Application of DA-Preconditioned RBFN with Global Structure to Power System Fault Detection

Abstract

In this paper, a hybrid method of data precondition techniques and an artificial neural network (ANN) is proposed to deal with fault detection in power systems. The proposed method makes use of FFT and DA clustering as a precondition technique. FFT is used to extract features of fault currents so that faults to be studied are characterized by frequency domain. DA clustering classifies input data into clusters in a sense of global clustering. DA contributes to the universal clustering that is not affected by the initial conditions. For each cluster, an ANN model is constructed to estimate the location and the type of fault. As ANN, this paper focuses on RBFN (Radial Basis Function Network) due to the better nonlinear approximation. DA clustering is also proposed to determine the centers of RBFN appropriately. Thus, the RBFN model results in one with global structure. The proposed method is successfully applied to a sample system. A comparison is made between the proposed and the conventional methods.

Reconstructing clusters for preconditioned short‐term load forecasting

“…DA is known for its high accuracy and its independence of the initial values. Thus, when determining the RBF center, DA offers more accurate output than k-means [17].…”