Orthogonal experimental design method used in particle swarm optimization for multimodal problems

Orthogonal Experimental Design (OED) method is usually used to study the effect of several factors simultaneously and the best combination of factor levels can be found in several tests. The Particle Swarm Optimization (PSO) can utilize OED to improve the searching ability. However, the main effect of OED holds only when no or weak interaction of factors exists. This limitation of OED makes PSO search effective on unimodal or simple problems but very vulnerable on complex multimodal problems. This paper presents an effective method utilizing OED on multimodal problems. A new vector is formed through learning particle's previous and neighborhood's best vector. Instead of treating the new vector as exemplar for others to follow, this new vector is treated as base vector which needs to be explored further. Experimental studies on a set of test functions show that OED method used in this way has better robustness and converges closer to the global optimum than several other peer algorithms.