Explainable deep learning framework for fault detection and diagnosis in smart power grids

초록

Reliable fault detection and diagnosis are critical for ensuring the stability, safety, and efficiency of modern power systems. With the increasing integration of renewable energy sources and the growing complexity of smart grids, traditional rule-based and model-driven approaches often struggle to capture nonlinear dynamics and rapidly evolving system conditions. Although deep learning techniques have demonstrated superior performance in fault detection tasks, their black-box nature limits their adoption in safety-critical power system operations. In this study, an explainable deep learning framework proposed for fault detection and diagnosis in smart power grids. The proposed approach integrates a hybrid temporal deep learning model with explainable artificial intelligence (XAI) techniques to provide both high predictive accuracy and interpretable decision-making. Time-series data from supervisory control and data acquisition (SCADA) systems and phasor measurement units (PMUs) are used to capture system dynamics under normal and fault conditions. A Long Short-Term Memory (LSTM)-based architecture is employed to model temporal dependencies, while SHapley Additive exPlanations (SHAP) are utilized to quantify the contribution of each feature to the model predictions. The framework enables not only accurate detection of multiple fault types but also provides transparent insights into the underlying causes of faults by highlighting critical system variables. Experimental results demonstrate that the proposed model outperforms conventional machine learning and deep learning baselines in terms of detection accuracy and robustness. Furthermore, the integration of XAI significantly enhances interpretability, supporting operator trust and facilitating real-time decision-making. Overall, this work presents a reliable and interpretable solution for intelligent fault diagnosis in power systems, contributing to the advancement of trustworthy AI-driven smart grid operations.

키워드