Cluster-Based Sampling Allocation for Multi-Fidelity Simulation Optimization

Simulation optimization is widely used to optimize complex systems. High-fidelity simulation can be expensive, especially when the number of designs is large. In practice, fast but less accurate low-fidelity simulation is often available and can provide valuable information. In this paper, we propose a sampling algorithm that utilizes information from multiple fidelity simulation models to improve the efficiency of searching for the best design. A k-means algorithm is introduced to help capture the performance clustering phenomenon among designs, and a cluster validity index is proposed to determine the optimal number of clusters. The proposed sampling algorithm can incorporate the information of performance clusters and approximately minimize the expected opportunity cost of the selected best design. Numerical results substantiate the superior performance of the proposed algorithm.