Application of Feature Selection in Distribution Transformer Design and Manufacturing Using Feed Forward Artificial Neural Network and Equilibrium Optimizer Algorithm

In this study, a Feedforward Artificial Neural Network is used as a regression model to predict the dataset collected from 170 distribution transformers with a nominal power of 1000 kVA that have already been designed and manufactured. There is a total of 17 design variables (features) during the design process. The objective of the feature selection in the transformer design problem is to reduce computational complexity and improve the manufacturing process. The Equilibrium Optimization (EO) algorithm is applied to solve the feature selection (FS) problem by minimizing the regression error with fewer numbers of features as compared to the regression with the original dataset. The results of the study reveal that out of the 17 design variables, six features have the highest priority and level of importance in the design process, while six features have less importance and can be set to a constant value.