Machine learning-based power output and load prediction for a solar photovoltaic mini grid.

Abstract

Access to sustainable, affordable and reliable energy is one of the essential sustainable development goals. This though has not been achieved in most of the rural communities worldwide. To address this challenge, Renewable energy technologies have been embraced to provide electricity to rural communities especially those with no access to the main grid. Solar photovoltaic (PV) is the most commonly used renewable energy technology in providing electricity to many rural communities in developing countries such as Uganda. However, the power output of solar PV systems is affected by the constantly changing nature of weather conditions and load demand. This greatly affects power production planning during the operation of these systems. In this project, two separate machine learning models are proposed to predict PV power output and the load demand for solar photovoltaic systems. The case study for this project was the 40kWp + 190kWh Utility 2.0 PV minigrid in Mukono district, Uganda. The proposed model for PV power output prediction achieves the RMSE of 9.793 kW and MAE of 5.374 kW and nRMSE of 0.092.For load prediction, the developed model an RMSE of 0.488 kW and MAE of 0.303 kW were obtained. The PV power prediction model is evaluated on synthesized power data using the simulated PV system in MatLab and the load prediction model on open source datasets with varying hyperparameters’ search spaces. Both models developed were deployed in a web application with a dashboard to aid in decision making for PV minigrid operators. More research should be done on other factors that may affect the load prediction model other than considering the historical load data only. Also there is need to integrate machine learning predictive models into an optimization algorithm for solar PV system