Isolation Forest-Based Anomaly Detection for Solar PV Performance Monitoring: CLEMESÍ Solar Plant Case Study

Arellanos-Tafur E. · 2026 IEEE 15th International Conference on Communication Systems and Network Technologies, CSNT 2026 (2026) · DOI: 10.1109/csnt69054.2026.11502170

Resumen

This study implements the Isolation Forest algorithm for anomaly detection at the CLEMESI solar power plant (115 MW), operated by Orygen Peru S.A.A. in Moquegua, Peru. A total of 731 days of operational data, from May 8, 2023, to May 7, 2025, were analyzed using unsupervised machine learning techniques. The methodology identified 28 significant anomalies (3.83%), with 89.3% concentrated within the first sixty operational days of the commissioning phase. The facility achieved 607,937.49 MWh of total generation, with an average daily production of 831.65 MWh and a capacity factor of 30.13%. Seasonal analysis showed optimal performance during Southern Hemisphere spring, reaching a 37.15% capacity factor, while winter exhibited lower but stable generation at 21.33%. Validation using Z-score and Interquartile Range methods confirmed the robustness of the detected anomalies. Post-commissioning performance demonstrated high stability with near-zero anomaly rates, indicating effective algorithm adaptation. The study validates unsupervised learning for solar monitoring, particularly in newly commissioned plants lacking historical fault data, showing that automated anomaly detection improves operational efficiency through proactive maintenance and reduced downtime.

Instituciones: Universidad Continental