Earticle

초록

영어

Photovoltaic (PV) inverters are critical components that strongly affect energy yield and power quality in gridconnected PV systems, yet they are continuously exposed to fluctuating meteorological and operating conditions that often lead to latent performance degradation and anomalous events. This paper proposes an explainable anomaly detection framework for PV inverters that combines state-index-based labeling, enhanced meteorological features, and hybrid machine-learning models. First, long-term inverter, basic weather, and derived meteorological data are integrated into an hourly time series, and anomaly labels are defined using thresholds on the Inverter Efficiency Drop and Power Stability Index. The resulting dataset contains 20,303 time steps, of which 5.3% are labeled as anomalous. Second, point level anomalies are detected using an XGBoost classifier trained on 20 input features excluding the state indices, achieving 0.915 test accuracy and 0.9045 ROCAUC. Third, sequence-level anomalies over 24-hour windows are identified using an LSTM Autoencoder based on reconstruction error. Finally, feature-importance analysis and reconstruction-error distributions reveal that enhanced meteorological features, particularly Effective Solar Irradiance, play a key role in detecting inverter anomalies. The proposed framework provides both accurate and interpretable anomaly detection, supporting practical early warning and maintenance decisions for PV inverter operation.

목차

Abstract
1. INTRODUCTION
2. Preprocessing and State-Index-Based Anomaly Labeling
3. PROPOSED FRAMEWORK AND EXPERIMENTAL RESULTS
3.1 Overall Framework and Input Features
3.2 Dataset Configuration and Evaluation Protocol
3.3 Point-level Detection with XGBoost
3.4 Sequence-level Detection with LSTM Autoencoder
3.5 Comparative Analysis and Result
4. Conclusion
ACKNOWLEDGEMENT
REFERENCES

키워드

저자

• Sang-Bum Kim [ Assistant Prof., Department of Robotdrone Engineering, Honam University, Gwangju, Korea ]
• Kyu-Ha Kim [ Assistant Prof., Industry Academic Cooperation Foundation, Honam University, Gwangju, Korea ]
• Sang-Hyun Lee [ Associate Prof., Department of Computer Engineering, Honam University, Gwangju, Korea ] Corresponding Author

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  International Journal of Advanced Culture Technology(IJACT)

• 간기

  계간

• pISSN

  2288-7202

• eISSN

  2288-7318

• 수록기간

  2013~2025

• 등재여부

  KCI 등재

• 십진분류

  KDC 600 DDC 700