Unsupervised Temporal Clustering of Solar Power Plants using K-Means and DBSCAN for Production Pattern Recognition

Arellanos-Tafur E. · 2026 IEEE 15th International Conference on Communication Systems and Network Technologies, CSNT 2026 (2026) · DOI: 10.1109/csnt69054.2026.11502417

Resumen

This research addresses operational fragmentation within Peru's solar energy portfolio through the application of advanced unsupervised learning methodologies. The analysis encompassed 1,096 consecutive daily production measurements (2022-2024) from eight major Peruvian photovoltaic installations: Colca Solar, Engie, Majes, Moquegua Solar, Orygen Peru (Rubí), Panamericana Solar, Repartición, and Tacna Solar. The analytical framework integrated descriptive statistics, K-Means partitional clustering, and DBSCAN density-based algorithms to characterize operational behaviors and detect performance anomalies. Three distinct capacity-differentiated clusters emerged: Cluster 0 encompassing six facilities (mean output 114.78 MW), Cluster 1 containing one high-capacity plant (mean output 1229.53 MW), and Cluster 2 comprising one intermediate facility (mean output 300.93 MW). K-Means clustering demonstrated internal validity with a silhouette coefficient of 0.533. DBSCAN identified minimal anomalous behavior, detecting aberrant performance in only 0.09% of operational days across Majes and Moquegua Solar facilities. Seasonal analysis revealed spring-to-winter production variance reaching 18.5% for Cluster 0 installations. This research represents one of the first comprehensive applications of temporal clustering in Peru's solar sector, establishing quantitative foundations for operational improvement, predictive maintenance scheduling, and evidence-based renewable energy policy development.

Instituciones: Universidad Continental