Enhancing the power quality of a grid-connected rooftop solar PV system under varying environmental conditions: Strategic optimization of influencing parameters using machine learning and desirability-driven approach

Power quality (PQ) is a critical concern in grid-connected solar PV systems, as non-linear inverter behavior and environmental variations introduce harmonics and reduce energy output. Most studies analyze PQ in isolation, with few addressing the combined effects of temperature, solar irradiance, and relative humidity using advanced predictive models. This study proposes a machine learning (ML) and Response Surface Methodology (RSM) framework for real-time prediction and optimization of PQ. Ten supervised ML regression models were trained on operational data from a 100 kWp grid-connected solar PV system (338 polycrystalline panels, 320 Wp each, 28.6833° N, 77.4500° E). Ambient temperature, solar irradiance, and relative humidity were used as inputs, while total harmonic distortion (THDi), power factor (PF), and apparent power (kVA) were measured as outputs. Model performance was evaluated using R² and RMSE. Random Forest achieved the highest accuracy with R² = 0.894 and RMSE = 5.87, outperforming traditional regressors. Optimization revealed optimal operating conditions at 948.79 W/m² irradiance, 37 °C temperature, and 36 % relative humidity, resulting in THDi of 7.9 %, PF of 0.995, and apparent power of 63.97 kVA, all compliant with IEEE Standard 519–2014. Compared to previous ANN and MLR-based approaches (R² ≈ 0.75–0.85), the proposed ML-RSM framework offers enhanced predictive performance and provides a practical tool for real-time PV system monitoring and control.