ANN model identification: Parallel big bang big crunch algorithm

Abstract

This paper proposes a modification to existing big bang big crunch optimization algorithm that uses the concept of more than one population. In this the search begins with all the populations independently in parallel and as the algorithm proceeds the local best of the individual populations interact with global best to avoid local minima. In order to validate the proposed approach the authors have identified two models one from control field namely rapid battery charger and second a rating system for institutes of higher learning and compared its results with simple BB-BC based approach.The author further compared results of the proposed approach with the results of other recent soft computing based algorithms for ANN model identification. The proposed algorithm outperformed all of the other 7 algorithms in terms of MSE and convergence time.