Hybrid deep learning-driven explainable AI framework for fault detection and classification in smart power grids

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Power Systems

Background:

• Stable operation and reduced downtime are critical for smart power grids.
• Accurate and interpretable fault detection is essential for grid reliability.
• Existing methods may lack accuracy or interpretability in complex grid environments.

Purpose of the Study:

• To propose a hybrid deep learning model for accurate and interpretable fault detection in smart power grids.
• To integrate Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) with Explainable Artificial Intelligence (XAI).
• To capture spatio-temporal dynamics of electrical signatures for robust fault classification.

Main Methods:

• Developed a hybrid CNN-LSTM model incorporating Explainable Artificial Intelligence (XAI) techniques like SHAP and attention mechanisms.
• Utilized real-time sensor data and simulated fault conditions (LG, LL, LLG, three-phase faults).
• Evaluated model performance on accuracy, F1-score, robustness in noisy environments, and inference latency.

Main Results:

• Achieved high classification accuracy (97.84%) and F1-score (97.08%), surpassing conventional models by over 3%.
• Demonstrated stable performance in noisy environments with <2% decrease and low inference latency (18 ms) for real-time deployment.
• XAI integration successfully identified crucial features for fault prediction, enhancing model interpretability.

Conclusions:

• The proposed hybrid deep learning model offers a powerful, scalable, and interpretable solution for intelligent fault management in smart grids.
• The model's accuracy, robustness, and real-time capabilities make it suitable for contemporary smart grid applications.
• Explainable AI significantly enhances the trustworthiness and practical applicability of fault detection systems.