부산대학교 도서관

Detecting and identifying faults is essential for protection systems to minimize downtime and prevent cascading failures. This work proposes a real-time deep learning-based fault diagnosis framework to detect the fault and identify the fault type information. Ground faults, high impedance faults and downed conductors are identified. The framework comprises dynamic state estimation-based fault detection and supervised learning for fault classification. The identified fault is utilized in a time domain dynamic state estimation to estimate the fault location. The fault locating method is not addressed in this paper. The fault classification method uses a 1D convolutional neural network (1D-CNN) for offline training and online classification. With the sample data obtained from merging units, this framework can automatically run dynamic state estimation algorithm to detect the internal fault in a protection zone. After that, the faulted samples are fed into the CNN model to output the fault type. The method is demonstrated with three events which include different types of faults, including downed conductors. The results have demonstrated the capability of the proposed framework to detect faults and identify the fault type and faulted phase.