Development of No-Load Noise Power Transformer Model using Back Propagation Neural Network

Power transformer noise contributed to environmental noise pollution. When a power transformer is connected to a voltage source and not connected to a load, noload noise will emerge. No-load noise has a dominant portion of the total noise of the power transformer. Therefore, knowledge of no-load noise from the design stage is very necessary. No-load noise must be predicted since the design stage so that it will not exceed the predetermined limit. This research develops a noload noise power transformer model using a backpropagation neural network. No-load noise is modeled by inputting data in the form of power rated, magnetic flux density, and geometry factor. The geometry factor is the logarithm of the ratio between core weight and core cross-sectional area multiplied by the height of the core, added to the number of legs. The method employed to predict the no-load noise was a back-propagation neural network with 4 hidden layers and 5 neurons in each layer. The modeling results showed that no-load noise prediction results provide better accuracy compared to multiple linear regression method.