Entropy-guided ensemble Kalman filtering with synthetic feature reliability for resilient power quality event forecasting in smart distribution systems
This study introduces a robust hybrid predictive framework for early detection and classification of power quality (PQ) events in carbon-aware smart distribution systems. The proposed approach synergistically combines the Ensemble Kalman Filter (EnKF) with a reliability-aware synthetic data integration strategy, where the contribution of each synthetic sample is adaptively weighted using entropy-derived confidence metrics. This entropy-guided data fusion enhances robustness in environments characterized by incomplete, noisy, or scarce PQ observations. The framework begins with the extraction of discriminative time-frequency representations using the Adaptive Q-Factor Wavelet Transform (AQWT), enabling precise localization of transient disturbances. Subsequently, EnKF-based temporal estimation compensates for missing or corrupted features, while the entropy-calibrated weighting ensures that synthetic augmentation contributes constructively during state updates. A multi-class classifier, trained on the temporally smoothed and reliability-scored features, enables accurate PQ disturbance categorization and proactive event forecasting. Evaluations on publicly available PQ datasets and challenging noise scenarios demonstrate superior classification accuracy, early warning capability, and resilience under limited ground-truth conditions, outperforming classical data fusion and standalone Kalman filter models. The framework is scalable and well-suited for edge-based deployment in modern smart grid monitoring infrastructures.