Integrating Clustering Methodologies and Routing Optimization Algorithms for Last-Mile Parcel Delivery

Abstract

This paper aims to design a two-echelon parcel distribution network modeled as the Two-Echelon Vehicle Routing Problem (2E-VRP). In this problem, e-cargo bikes perform the last-mile delivery. In fact, this transportation mode is positioned as a promising alternative to make last-mile delivery. Studies show cost and carbon dioxide equivalent (CO2e) emissions savings with cargo bikes setup compared to conventional vans. To solve this problem, a three-stage decomposition algorithm is proposed. In the first stage, the non-supervised machine learning clustering method 2D-k-means is considered to cluster the clients to the satellites. The second and third stages comprise the second and first echelon routing. The last two stages use a heuristic based on the Nearest Neighbor (NN) procedure. Two local search operators were used as improvement algorithms for the solution given by the NN in the second stage. There are scarce studies that use the 2D-k-means algorithm in this urban distribution network context. Experiments are run using a small instance based on real data from a delivery company in the city of Paris, France. Results show that the fixed costs and the cost of energy consumption of the e-cargo bikes are cheaper than the van used in the first echelon. Also, a reduction of 8.2% in terms of travel time is obtained when the Relocate local search is applied. Additional savings are achieved in performance indicators.