An Improved Particle Swarm Optimization Algorithm to Optimize Modular Neural Network Architectures

Abstract

According to the literature of Particle Swarm Optimization (PSO), there are problems of local minimum and premature convergence with this algorithm. A new algorithm is presented called the Improved Particle Swarm Optimization using the gradient descent method (BP algorithm) as operator of particle swarm incorporated into the Algorithm, as a function to test the improvement. The Gradient Descent Method (BP Algorithm) helps not only to increase the global optimization ability, but also avoid the premature convergence problem. The Improved PSO Algorithm IPSO is applied to Neural Network to optimize the architecture. The results show that there is an improvement with respect to using the conventional PSO Algorithm.