Fast and efficient heuristics for the (s, S, T) inventory policy

Abstract

We consider a single-echelon inventory system facing stochastic demand controlled by the standard ( s , S , T ) periodic review policy. While an exact algorithm for optimising all three policy variables exists, high computational requirements prohibit real industrial applications. To deal with this problem, here we propose three search heuristic algorithms that can approximate the optimal policy solution very fast and accurately. All heuristics make use of the empirical observation that the optimisers of the ( s , S , T ) policy are closely related to those of the classical ( r , nQ , T ) policy. For the latter, however, a fast near-optimal heuristic exists. Therefore, starting with the ( r , nQ , T ) optimal solution, all algorithms employ local neighbourhood search to determine the required ( s , S , T ) policy solution approximation. Experiments with well-known meta-heuristics such as Simulated Annealing, Tabu Search, and others show that the heuristics compare favourably in both solution time and quality. Moreover, compared with the existing exact algorithm heuristic solutions are very satisfactory in quality and are obtained in a fraction of CPU-time.