Particle Swarm Optimization: Efficient Globally Convergent Modifications

In this paper we consider the Particle Swarm Optimization (PSO) algorithm [10, 7], in the class of Evolutionary Algorithms, for the solution of global optimization problems. We analyze a couple of issues aiming at improving both the effectiveness and the efficiency of PSO. In particular, first we recognize that in accordance with the results in [5, 6], the initial points configuration required by the method, may be a crucial issue for the efficiency of PSO iteration. Therefore, a promising strategy to generate initial points is provided in the paper. Then, we address some very preliminary aspects of PSO global convergence towards stationary points, for some Ship Design problems. To this purpose observe that the class of Ship Design applications includes several challenging smooth problems, where expensive simulations provide information to the optimizer, and each function evaluation may require up to hours of CPU-time. In addition, the final solution provided by the optimization method is also required to be a stationary point.