Hybrid Transformer Forecasting for Renewable Microgrids

Abstract

Study aimed to develop an intelligent system for short-term forecasting of electricity generation by wind and solar power stations, based on hybrid transformer neural networks, with a view to improving the efficiency and stability of microgrids in western Ukraine, particularly within the Volyn region. During the study, empirical data on meteorological parameters – such as wind speed, solar irradiance, temperature and humidity – were collected, along with historical generation data from renewable energy facilities in the Volyn region over three years. A hybrid model of transformed neural networks was developed and trained, combining the advantages of different architectures for processing time series. The main results showed that the proposed system achieved an average absolute error in forecasting wind power generation of 4.2% and solar generation of 3.8% over a horizon of 1 to 6 hours. This is 25% better than standard neural networks and 31% better than statistical models. Testing of the system on a real microgrid in the Volyn region with a capacity of up to 10 MW demonstrated a 36% improvement in the microgrid’s operational stability, a 21% reduction in peak load deviations, a 17.6% reduction in power system balancing costs, and the maintenance of the grid frequency within the regulatory limits (±0.05 Hz). Furthermore, thanks to optimisation, the share of renewable energy sources was increased by 14% without the risk of overloading the equipment. The results obtained demonstrated the high efficiency of hybrid transformable neural networks under the variable weather conditions of western Ukraine. Practical value of results lies in the possibility of directly implementing the developed system into existing microgrids in the Volyn region; furthermore, microgrid operators and energy companies can utilise the hybrid transformer model for short-term forecasting of wind and solar generation.