Very Short-Term Reactive Power Forecasting Using Machine Learning-Based Algorithms

The growing popularity of microgrids and distributed generation has encouraged further research into the accurate regulation of the electrical grid, especially with consideration to the intricate variations in reactive power and fluctuating power factors. Reactive power, which encompasses the power consumed and produced by the inductive and capacitive elements of a power system, is crucial for maintaining stable and secure grid operation as well as for minimising power losses and enhancing voltage profiles. Reactive power forecasting (RPF) is a critical aspect of power forecasting, especially in the context of ensuring the stability and reliability of the power grid. The application of machine learning (ML) algorithms in RPF provides benefits such as enhanced forecast precision, the capacity to address challenges related to renewable energy integration, maintaining efficient energy resource management, and ultimately improving the consistency of the power grid. This paper delves into the research and implementation of real-time very short-term RPF by employing long short-term memory (LSTM) and gated recurrent unit (GRU) networks, and extreme gradient boosted decision trees (XGBoost) in a large hospital complex situated in Adana, Turkiye. Consequently, utilised algorithms have been compared in terms of coefficient of determination (R-2), mean absolute error (MAE), mean absolute percentage error (MAPE), and mean absolute scaled error (MASE). There is a lack of real-time applications of RPF in the existing literature and this study aims to address this gap while also providing support to future researchers in this area.