Published November 9, 2026 | Version v1

Real-World Load and Wind Power Forecasting in Greek Island Energy Systems: A Deep Learning Perspective

Description

Accurate short-term forecasting of electricity load and renewable generation is essential for the reliable operation of energy systems. This paper presents a comparative evaluation of classical statistical methods and deep learning models for minute-level load and wind power forecasting using real-world data from the Greek islands of Samos and Ikaria. Autoregressive models (ARIMA, SARIMAX, AutoReg) are benchmarked against Long Short-Term Memory (LSTM) networks and Deep Neural Networks (DNNs). The models are evaluated using both standard one-step-ahead prediction and realistic rolling and recursive multi-step forecasting scenarios. Experimental results show that deep learning approaches significantly outperform statistical methods across all tasks, achieving up to one to two orders of magnitude lower error. Among them, the DNN demonstrates the most stable and robust behavior, especially under long-horizon deployment. The study also highlights the increased difficulty of wind power forecasting compared to load prediction due to strong intermittency and non-stationarity.

Files

Real-World Load and Wind Power Forecasting in Greek Island Energy Systems A Deep Learning Perspective_Camera_Ready.pdf

Additional details

Funding

European Commission
SINNOGENES - STORAGE INNOVATIONS FOR GREEN ENERGY SYSTEMS 101096992