AI/ML-Based Short-Term Solar PV Power Forecasting: A Comparative Study of ANN, Random Forest, And SVR

Abstract

The increasing penetration of solar photovoltaic (PV) generation into modern power systems has created a growing need for accurate short-term PV power forecasting to support reliable grid operation, energy management, and renewable energy integration. This study presents a comparative analysis of three machine learning models, namely Artificial Neural Network (ANN), Random Forest (RF), and Support Vector Regression (SVR), for short-term solar PV power forecasting. The forecasting framework utilizes solar irradiance, ambient temperature, module temperature, temporal variables, and historical PV power as input features. A chronological training-testing strategy is adopted to preserve the time-series characteristics of PV generation. The performance of the developed models is evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), normalized RMSE (nRMSE), and coefficient of determination (R²). Initial results obtained from the experimental dataset demonstrate that Random Forest provides superior forecasting performance compared with ANN and SVR, achieving lower prediction errors and a higher coefficient of determination. The results indicate that ensemble-based machine learning techniques can effectively capture the nonlinear relationship between environmental conditions, historical generation, and PV power output. The proposed comparative framework provides a foundation for the development of advanced PV forecasting systems for grid-connected solar power plants. Future work will incorporate Long Short-Term Memory (LSTM) networks and validate the proposed models using real-world Indian PV plant data under different forecasting horizons and operating conditions.