| In this research, we introduce the production planning - multi-depot location routing problem (PP-MDLRP). The objective of the PP-MDLRP is to minimize transportation and warehousing costs, i.e., inventory and usage, by selecting warehouses from a set of possible warehouses, creating routes that visit each customer exactly once, and developing a production plan to satisfy demand in each period. The PP-MDLRP is a multi-period extension of the multi-depot location routing problem (MDLRP). Because the PP-MDLRP is NP-hard, we develop decomposition based heuristic approaches.;In Chapter 3, we develop a capacitated hierarchical clustering heuristic to solve the MDLRP. Because hierarchical clustering does not require a priori assumptions about the number of clusters, we propose two stopping criteria: minimum number of clusters required and change in within cluster variation.;In Chapter 4, we test the performance of the two stopping criteria introduced in Chapter 3 under varying assumptions and compare the performance of our heuristic approach with existing heuristics and metaheuristics proposed in the literature. Results indicate that on average the minimum number of clusters stopping rule provides better solutions than the change in variation stopping rule after the descent heuristic improvement phase. However, each stopping rule finds a better solution for 1/3 of the cases. Our heuristic results compare favorably with other heuristic approaches proposed in the literature. For instances with capacitated warehouses, our simulated annealing implementation performs better than various alternative metaheuristics proposed in the literature. For instances with uncapacitated warehouses and fixed vehicle costs the results are mixed.;In Chapter 5, we develop two decomposition based heuristic approaches for the PP-MDLRP. Results show that after the improvement phases, the mixed decomposition approach provides better solutions than the functional decomposition approach in 94% of our test cases. The relative difference in heuristic performance is affected by the underlying problem structure. In particular, the interaction among number of warehouses, number of areas with a high concentration of customers, and the fraction of customers that are located in these areas is statistically significant. |
