International Journal of Computational Intelligence Research

This paper reports the results of the adoption of a probabilistically defined communication structure in a special algorithm coined as EPSO – Evolutionary Particle Swarm Optimization, which is classified as an evolutionary algorithm using a particle movement rule as the recombination operator. Alternatively, EPSO may be seen as an algorithm of the family of PSO (Particle Swarm Optimization) but with a self-adaptive mechanism applied to make the weights of the movement rule evolve improving the performance of the algorithm. The paper presents results showing that a probabilistically controlled communication (to the particles of a swarm) of the location of the best-sofar point leads to better convergence and that the optimal value of the probability of communication depends on the topology of the surface being searched. Also, full communication (similar to classical PSO) has in all cases been shown to be worse than probabilistically constrained communication. This is demonstrated by comparing results in different test functions and also in the application of EPSO to an industrially relevant application – the reactive power planning in large scale power systems.