Fault location in distribution system using convolutional neural network based on domain transformation

Distribution lines are integral parts of the modern power system, which can affect the security and stability of power supply directly. An effective power system protection scheme should be able to detect all occurring faults as soon as possible. There are two tasks in fault diagnosis. One is the fault classification, where high accuracy rates have already achieved. Thus, this paper focuses on the other task, i.e. fault location. Enlightened by Fourier transform, this paper proposes an online data-driven method, which transforms signals from time domain to image domain through signal-to-image (SIG) algorithm and then process the transformed images with framework based on convolutional neural network (CNN). On the one hand, we can extract more crucial characteristic and information from image domain. On the other hand, the CNN-based structure is much smaller than others. It needs less memory space and would be easier to be transplanted to hardware platform. Moreover, the proposed algorithm does not require synchronous devices. The numerical comparison shows that the proposed SIG-CNN fault location model achieves robust and accurate results compared with other data-driven algorithms.