An Artificial Neural Network–Based Approach for Harmonic Components Prediction in Grid-Connected PV Systems

Abstract

With the increasing penetration of solar photovoltaic (PV) generation, non-sinusoidal current components can be injected into the grid due to power-electronic conversion, degrading power quality and potentially affecting system stability. In practical mitigation, response delays can reduce the effectiveness of active filtering; therefore, fast prediction of harmonic distortion is valuable for enabling adaptive control. This study investigates an artificial neural network (ANN) model to predict total harmonic distortion (THD) and harmonic components in a grid-connected PV system using only two environmental inputs: solar irradiance and ambient temperature. The model was trained and tested using measurements from a 6.4 kW PV setup at the Department of Electrical Engineering, IOE Pulchowk Campus. Based on the recorded measurements, THD under the observed operating conditions ranged from 9.12% to 32.45%. The proposed ANN achieved an overall prediction accuracy of 83% on the available datasets, demonstrating that a lightweight model driven by minimal environmental features can provide timely harmonic estimates. These results support the feasibility of using ANN-based prediction as a foundation for real-time monitoring and adaptive harmonic filtering in PV-connected distribution systems.