Dissolved Gas Analysis of Renewable Energy Generation Transformers using Statistical Machine Learning

Renewable energy has been increasingly part of the available generation on the electric power grid. The availability of the renewable energy generation sources is reliant on the ability of the renewable energy source to continuously and reliably generate energy for the grid. A key equipment in connecting renewable energy sources to the grid are padmount transformers. Padmount transformers used in renewable applications have been subject to unkown failures that do not follow the electrical standards for interpreting transformer condition based on dissolved gas analysis (DGA) samples. This is due to the electrical, mechanical, and thermal impact renewable energy has on padmount transformers that are designed for utility load applications. This thesis uses various statistical machine learning techniques to predict renewable padmount transformer condition and failure based on DGA values. Principle Component Analysis (PCA), Support Vector Machine (SVM), PCA-SVM, and Multilayer Perceptron (MLP) are applied to industry level data to provide accurate transformer condition prediction. A solution is presented for an industry level phenomenon that has not been able to be solved using conventional industry standards and approaches. This thesis reviews the challenge and unsolved issue in the renewable energy industry, and provides in depth research and results to be able to accurately predict renewable padmount transformer condition.