A0316
Title: Electricity price forecasting with LSTM: Evidence from the Italian day-ahead power market
Authors: Filippo Beltrami - Eurac Research (Italy) [presenting]
Luigi Grossi - University of Parma (Italy)
Patrick Thoeni - Eurac Research (Italy)
Riccardo Lucarno - Eurac Research (Italy)
Abstract: The aim is to present a comprehensive forecasting framework for electricity prices in the Italian day-ahead power market, leveraging both traditional time-series models and advanced artificial intelligence (AI) techniques. The focus is on a novel, high-resolution proprietary dataset which has not yet been used in the existing literature, spanning from 2021 to 2024 at 15-minute intervals and encompassing, among other variables, market volumes, zonal prices, detailed weather metrics (e.g., temperature), cross-border exchanges with Switzerland, France, and Austria, as well as plant maintenance schedules. The modeling pipeline systematically implements a suite of long short-term memory (LSTM) architectures on this uniquely rich and granular dataset. Each model is rigorously benchmarked against established econometric and machine-learning references, namely a refined lasso-estimated autoregressive (LEAR) model and a two-layer DNN. Model performance is evaluated using a comprehensive set of metrics (rMAE, MAE, MAPE, sMAPE, RMSE) alongside measures of computational effort, with the goal of identifying the optimal trade-off between forecasting accuracy and computational cost and thus aligning cutting-edge methodological advances with the practical performance requirements of industry stakeholders.
