applied to improve the results in terms of the second echelon routing, considering the computational complexity of a large-scale instance. A case study based on real data from a delivery company in the city of Paris, France is adopted to perform the experiments. The outcomes of this paper show an improvement of 22.6% in travel time and distance. This reduction is also assessed with performance indicators like land use, fixed costs, energy consumption, carbon dioxide equivalent, and fine particles emissions.