An Improved Optimization Strategy and Its Application to Clustering Analysis

Abstract

In this paper, a new optimization strategy is put forward which locates as many potential unimodal regions as possible in the search space. The potential optima can be further explored by a global optimization method for searching in the identified unimodal regions. The proposed strategy was evaluated by the optimization of test functions. The results obtained by this approach are comparable with those achieved by variable step size generalized simulated annealing (VSGSA) and a genetic algorithm (GA). Finally, we used this strategy in a clustering analysis of a tobacco data set.