Development of a Hybrid Framework for Knowledge Discovery in Smart Grid Data From Underground Substations

Smart Grids rely on extensive monitoring of operational data to ensure reliable and efficient energy distribution. Underground substations, widely used in dense urban areas, pose unique challenges due to high ambient temperatures, constrained spaces, and noisy or irregular data, which limit the effectiveness of conventional data-driven analytics. This study proposes and validates a hybrid Knowledge Discovery in Databases (KDD) framework specifically designed to analyze operational data from underground substations. Real-world data from 43 substations managed by CEEE Equatorial in Porto Alegre, Brazil, were preprocessed and analyzed using the Expectation-Maximization (EM) algorithm to uncover latent operational patterns and the Apriori algorithm to extract association rules that explain and validate these patterns. The framework integrates rigorous data cleaning, normalization, discretization, and interpretable clustering and rule-mining techniques implemented in R and WEKA. Experimental results identified meaningful clusters representing distinct operational regimes, including a critical anomalous cluster characterized by high temperature combined with low-voltage and low-current conditions. Association rules reinforced the interpretability of these clusters and highlighted operational anomalies with high confidence and support. The proposed framework demonstrates practical relevance for intelligent monitoring, anomaly detection, and decision support in Smart Grid systems. Future work includes temporal modeling, incorporation of external contextual data, and the use of hybrid or deep learning approaches to enhance model scalability and diagnostic accuracy.