Fault section estimation in electric power distribution system with Elman neural network

In this article is investigated 6 kV distribution system in large pumping station. Novel method for patterns development with fault trees modeling is presented too. Minimal paths are used for representing of normal working conditions and successful fault clearance. Neural network with back propagation of error (BPNN), neural network with radial basis transfer function (RBNN) and Elman recurrent neural network (Elman NN) behaviors are investigated for finding faulted sections using data from circuit breakers and relay protections. During education and simulations was used big data set, consisting 7225 patterns. For experiments are prepared sets of input vectors, consisting condition of circuit breakers and corresponding relay protections, which are modified input patterns with added Hamming distance as a disturbance. Modeling of missing data (Hamming distance on 1), overlapping of cascading events (Hamming distance on 0), and random case of distribution system condition influence to SCADA system operation (Random Hamming distance on 0 or 1) was tested with BPNN, RBNN and Elman NN. Results are showing that Elman NN can be used for fault section estimation (FSE).