Application of a PCA-MLR Load Noise Prediction Model for The Power Transformer

The main problem with load noise is that it is difficult to accurately quantify the value of the load noise that the transformer will generate. If a transformer’s load noise measurement results are within the standard limit or are not by customer demand, efforts to reduce load noise are challenging. This happens because load noise has not been part of the comprehensive design aspect since the beginning. This research uses principle component analysis and multiple linear regression to model load noise in power transformers. Input variables consist of 18 types that are thought to have a relationship with load noise. The principle component analysis (PCA) process selects variables with an eigenvalue greater than 1. The PCA process of 18 variables gives the results of 5 variables with an eigenvalue of more than one and data representativeness of 92.6%. The selected variable is used as an independent variable in the MLR model. The selection of independent variable combinations is done by looking at the Cp Mallows value. The Cp Mallows value indicates that only three principle component (PC) variables were used to model load noise with multiple linear regression (MLR). The final MLR model results are returned to the original 18 variables. The results showed that modelling with the PCA-MLR method had better results than the Reiplinger method models. The PCA-MLR model has an MSE value of 10.2 and a regression determination of 0.87.