Machine Learning based Stacked Ensemble Model for A Photovoltaic System Power Generation Forecasting: A Case Study of Dammam Saudi Arabia

As Saudi Arabia makes a significant transition towards renewable energy, solar energy have emerged as a feasible alternative owing to the country's favorable geographical conditions. However, this transition poses critical challenges that necessitate careful consideration. Due to the current energy mix's predominance of oil, the integration and management of solar power plants into the existing grid necessitate accurate forecasting. This, in turn, can help minimize grid instability and enhance the overall sustainability and resilience of the energy infrastructure. The issue is addressed in this work by using machine learning algorithms. For precise solar power prediction, a stacked ensemble model that employes Artificial Neural Networks (ANN) and k-Nearest Neighbours (KNN) as the base models complemented by linear regression as the meta model is proposed in this work. The ensemble model was compared against standalone KNN, ANN, stochastic gradient descent, Adaboost and XGBoost models. The results demonstrate superior performance of the ensemble model compared to the standalone models. The performance of the proposed model is verified using datasets obtained from a 5 MWp solar PV power plant based on real time meteological data from meteonom database situated in Dammam, Saudi Arabia. The proposed stacked ensemble model in this study was compared to existing ensemble models . The outcomes showed that, for a polycrystalline photovoltaic system, the suggested model outperformed its predecessors in terms of performance indicators. The work proposed in this study is versatile and can be applied to various regions and power plants of diverse sizes.