Ammouriova, M.: Approaches to Enhance the Performance of Simheuristic Methods in the Optimisation of Multi-echelon Logistics Distribution Networks. Series "Fortschritte in der IT in Produktion und Logistik", Vol 3. Göttingen: Cuvillier 2021.

The management of logistics distribution networks is a challenging task. These networks consist of entities, such as stock keeping units (SKUs), and sites. Items are stored as SKUs in the sites and delivered to customers according to their placed orders. Decision-makers select actions, such as centralising an SKU in a site, to reduce costs and increase the service level. These actions might have a conflict impact on the costs and the service level, which triggers the complexity of the logistics distribution network. Moreover, the number of entities and the possible actions are influenced by the size of the networks. The selection of actions forms an NP-hard combinatorial optimisation problem. Thus, decision-makers utilise tools to assist them in selecting actions, such as a logistics assistance system (LAS). Logistics assistance systems have been developed to recommend action plans to assist decision-makers. Action plans include actions to be applied on the network in a given order. These systems can be based on a simheuristic approach to benefit from the power of simulation and metaheuristics. In optimising logistics distribution networks, a metaheuristic algorithm is used to form action plans, and simulation is used to study the impact of these action plans on the costs and the service level. However, simulation is a computationally expensive tool because of the long simulation runs of large networks. Thus, the performance of a LAS is defined in terms of the impact of recommended action plans on a network with respect to the number of simulation runs needed to recommend action plans. This research aims to propose approaches to enhance the performance of a LAS that is based on a simheuristics approach. The first approach utilises problem information called “domain-specific information” (DSI) to guide the search for promising actions, such as the type of changes applied by actions, the success of actions, and correlations between actions. According to the type of changes, actions are classified as structural actions and parametrical actions. Structural actions add entities or remove them from the network, while parametrical actions modify the entities’ parameters, such as the stock level of an SKU. The success of actions is defined based on the impact of the actions on the network. Relations between actions and their impact on the network define their correlations. The DSI approach alters the selection probabilities of actions to guide the metaheuristic algorithm in selecting promising actions. The other two approaches aim to reduce the number of simulation runs. The first approach reduces the number of actions. Actions are grouped, and the grouped actions replace actions in the space of possible actions. The second approach defines conditions to skip simulation runs. In this approach, equivalent action plans are defined. Equivalent action plans have the same impact on the performance of the network. The three approaches were evaluated on a logistics distribution network. The DSI approach found action plans with lower costs and a higher service level compared to a random selection of actions. Grouping actions formed a smaller space of actions that helped the LAS to recommend actions plans in a lower number of simulation runs. The last approach, defining equivalent action plans, helped to reduce the number of simulation runs. In conclusion, the evaluation showed that these approaches could be used to recommend promising action plans in a lower number of simulation runs.