Adaptive Filtering for Unstable Power System Harmonics using Artificial Network

Conventional approaches for harmonics filtering usually employ either passive, active filtering techniques or hybrid filters. This paper proposes an adaptive harmonic filtering approach using a modified discrete Hopfield network model. The advantage of the scheme is that it can extract the fundamental component of the distorted current and provide a suitable compensation current as the power harmonics may vary in amplitude and frequency from time to time. Therefore, the time-variant harmonic environments in real-time machine systems can be adapted successfully. Real-time performance experiments verify that the proposed scheme is feasible in term of real-time tracking, adaptive low frequency harmonics filtering, fast training and convergence speed.