Fuzzy neural network based estimation of power electronic waveforms

Abstract

Neural networks and fuzzy logic are showing a good promise for application in power electronics and motion control systems. They have been applied in feedback control of converter and drives, estimation of waveforms and signals, and performance enhancement control. Fuzzy logic and neural networks are appropriate where the plant model is ill-defined, non-linear and has parameter variation problems. Besides, such technologies have distinct advantages when compared to a digital signal processor (DSP) based implementation, like fast response, robustness and immunity to harmonic noise. They are model-free estimators since they "learn from experience" with numerical and linguistic data. The present work uses a fuzzy-neural-network (FNN) where a neural network topology emulates fuzzy reasoning. Such neural network permits automatic identification of fuzzy rules and tunes the membership functions. The distorted line current waves in a threephase diode rectifier feeding an inverter-machine load have been taken into consideration and a FNN has been applied to estimate rms current and fundamental rms current. The results of the estimation have been compared with the actual values, and indicate good accuracy. Although the paper considers a relatively simple estimation problem, the fuzzy-neural-network technique can be extended to more complex waveforms and in the estimation of signals of scalar or vector-control drives.