Identify incipient faults through similarity comparison with waveform split-recognition framework

Introduction: Incipient faults of distribution networks, if not detected at an early stage, could evolve into permanent and result in significant economic losses. It is necessary to detect incipient improve power grid security. However, due the short duration unapparent waveform distortion, are difficult identify. In addition, usually have a small data volume, which compromises their pattern recognition. Methods: this paper, fault identification method proposed address these problems. First, Waveform Split-Recognition Framework (WSRF) provide two-step process: 1) split several segments according cycles, 2) recognize through similarity decomposed segments. Second, we design Similarity Comparison Network (SCN) learn by sharing weights two Convolution Neural Networks (CNNs), then calculate gap between them non-linear function high-dimensional space. Last, disassembled filters devised extract features from waveform. Results: The initializing can speed Accuracy training, some existing datasets like MNIST consisting 250 handwritten numbers different people able initial adaptive method. This paper uses field simulation verify performance SCN WSRF. Discussion: WSRF achieve more than 95% identifying faults, much higher three other methods literature. And good results locations times. Which