Machine Learning-Based Forecasting of the Automatic Frequency Restoration Reserve Demand

Due to decarbonization, significant changes are occurring in the energy sector. One of these changes is the separation of standard power delivery and the provision of ancillary services, where small hybrid sources could become service providers. The information that can improve control of these resources providing this service is an accurate forecast of regulatory power demand. This paper proposes a novel forecasting model predicting automatic Frequency Restoration Reserve demand. Our approach is based on convolutional neural networks (CNN) and it outperforms traditional forecasting methods applied to this problem. Our experiments, which were conducted on real-world data, demonstrate the effectiveness of our approach. These findings have significant implications for controlling hybrid energy resources and may lead to improved performance and efficiency in the energy sector.