Real-parameter compact supervision for the Particle Swarm Optimization (RCSPSO)

This paper proposes the Real-parameter compact supervision for the Particle Swarm Optimization (RCSPSO) in order to optimize problems with continuous parameters. RCSPSO uses the evolutionary configuration of the Real-valued Compact Genetic Algorithm (RCGA) and the search philosophy of the Particle Swarm Optimization (PSO). As a Compact Evolutionary Algorithms (CEA), RCGA rather than operating on a population of individuals processes a statistical representation of that population. Thus, it shows a very explorative behavior. In contrast, PSO despite having access to several solutions only uses the best solution in order to explore the search space. Although it hardly consumes any additional memory, it provides great insight on the potential exploration areas to the particles. Moreover, to improve the update operation of the probability vector in RCGA, it uses an algorithm that prevents inaccurate influence of particle´s fitness on its gene´s fitness by evaluating each gene separately. Furthermore, to improve sampling the search space, the algorithm uses a combination of Cauchy and Gaussian distributions. To show the algorithm´s viability, we use Differential Evolution (DE), CEAs and PSO on some well-known benchmark functions under identical initial conditions. The results show that the proposed method outperforms the aforementioned algorithms in majority of simulation scenarios.