Machine Learning-Based Short-Term Photovoltaic Power Forecasting Using Exogenous Meteorological Data

Abstract

Machine Learning (ML) plays a transformative role in optimizing energy management processes. Hence, this study evaluates the capability of predicting short-term Photovoltaic (PV) power generation utilizing exogenous weather data through ML techniques, including Support Vector Machine (SVM), Random Forest (RF), and Multilayer Perceptron (MLP). The collected data underwent preprocessing, and the respective models were trained. The aim is to achieve higher prediction accuracy, reduced prediction errors and efficient use of time computation. The results demonstrate promising performance of the proposed Recursive MLP (RMLP) compared to others in forecasting short-term photovoltaic power generation, showcasing a low prediction error with RMSE of 0.0911 kW, and significant correlation R^2 of 98%. The primary findings drawn from this investigation emphasize the effectiveness and capacity of machine learning based approaches to forecast short-term solar energy production, thereby presenting optimistic prospects for the enhanced integration of renewable energies into the power grid.