A data-driven method for fault diagnosis of microgrid inverters and switches using single-point measurements and feature engineering

Abstract

Accurate and efficient fault diagnosis of microgrid components, particularly inverters and switches, is critical for ensuring system reliability and operational stability. Traditional diagnostic methods typically depend on complex multi-point measurements, resulting in increased system complexity, higher maintenance costs, and reduced practicality under dynamic operating conditions. To overcome these limitations, this study proposes a novel data-driven hierarchical fault diagnosis approach for microgrids by combining single-point electrical measurements with advanced feature engineering techniques. The primary contributions of this research include:

(1) Establishing the theoretical viability and practical effectiveness of using single-point measurements (voltage, active power, and reactive power) for fault diagnosis in microgrid inverters and switches, significantly simplifying instrumentation and reducing deployment complexity.

(2) Developing an advanced feature engineering approach integrating physical power theory with statistical analysis, enhancing diagnostic robustness under noisy and transient operating conditions.

(3) Constructing a hierarchical two-stage diagnostic architecture, where the first stage employs an XGBoost classifier for rapid fault detection in inverter components, followed by a second stage utilising an ensemble model comprising XGBoost, LightGBM, and CatBoost classifiers for detailed multi-class fault identification of switches.

Simulation studies validate that the proposed approach achieves superior accuracy, robustness, and computational efficiency compared to conventional diagnostic methods. It provides a practical, scalable, cost-effective solution that significantly enhances fault diagnosis capabilities in modern microgrid applications.