Meta-heuristics for placing strategic safety stock in multi-echelon inventory with differentiated service times

The computational resolution of multi-echelon safety stock placement problems has attracted ample attention in recent years. Practitioners can obtain good solutions for large supply networks with general structure using the guaranteed service model. The mainstream assumption in this model is that a stock point quotes identical service times to its successors. In business, it is common to assign customers to different customer- and service classes, as it can yield significant cost improvements. Nevertheless, differentiated service times have rarely been considered in computational methods of safety stock placement. We relax the assumption of identical service times in the guaranteed service approach and allow stock points to prioritize between their successors. This increases the complexity of the problem considerably, so that meta-heuristics become the methods of choice. Meta-heuristics need a mapping between safety stock levels in the supply network (from which they compute the holding cost) and an internal representation of stocking decisions (from which they generate new solutions). The design of a representation is non-trivial because of complex interactions between stocking decisions and stock levels. We propose a representation for the safety stock allocation problem with differentiated service times that can be used in general-acyclic supply networks. We apply a local search, a simple genetic algorithm and a problem-adjusted simulated annealing to 38 general-acyclic real-world instances. Results suggest that service time differentiation indeed decreases total holding cost in the network. Simulated annealing outperforms the other meta-heuristics within the set of tested methods with respect to speed and solution quality.