Development of improved functional neural network based autoregression models for power quality improvement

Abstract

This paper presents two improved and adaptive models based on functional neural network and Autoregression (FNNAR) analysis. These models have been developed for estimating the fundamental component of nonlinear and varying load current and computing the exact compensation required in a power distribution system. The proposed FNNAR analysis involves two steps: The first step is designed to estimate the fundamental current in terms of polynomial or trigonometric functional expansion terms; while, the second step involves computations based on the weighted sum of the delayed output terms. An activation function is additionally incorporated to account for the nonlinearity and sudden variations of load current. Both the FNNAR models are developed and their parameters computed in an adaptive manner from the input–output data. The simulation results on a single-phase 110 V, 50 Hz system power distribution system are validated by a scaled down experimental model showing hardware results depicting load compensation. Adequate comparison of the two developed models is also discussed in the paper with two advanced variants of conventional algorithms viz. Least means square algorithm and second order generalized integrator based filtering technique.