Artificial Intelligence for Energy Sustainability: Predicting Photovoltaic Generation in P2P Networks

Abstract

The incorporation of renewable sources into electrical systems is essential for advancing toward sustainable energy models. However, intermittency and meteorological variability introduce uncertainty into supply operations and planning, necessitating reliable predictive tools. In this context, artificial intelligence (AI) offers alternatives for modeling generation behavior and supporting decision-making. This paper presents an AI-based knowledge discovery (KDD) model for predicting electricity generation in a peer-to-peer (P2P) network. The methodology was structured in four phases: exploration, analysis, construction, and validation. LSTM–CNN and Transformer deep learning architectures were implemented and trained on multivariate time series that integrate electrical, meteorological, and weather variables. The data come from five institutions in the department of Nariño, Colombia. The results show robust performance in both architectures (R² > 0.83), demonstrating the strategic potential of AI in distributed energy systems.