Quaternion Gated Recurrent Units for Renewable Energy: Improving Power Forecasting

The unstable nature of renewable energy sources makes it necessary to optimize their management to maximize efficiency and reduce final emissions. In this work, we propose the quaternion-valued gated recurrent unit (QGRU) for the (deterministic) energy forecasting task. The presented models are used for the prediction of Photovoltaic and Wind Power sources, focusing on the GEFCom2014 dataset, which consists of several pieces of information, such as weather data and the corresponding power production collected over different years. The quaternion embedding of the data has been also described. Numerical results and comparisons with state-of-the-art approaches show the effectiveness of the proposed idea, highlighting that the QGRU outperforms most of the time the correspondent GRU model.