Under high-penetration distributed generation (DG) integration, distribution networks exhibit ambiguous fault characteristics, insufficient information utilization, and poor self-healing robustness. This paper proposes a multi-source heterogeneous data-driven fault diagnosis and self-healing control algorithm. First, a DTW-matrix model fusion framework is established. Based on Dynamic Time Warping (DTW), it achieves time alignment and feature-level fusion of multi-source heterogeneous data from SCADA, μPMU, TCM, and other sources, generating a unified spatiotemporal scale fault feature matrix. Second, a hybrid diagnostic model combining “deep learning and rule-based inference” was designed. It utilizes LSTM to extract time-series features, which are then validated and refined against SCADA and TCM event rule libraries to achieve highly accurate and interpretable fault localization. Finally, an interval-based self-healing control strategy is proposed, modeling DG output uncertainty as intervals. A two-stage robust optimization model dynamically generates islanding partitioning and load restoration plans, ensuring system stability under worst-case scenarios. Experimental results on an enhanced IEEE 33-node system demonstrate that the proposed algorithm achieves a fault diagnosis accuracy of 98.5% and a recall rate of 99.0%. Its self-healing control exhibits robustness exceeding 98% under DG fluctuation scenarios, significantly outperforming traditional methods and single data-driven approaches. This effectively enhances the fault response capability and power restoration reliability of high-penetration distribution networks.
Lin et al. (Sun,) studied this question.