Estimation model of transformer iron loss using neural network

When we want to calculate the transformer iron loss in operation, in addition to considering the nonlinear hysteretic phenomenon of transformer itself and the natural unbalanced characteristic, the actual situation that the transformer is operated under three phase unbalance state should also be considered. This results in very non-regular change of the transformer iron loss, and the accuracy of the polynomial model that is commonly used to estimate the iron loss of the transformer in the past is thus reduced. Since neural network has parallel processing capability, which can process highly nonlinear function problem, hence, in this study, we try to use neural network model to set up the nonlinear relationship between the iron loss and voltage of the transformer. Therefore, we can only measure the voltage value to get accurate transformer iron loss. As we compare the neural network model set up in this study to the conventional polynomial method, we can find that neural network model has lower average error rate; this is especially in the prediction of the total transformer iron loss in the three phase balance system, and it is found that the prediction error can be reduced by 50%.