Coreloss Estimation Via Long Short-Term Memory Model (Lstm) of Dry-Type Transformer Based on Fea

Abstract

Accurate estimation of losses is very important in transformer designs for energy systems. Therefore, in this study, a long short-term memory model (LSTM) was performed to predict the core loss of three-phase dry-type transformers based on Finite Element Analysis (FEA) analysis. Since, in ordinary multilayer networks, learning problems occur when the gradient value gets too small during backpropagation. LSTM, on the other hand, can store information better thanks to its extra layers that communicate. Thus, the learning process takes place more efficiently. The analysis and estimation processes were performed using a primary number of turns, excitation voltage, and three different cross-section area parameters. 486 data randomly selected from 506 data obtained by ANSYS/MAXWELL in the training of the LSTM model were used. The remaining 20 data were used in the testing process to measure system performance. The error obtained by the validation test is 0.15. It is very close to the simulated value, thus LSTM can be used as a reliable estimation method during the design stage.