Improvement of Doernenburg Ratio Method Dissolved-gas Analysis using 5-fold Cross Validation Artificial Neural Network

Abstract

Dissolved-gas Analysis (DGA) is the common method in determining fault occurring in a
transformer. Fault is identified based on the composition of dissolved gases. This paper discusses
the improvement of conventional DGA interpretation method using 5-fold cross validation
Artificial Neural Network (ANN). Conventional ANN requires a large number of data, while
collecting data requires amount of times. This can be overcome using cross validation method.
data is divided into five groups. Four of which are selected as train data, while the other one is
used as test data. These processes are repeated until all groups are used as test data. Using cross
validation, a valid and consistent decision can be made using limited amount of data.