A Hybrid Time-Series PV Power Forecasting Model Implementing Facebook Prophet and Neural Prophet Algorithms
Abstract
Globally, countries are limiting their dependency on conventional energy as their primary energy source. From this perspective, the ongoing integration of renewable energy sources with existing systems is the focus of researchers and power utilities. Photovoltaics (PV) is one of the leading solar technologies driving substantial growth in renewable energy applications. Consequently, power utilities are faced with large penetration of PV systems within the existing electrical grid. This ongoing integration introduces new electricity generation profiles, which raises technical challenges affecting grid reliability. Therefore, this work proposes building a PV power short-term forecasting model, which can avoid adverse outcomes of large penetrations by balancing the load demand.
Extensive research on this application was used to build a hybrid PV power forecasting model. Emerging machine learning algorithms, known as Prophet and Neural Prophet, were selected to develop a model for a PV power plant located in the United Arab Emirates (UAE)
Three univariate meteorological sub-models were built to provide meteorological forecasts for a seven-year, fifteen-minute resolution duration (2014-2020). These forecasts were injected into the hybrid model to forecast the power output for a duration of a month ahead. The results indicated good performance compared to the individual prophet model, which scored an RMSE value of 0.805589.
