کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Investigation of Statistical and Mathematical Indices for Transformer Frequency Response Analysis and Winding Condition Diagnosis Using Machine Learning Methods

نویسندگان :

Ali Ghorbani¹ Hossein Parsadust² Ali Khani³ Mohammad Molaei⁴

1- شرکت برق منطقه ای خراسان 2- شرکت برق منطقه ای خراسان 3- شرکت برق منطقه ای خراسان 4- شرکت برق منطقه ای خراسان

کلمات کلیدی :

Power transformer،Frequency Response Analysis (FRA)،Statistical indices،Machine learning

چکیده :

Power transformers are among the most expensive and critical components in the power industry. Therefore, to enhance the reliability of the power network, it is necessary to continuously monitor the transformer condition from the moment of installation in the substation, in order to prevent imposing significant costs on the network. Various methods exist for transformer monitoring, one of which is Frequency Response Analysis (FRA), which can be used to detect physical changes and insulation issues. In this study, eight statistical and mathematical indices used for analyzing differences between two FRA curves were evaluated and ranked based on features such as uniformity, sensitivity to amplitude shifts, and suitable variation ranges. Four of these indices were ultimately selected as the most reliable. Subsequently, machine learning methods were employed for winding condition assessment; appropriate features were first extracted based on the selected indices, and then the well-known Support Vector Machine (SVM) classification algorithm was applied. All results were obtained from real FRA measurements of transformers across the country, demonstrating the high accuracy of the proposed approach.