A Comparative Study of Transformers Based Machine Learning Models for Solar PV Power Forecasting

Precise solar power forecasting is essential for ensuring optimal power quality and system dependability due to the growing integration of photovoltaic (PV) facilities into power systems. The intermittent and unpredictable nature of PV power poses a substantial problem for system operators when it comes to successfully managing and distributing power from PV power stations. Conventional models commonly rely on meteorological data to accurately predict PV power generation in the day ahead. Nevertheless, these forecasts frequently prove inaccurate during severe weather conditions. In this paper, the effectiveness of novel time series forecasting methods, including Temporal Convolutional Mixers (TCMix), Decomposition Linear (DLinear), and a hybrid of these two, for short-term and long-term solar power forecasting is evaluated. The projected outcomes from the suggested photovoltaic power forecasting model closely correspond with the recorded measurement data. Moreover, this model showcases enhanced prediction precision in contrast to conventional approaches that employ artificial neural networks (ANN).