Fault Identification in Power Distribution Systems based on Domain Adaptation Probabilistic Learning

To improve the fault identification accuracy in power distribution systems, a diagnosis method for incipient faults is proposed based on Domain Adaption Probabilistic Learning. The proposed method finds the best linear mapping by waveform decomposition and feature similarity, which maps simulation data and field data into the same feature space. In this space, the distribution difference between two kinds of features reaches its minimum, allowing the model to use simulation data for training and field data for test. The whole process solves the essential problem of few samples in fault identification. Experiments show that this method is much better than other commonly-used classification models, such as Convolutional Neural Network, Support Vector Machine and K-nearest Neighbors, in the aspects of reliability and accuracy. This research provides a new idea for fault identification in power distribution systems under the condition of domain adaptation.