A fast simulation-based optimization method for inventory control of general supply chain networks under uncertainty

Simulation-based optimization can significantly improve operational efficiency of a supply chain network under uncertainty. However, both the noisiness and complexity render the simulation as a black-box function. We propose a novel regional surrogate based framework for inventory optimization in general supply chain networks under demand uncertainty. Both the objective value and service level constraints are estimated by the kriging method using regional information. The aggregated surrogate models are optimized by a trust-region framework. For a case study with 15 inventory storing nodes, the proposed algorithm returns an optimal solution in 2,994 seconds with 6,721 functional evaluations while the genetic algorithm (GA) returns a 36.2% higher objective value after 46,000 function evaluations.