Forecasting Electricity Prices in Energy Communities: A Comparative Analysis in the Portuguese Retail Market

Accurate electricity price forecasting is pivotal for selecting optimal retail options, emphasizing the need for precise predictions. This paper provides a comparative analysis of three machine learning techniques—Artificial Neural Network (ANN), K-Nearest Neighbors (KNN), and Long Short-Term Memory (LSTM)—for predicting electricity prices in the Portuguese retail market, with applicability to energy communities, households, or companies. The models were trained on a comprehensive dataset from the Iberian electricity market operator (OMI) spanning from January 2020 to December 2023, but the analysis focuses on forecasts for January to March 2024. We detail the use of [1], [460] past daily lags, a [90-day] forecasting horizon, and specific features for the models. Through a detailed case study, we compare model-generated forecasts with actual prices and OMI forecasts, evaluating various fixed and indexed retail tariffs to assess potential cost savings. Discussion includes implications for energy community management, highlighting the insights despite the limited three-month evaluation period.