Data-Driven Deep Learning Algorithm for Harmonics and Interharmonics Flicker Prediction and Mitigation in the Smart Grid System

Abstract

In grid-connected smart distribution systems, the prediction and mitigation of flicker caused by harmonics and interharmonics present a significant challenge for stable grid operation, particularly in the presence of distributed renewable energy sources (DRESs). The intermittent nature of DRES introduces dominant low-frequency components that exacerbate flicker issues in smart grid environments. To address these challenges, this research proposes a deep convolutional neural network (DCNN) model, employing a mean squared error loss function, designed to outperform conventional active power filters and static Var compensators (SVCs) in flicker mitigation. The training dataset for the proposed DCNN was obtained from real-time measurements at the Muppandal Wind Farm in Tamil Nadu, India. Numerical evaluations based on flicker sensation, prediction accuracy, perceptibility, and error rates demonstrate the superior performance of the proposed method compared to existing techniques. The results confirm that the proposed DCNN model is a viable solution for real-time flicker prediction and mitigation in smart grid applications, especially those integrating intermittent renewable energy sources.