A Transformer Protection Scheme Based on The Deep Forest Algorithm

Abstract

A transformer protection scheme based on the deep forest is proposed to improve the performance of transformer protection. Considering that the magnetization curve can reflect the essential cause of the inrush current, but it is difficult to measure, the voltage and differential current sequences are selected as alternative input of the deep forest. The classification deviation of each layer is defined to automatically determine the depth of the deep forest. The protection acts based on the algorithm output. The deep forest is trained by the simulation sampling data from PSCAD/EMTDC simulation. The performance of the deep forest algorithm is verified through simulation data and dynamic model experimental data. The test results show that the deep forest algorithm can identify the operating states of transformer quickly and reliably. It has good generalization and no requirement of high sampling frequency, and has certain adaptability to CT saturation and over excitation.