Optimized Photovoltaic Power Forecasting with Feature Engineering and Stacked Machine Learning Models

A precise prediction of solar energy is crucial to enable a greater degree of incorporation of renewable energy into the regulation of the current power system. With the increasing implementation of solar photovoltaic plants, research and development efforts have achieved substantial advancements in evaluating and studying the performance of solar installations to enhance their dependability. Given the abundance of data at unusually detailed levels, there is a chance to apply data-based algorithms to improve solar power production forecast accuracy. This work presents a layered machine learning model that generates hourly forecasts for a 100 Kw photovoltaic (PV) system for Dehradun, Uttarakhand, India. This methodology stacks three models: Random Forest (RF), Extreme Gradient Boosting Machine (XGB), and Multi-Layer Perceptron (MLP) as the base learner, with Multiple Linear Regressor (MLR) serving as the meta learner. The findings indicate that the stacked ensemble model does better than the single models that were used for comparison, with nRMSE values of 1.06% and 3.25%, respectively. These results open up new ways to improve PV prediction in this area.