Artificial Intelligence in Energy Management: Enhancing Electricity Demand Forecasting with Recurrent Neural Networks (RNN)

One of the main challenges in increasing the share of renewable energy in total energy consumption lies in balancing production and distribution to ensure a continuous alignment between availability and demand. The rise of renewable energy sources has led to significant changes in the generation and utilization of electricity, complicating production and supply planning. Accurate electricity demand forecasting has thus become a critical aspect of grid management, enabled operational optimization and enhance decision-making processes related to energy production and distribution. With the growing availability of consumption data, traditional forecasting methods have evolved significantly toward data-driven approaches, particularly through advancements in machine learning (ML). This project focuses on hourly electricity demand forecasting in Italy, utilizing Recurrent Neural Networks (RNNs) and specifically implementing Long Short-Term Memory (LSTM) models. These networks are designed to capture both short- and long-term dependencies in sequential data. Using electricity demand data from Terna - Rete Elettrica Nazionale S.p.A., the LSTM model demonstrated comparable accuracy to Terna's forecasts, effectively predicting hourly demand over both short- and medium-term periods. The model accurately reproduced energy demand peaks, highlighting its robustness and reliability. This study concludes that well-constructed machine learning models offer a promising solution for improving energy demand forecasting accuracy and supporting grid management.