Optimal Routing and Scheduling in E-commerce Logistics using Crowdsourcing Strategies

A Master of Science thesis in Engineering Systems Management by Eman Mohamed entitled, "Optimal Routing and Scheduling in E-commerce Logistics using Crowdsourcing Strategies," submitted in July 2017. Thesis advisor is Dr Malick Ndiaye. Soft and hard copy available.

Abstract

The internet penetration has transformed the way people shop and has enlarged online shopping popularity. From 2014 to 2016, electronic commerce has increased by 50% and the forecast predicts a three fold increase by 2020. This growth creates various challenges for logistics providers when it comes to home deliveries. Researches showed that the last mile delivery accounts for 13% to 75% of logistics costs and is currently regarded as the most problematic and most polluting section of the entire supply chain. Till now, there is no straight forward solution but technology is paving the way to optimize this part of the logistical process. At the same time, the popularity of mobile applications is an opportunity for companies as they create valuable connections between companies and customers. This research aims to benefit from these connections to finish the last leg of delivery through the use of crowdsourcing. The idea is to use a pool of citizens to deliver products from several lockers distributed around thecity to consumers' doorsteps. Deliveries occur during a set time window for each customer taking into consideration drivers' capacity and availability time. The problem as described is proven to be NP-Hard, thus its solution should be verified using exactand heuristic-based techniques for large size instances. We propose two Integer Linear programming (ILP) formulations with the objective of minimizing the total reward paid to crowd-workers through optimum assignment of parcels to drivers and optimum routing for each driver. The proposed models are first solved using Lingo software and their run time compared. Furthermore, a Variable Neighborhood Search (VNS) algorithm that uses a Variable Neighborhood Descent (VND) in its local search phase is developed using C++ programming language to solve large-size problems in a reasonable time. The proposed algorithm returned exact and near-optimal solutions within few seconds depending on the problem's size and all results were obtained within an average of 20 seconds. A sensitivity analysis is conducted by varying the main model parameters to determine inputs' their impact on the problem. Although there was a clear correlation between some parameters and the total delivery cost, some other experiments were inconclusive.