ABSTRACT
This chapter describes two new algorithms for power transformer differential protection. They are dimensionality reduction methods based on the principal component analysis (PCA) and curvilinear component analysis (CCA). The CCA algorithm applies a cost function to unfold closed data surfaces, a distinctive feature in nonlinear problems. PCA aims to reduce a database with many variables to a lower dimensional space while losing as little information as possible. The PCA extracts the characteristic features from transformer differential current waveform during distinct scenarios; these features allow to classify the current signal as inrush transient or short circuit inside the power transformer. The current transformers used on both sides of the power transformer differ in their electrical behavior, have different magnetic saturation characteristics. The main protection used in power transformer is the differential protection, which is based on Kirchhoff's laws of electric circuits.
