A novel fault classification approach using manifold learning algorithm

Abstract

This paper provides a novel solution for power system transmission line fault classification. It is based on Clarke-Concordia transform and manifold learning algorithm using one-end current signals. We first convert the currents into “α α 0” phases, and construct the trajectories in a 3-diemensional space. Manifold learning is used to extract characteristic features and identify different fault patterns. A weight factor is introduced in the neighborhood selection algorithm in manifold learning, which helps to solve the local nonlinear confusion problem. Fault pattern formed in 3-dimensional “α α 0” space better illustrate the fault's distortion and displacement from the normal state. Simulation results have proven the feasibility of this approach.