Predictive Modeling of Harmonic Distortion in Electrical Systems Using Machine Learning

Conventional techniques used for current harmonic assessment suffer from limitations of inaccurate detection of total harmonic distortion due to inherent delay produced by the sampling window. To overcome this limitation, this paper proposes an AI model to predict harmonic distortions in grid current caused by interfaced renewable energy sources like PV and EV charging stations. The single AC source and single nonlinear load are simulated in MATLAB/Simulink, creating harmonics in the charging current and voltage supplied by the AC source. Sufficient data corresponding to various harmonic components ranging from third to twenty-one is generated using a simulation study. The generated data trains the machine learning model using different regression-based algorithms. Regression algorithms like Random Forest, XGBoost, and CatBoost showed strong accuracy in predicting harmonic distortion in grid current. The stacking strategy is used to combine the best-performing algorithms, which used linear regression as the meta model. The result captured from stacking shows the highest accuracy. This accuracy helps in the accurate assessment of harmonic components in charging drawn by nonlinear loads like EV chargers. The machine learning regression techniques precisely estimate various harmonic components. The paper follows IEEE-519 standard for harmonic control in electrical distribution networks.