Time series forecast of power output of a 50MWp solar farm in Ghana

Abstract

The energy industry in Ghana is working towards the strategic objective of accelerating the development and use of energy efficiency and renewable energy technology to attain a 10 % penetration of the country's electricity mix. However, due to the inherent unpredictability of solar energy compared to conventional sources, adjustments in power system planning and operations will be required to achieve these targets. The variations in solar energy output can cause problems for the grid infrastructure, especially for large-scale solar farms, potentially leading to poorer power flow quality. An autoregressive model (AR) serving as a benchmark model was developed as a reference for the Facebook prophet model. The Prophet outperformed the AR model in percentage-based metrics, with a Mean Absolute Percentage Error (MAPE) of 12.1 % and a Median Absolute Percentage Error (MdAPE) of 13.8 % , both lower than the AR model's 16.28 % and 17.23 % respectively. However, the AR model demonstrates stronger performance in absolute error metrics, suggesting it better captures magnitude changes, whereas Prophet excels in relative error metrics, indicating better robustness to scale and variability. It is expected that the results of this study will improve Bui Power Authority (BPA) confidence in the effective decision-making of energy generation and supply. Moreso, this study also contributes to existing research, particularly in Ghana, providing insights to optimize energy production, improve grid stability, and enhance revenue streams.