Neural Networks based Shunt Hybrid Active Power Filter for Harmonic Elimination

Iqbal, Muzammil, Jawad, Muhammad, Jaffery, Mujtaba Hussain, Akhtar, Saleem, Rafiq, Muhammad Nadeem, Qureshi, Muhammad Bilal, Ansari, Ali and Nawaz, Raheel ORCID: https://orcid.org/0000-0001-9588-0052 (2021) Neural Networks based Shunt Hybrid Active Power Filter for Harmonic Elimination. IEEE Access, 9. pp. 69913-69925.

Preview

Published Version Available under License Creative Commons Attribution. Download (1MB) | Preview

Abstract

The growing use of nonlinear devices is introducing harmonics in the power system networks that results in distortion of current and voltage signals causing damage to the power distribution system. Therefore, in power systems, the elimination of harmonics is of great concern. This paper presents an efficient techno-economical approach to suppress harmonics and improve the power factor in the power distribution network using neural network algorithms-based Shunt Hybrid Active Power Filter (SHAPF), such as Artificial Neural Network (ANN), Adaptive Neuro-Fuzzy Inference System (ANFIS), and Recurrent Neural Network (RNN). The objective of the proposed algorithms for SHAPF is to reduce Total Harmonic Distortion (THD) within an acceptable range to improve system quality. In our filter design approach, we tested and compared conventional pq0 theory and neural networks to detect the harmonics present in the power system. Moreover, for the regulation of the DC supply to the inverter of the SHAPF, the conventional PI controller and neural networks-based controllers are used and compared. The applicability of the proposed filter is tested for three different nonlinear load cases. The simulation results show that the neural networks-based filter control techniques satisfy all international standards with minimum current THD, neutral wire current elimination, and small DC voltage fluctuations for voltage regulation current. Furthermore, all three neural network architectures are tested and compared based on accuracy and computational complexity, with RNN outperforming the rest.