An approach to fault point identification using neocognitron and BP neural network

This paper presents a new approach to identity a fault point in a power system on real time. Based on the topology theory, the characteristics of a fault in the power system are treated as a pattern of the fault. Therefore, the calculation complexity of traditional approaches can be avoided. The aim of the proposed method is to identify fault points through directly analyzing the type of fault. Since each type of fault has each characteristic pattern of power flow, fault points can be identified by abstracting characteristics of power flow at each node of the power system. In order to abstract fault characteristics, neocognitron in which symmetrical three phase decomposition and data normalization are calculated using power flow at each node is introduced. Thus, the impacts of voltage grade and unbalanced load can be removed. Since each neocognitron corresponds to one node of the power system, hierarchical autonomous decentralization can be realized. Therefore, the proposed approach can be applied to a large power system. Fault point location is done by BP network. Since the neuro of BP network only corresponds to the nodes of the topologized power system, the training of the neural network can be performed independently. From this point of view, the applicability and flexibility of proposed approach are high. The effectiveness and applicability of the proposed approach are demonstrated on a simple power system model.