Explainable Deep Learning for Voltage Disturbance Classification in Microgrid AC Bus Using LSTM and SHAP

Abstract

The present paper introduces an explainable deep learning framework for the classification of voltage disturbances in microgrid AC buses. This framework employs Long ShortTerm Memory (LSTM) networks in combination with SHapley Additive exPlanations (SHAP). The proposed model processes raw time-series voltage and current data to detect and classify 8 types of power quality events such as harmonics distortions, symmetrical dip and othes, without the need for manual feature extraction. The integration of SHAP values permits the extraction of interpretable information from the model’s predictions, thus ensuring the transparency of decision making processes. The experimental results demonstrate a high level of accuracy (F1-score $=0.99$) for all categories and reveal the features that have the most significant impact on each disturbance type. A comparative analysis demonstrates that the proposed method surpasses existing techniques in terms of both classification performance and interpretability, making it suitable for real-time implementation in the context of smart grid monitoring and diagnostics.