Genetic programming for partial discharge feature construction in large generator diagnosis

In this paper, the standpoint of feature construction is employed into partial discharge defects identification of large generators by another emerging simulated evolution technique- genetic programming (GP). Genetic programming can discover relationships among observed data and express them mathematically. The architecture of partial discharge feature construction is proposed. GP is applied to extract and construct effective features from raw dataset. In addition, in order to eliminate the bottleneck of insufficient sample size, a kind of statistical resampling technique called bootstrap is incorporated as a preprocessing step into genetic programming. The experimental results show the good ability in partial discharge defects identification.