New GOPSO and its application to robust identification

Modeling of complex nonlinear systems has become a challenging task in presence of outliers. In this scenario a robust norm with an evolutionary approach does a potential job. A modified evolutionary algorithm GOPSO (global selection based orthogonal PSO) is proposed which offers a more accurate and computationally efficient training compared to OPSO (Orthogonal PSO). The potential of the proposed algorithm has been demonstrated on six benchmark multi-modal optimization problems. Further, robust identification models has been developed by combining Wilcoxon norm with a functional link artificial neural network (FLANN) structure trained by the proposed GOPSO. Exhaustive simulation studies on five complex plants show superior performance of proposed models when output of plant gets corrupted upto 50% outliers.