Transformers are indispensable elements of all sectors related to electricity generation, transmission and distribution. In transformer design, leakage fluxes are a very important factor in terms of winding type, core shape, losses, operating performance, operating life and efficiency. Voltage and power transformers can be designed in different sizes and powers according to the field to be used and the condition of the load. In power transformers, especially in small and medium power, two-layer winding type is generally used due to advantages such as not occupying much space and ease of insulation. While the transformer is being designed, estimating the leakages by considering the other parameters of the transformer can help to come up with the most optimal design. Transformers can be easily designed with different power and dimensions by means of classical package programs used for design. Leakage flux is usually obtained by calculation after design and use. Therefore, machine learning methods that allow predicting losses and leakages can offer opportunities to the designer in this sense. In this study, the most popular machine learning (ML) methods, neural networks (NN), decision trees (DT), nearest neighbor (KNN), naive Bayesian (NB) and support vector machines (SVM) methods were applied to detect leakage flux, and these methods were compared with each other in terms of their results. As a result of the obtained accuracy rates, it was seen that the NN method came to the fore were able to produce very successful results in transformer leakage flux detection.
Anahtar Kelimeler: Transformers, Leakages, ML, KNN, SVM, NB, NN
|
