ORBIT CLUSTERING BASED ON TRANSFER COST

摘要

We propose using cluster analysis to perform quick screening for combinatorialglobal optimization problems. The key missing component currently preventingcluster analysis from use in this context is the lack of a useable metric functionthat defines the cost to transfer between two orbits. We study several proposedmetrics and clustering algorithms, including k-means and the expectation maximizationalgorithm. We also show that proven heuristic methods such as theQ-law can be modified to work with cluster analysis.