Fault localisation and diagnosis in transmission networks based on robust deep Gabor convolutional neural network and PMU measurements

In this study, the authors aim to develop a real-time and comprehensive fault detection and localisation method in the transmission networks in wide-area power systems and a two-stage process. In the first stage, the fault and faulty line are diagnosed and then the fault location is estimated accurately. For this purpose, the authors design a robust deep Gabor filter convolutional neural network (RDGCNN) to understand the features of the complex and non-linear signals during fault occurrence directly from the raw measured signals by the phasor measurement units (PMUs), on the basis of which a structure for fault detection, faulty line classification, and fault location estimation. The modulated Gabor-based convolutional layers are able to capture temporal features as well as enhance understanding spatial features with fewer parameters. Furthermore, to enhance the robustness in the noisy condition, which is unavoidable in modern power systems, a new loss function is reformulated. The performance of the proposed RDGCNN is examined in the IEEE 68-bus system in high noisy condition and the comparative results demonstrate that the designed robust deep network is fast, accurate, and reliable in the fault detection, faulty line classifying, and fault occurrence position.