Abstract Although the radial basis function neural network (RBFNN) offers a potential solution for fault section estimation (FSE) inpower networks, it has to be totally retrained for the case of powernetwork topology change or power network expansion and cannotprovide any explanations for its diagnosis results due to the blackboxnature of the neural network. In this paper, the functionalequivalence between RBF NN and fuzzy system (FS) is built up forFSE problem throughout the neural network training process.Furthermore, based on this point, a novel retraining strategy ispresented for RBF NN, which can extract the unchanged knowledgefrom the original RBF NN and then insert the knowledge back to thenew RBF NN about the changing part of the power network in thecase of network topology change or expansion. The retrainingstrategy has been implemented and tested in a 4-bus power system.The simulation results show that the advanced FSE system withhybrid FS and RBF NN works successfully and efficiently in powernetworks.
展开▼