Development of two matheuristics for production-inventory-distribution routing problem / Dicky Lim Teik Kyee

• Main Author: Dicky Lim , Teik Kyee
• Format: Thesis
• Published: 2018
• Subjects: Q Science (General); QA Mathematics
• Online Access: http://studentsrepo.um.edu.my/10252/2/Dicky_Lim_Teik_Kyee.pdf; http://studentsrepo.um.edu.my/10252/1/Dicky_Lim_Teik_Kyee_%E2%80%93_Dissertation.pdf; http://studentsrepo.um.edu.my/10252/
• Institution: Universiti Malaya

Production, inventory and distribution are amongst the most important components of a supply chain network. Integrating production, inventory and distribution decisions is a challenging problem for manufacturers trying to optimize their supply chain. In general, the problem of optimally coordinating production, inventory and transportation is called the production-inventory-distribution routing problem (PIDRP). The PIDRP model considered in this study includes a finite planning horizon, a single production facility, a set of customers with deterministic and time-varying demand, and a fleet of homogeneous capacitated vehicles. The aim of solving the model is to construct a production plan and delivery schedule which minimizes the overall costs while fulfilling customers’ demand over the planning horizon. We propose an optimization algorithm designed by the interpolation of metaheuristics and mathematical programming techniques, known as MatHeuristics algorithm, to solve the model. In this study, we develop two different two-phase approaches within a MatHeuristic framework, namely MatHeuristic1 and MatHeuristic2. In MatHeuristic1, Phase 1 solves a mixed integer programming (MIP) model which includes all the constraints in the original model except the routing aspects. The routing constraints are replaced by an approximated routing cost in the objective function. A variable neighborhood search (VNS) is proposed in Phase 2. The VNS is employed to improve the solution and it incorporates some aspects of Tabu search to escape from local optima. Whilst in MatHeuristic2, we adopt a slightly different approach. The algorithm starts with the pre-processing stage where the routes are determined to act as input to the MIP in Phase 1. Phase 2 uses VNS and we utilize both phases in an interactive manner. The models in both approaches are solved by using Concert Technology of CPLEX 12.5 Optimizers with Microsoft Visual C++ 2010. Both algorithms are tested on a set of 90 benchmark instances with 50, 100 and 200 customers and 20 periods and the results are competitive when compared to the Memetic Algorithm with Population Management (MA|PM), Reactive Tabu Search (RTS) and Scatter Search (SS). MatHeuristic1 performs well in large instances when compared to other metaheuristics but MatHeuristic2 performs well in smaller cases.