Large-scale system identification using self-adaptive penguin search algorithm

Abstract

From an engineering point of view, non-linear systems are essential to the operation ofcontrol systems, because all systems actually have a non-linear state in nature. In reality,there are many different kinds of non-linear systems hidden by this negative definition.For successful analysis and control, the identification of non-linear systems using unknownmodels is typically necessary. Till now, numerous approaches are developed for identifyingnon-linear systems, but it cannot be employed with a large number of components. More-over, system identification is typically restricted to output and input signals alone, alsosuch systems are rarely used in reality. This is the primary justification for using non-linearsystems in this research. So, this research proposed a non-linear model of system iden-tification for large-scale systems under the consideration of two systems: bilinear systemand Volterra system. Therefore, a novel algorithm named Self Adaptive Penguin SearchOptimization (SAPeSO) is introduced to attain the system characteristics properly andminimize the output variation. Finally, the effectiveness of the proposed work is com-pared with existing works in terms of various error measures. This research mainly focuseson the application-oriented engineering problems. In particular, the Mean Absolute Error(MAE) of the proposed work for the Volterra system at 4000 samples is 18.83%, 14.05%,8.88%, 29.72%, 19.91%, and 6.70% which is better than the existing bald eagle search(BES), arithmetic optimization algorithm (AOA), whale optimization algorithm (WOA),nonlinear autoregressive moving average with exogenous inputs- frequency response func-tion+principal component analysis (NARMAX-FRF+PCA), Global Gravitational SearchAlgorithm-Assisted Kalman Filter (CGS-KF), and sparse regression and separable leastsquares method (SR-SLSM) methods, respectively. Finally, the error is minimum for theproposed model when compared with the other traditional approaches.