Neural Network Training Using Particle Swarm Optimization - a Case Study

Abstract

This paper presents analysis of multiparameter optimization realized applying Particle Swarm Optimization (PSO). Model of Neural Network (NN) was selected as object. The main goal was adaptation of internal coefficients (weights) used in data processing. Selected problems appearing in real engineering applications are analyzed (for the sake of example selection of initial parameters and coefficients of optimization algorithm). Tests present implementation of neural network for two tasks. The first of them is calculation of selected variables from the Lorenz attractor. The next example shows application of neural network as a state variable estimator implemented in electric drive. For this analysis also experimental research is presented. Considered results show flexibility of neural models used for data mapping and effectiveness of Particle Swarm Optimization algorithm.