Prediction of Stress in Power Transformer Winding Conductors Using Artificial Neural Networks: Hyperparameter Analysis

Fausto Valencia,Franklin Quilumba,Hugo Arcos

doi:10.3390/en14144242

Fausto Valencia, Franklin Quilumba + Show 1 more

Open Access

https://doi.org/10.3390/en14144242

Copy DOI

Journal: Energies	Publication Date: Jul 14, 2021
Citations: 4	License type: CC BY 4.0

Affiliation: National Polytechnic School

Abstract

The purpose of this research is the evaluation of artificial neural network models in the prediction of stresses in a 400 MVA power transformer winding conductor caused by the circulation of fault currents. The models were compared considering the training, validation, and test data errors’ behavior. Different combinations of hyperparameters were analyzed based on the variation of architectures, optimizers, and activation functions. The data for the process was created from finite element simulations performed in the FEMM software. The design of the Artificial Neural Network was performed using the Keras framework. As a result, a model with one hidden layer was the best suited architecture for the problem at hand, with the optimizer Adam and the activation function ReLU. The final Artificial Neural Network model predictions were compared with the Finite Element Method results, showing good agreement but with a much shorter solution time.

Highlights

The model is applied to the transformer where the input data are the current through the windings in amperes, and the output data are the stresses on the winding’s middle disks in Pascals
The stress is calculated at the points where the currents have an extreme value
For Phase A, the stress is calculated in π and in 2π

Summary

Introduction

FEM needs the discretization of the medium, which results in the creation of nodes, each one represented as one row and one column in a matrix [6]. This process is not unique because FEM internally looks for the discretization that presents a slight field variation between nodes [7]. This is achieved through an iterative process, which lasts longer when the parameters of the problem have a nonlinear behavior—such as the permeability of the transformer core. Even if the dynamical analysis of the same reference is considered, 250 simulations should be performed at most, which correspond to 2 s at 60 Hz, at the highest value of short circuit current and the most pessimistic transient conditions

Objectives

Results

Conclusion