Forecasting electricity prices in the portuguese market: A machine learning approach

The Portuguese electricity market, integrated within the MIBEL, exhibits high volatility and structural complexity, primarily driven by the growing share of intermittent renewable energy, unpredictable demand patterns, and meteorological variability. These factors pose significant challenges for SMEs, which often lack the analytical tools required to anticipate price fluctuations and manage energy costs effectively. This dissertation proposes a machine learning-based forecasting model tailored to predict hourly electricity prices over a seven-day horizon. The model is specifically designed to support industrial SMEs in improving energy planning and mitigating exposure to price risk. A comprehensive dataset was constructed, comprising hourly observations from March 2020 to March 2025 and incorporating 37 variables across five key dimensions: energy production, consumption, market prices, cross-border exchanges, and weather conditions. The methodological framework combines robust preprocessing techniques, including outlier mitigation, robust normalization, and one-hot encoding, with advanced learning algorithms. LightGBM was selected for its predictive performance and scalability. Hyperparameter tuning was conducted using Bayesian optimization via Optuna. The final model achieved a MAE below 6 €/MWh, in line with industry standards for short-term forecasting. Results underscore the relevance of meteorological factors and cross-border dynamics in shaping market behaviour. This study contributes a practical and interpretable tool that enhances SMEs’ decision-making capacity, while also demonstrating the effectiveness of machine learning methods in navigating the complexities of modern electricity markets.