{"title":"基于老化参数的变压器相对老化率预测","authors":"Mohsen Hosseinkhanloo , Navid Taghizadegan Kalantari , Vahid Behjat , Sajad Najafi Ravadanegh","doi":"10.1016/j.array.2025.100433","DOIUrl":null,"url":null,"abstract":"<div><div>This paper presents a novel quantitative approach utilizing machine learning (ML) techniques to predict the aging rate of power transformers based on key aging factors, including loading level, temperature, moisture, and oxygen. Traditional methods for calculating aging rates are limited to discrete values of aging factors, which restricts their applicability in real-world scenarios. Proposed method extends the calculation of transformer aging rate using ML by training data which is achieved by the discrete and limited values obtained in experimental works in laboratories. For this purpose, ML models including Gaussian Process Regression (GPR), Support Vector Machines (SVM), Neural Networks (NN), Fine Tree (FT), Linear Regression (LR), Kernel and Ensemble are utilized to expand aging rate calculation. Values of metrics indicate that GPR had the best accuracy (RMSE = 0.055, R<sup>2</sup> = 1, MSE = 0.003, MAE = 0.028 and MAPE = 7 %). On the contrary, LR had the worst accuracy considering the values RMSE = 3.462, R<sup>2</sup> = 0.57, MSE = 11.986, MAE = 2.36 and MAPE = 2196 %. Taking prediction speed into account, NN had the higher value of 5800 (obs/sec), while GPR had the value of 2600 (obs/sec). Moreover, training time was lower in FT and LR (30.3 s and 34.8 s respectively) compared to other models with higher accuracy.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"27 ","pages":"Article 100433"},"PeriodicalIF":4.5000,"publicationDate":"2025-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Utilizing machine learning for transformer relative aging rate prediction based on aging parameters\",\"authors\":\"Mohsen Hosseinkhanloo , Navid Taghizadegan Kalantari , Vahid Behjat , Sajad Najafi Ravadanegh\",\"doi\":\"10.1016/j.array.2025.100433\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>This paper presents a novel quantitative approach utilizing machine learning (ML) techniques to predict the aging rate of power transformers based on key aging factors, including loading level, temperature, moisture, and oxygen. Traditional methods for calculating aging rates are limited to discrete values of aging factors, which restricts their applicability in real-world scenarios. Proposed method extends the calculation of transformer aging rate using ML by training data which is achieved by the discrete and limited values obtained in experimental works in laboratories. For this purpose, ML models including Gaussian Process Regression (GPR), Support Vector Machines (SVM), Neural Networks (NN), Fine Tree (FT), Linear Regression (LR), Kernel and Ensemble are utilized to expand aging rate calculation. Values of metrics indicate that GPR had the best accuracy (RMSE = 0.055, R<sup>2</sup> = 1, MSE = 0.003, MAE = 0.028 and MAPE = 7 %). On the contrary, LR had the worst accuracy considering the values RMSE = 3.462, R<sup>2</sup> = 0.57, MSE = 11.986, MAE = 2.36 and MAPE = 2196 %. Taking prediction speed into account, NN had the higher value of 5800 (obs/sec), while GPR had the value of 2600 (obs/sec). Moreover, training time was lower in FT and LR (30.3 s and 34.8 s respectively) compared to other models with higher accuracy.</div></div>\",\"PeriodicalId\":8417,\"journal\":{\"name\":\"Array\",\"volume\":\"27 \",\"pages\":\"Article 100433\"},\"PeriodicalIF\":4.5000,\"publicationDate\":\"2025-06-14\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Array\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S2590005625000608\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"COMPUTER SCIENCE, THEORY & METHODS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Array","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2590005625000608","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, THEORY & METHODS","Score":null,"Total":0}
Utilizing machine learning for transformer relative aging rate prediction based on aging parameters
This paper presents a novel quantitative approach utilizing machine learning (ML) techniques to predict the aging rate of power transformers based on key aging factors, including loading level, temperature, moisture, and oxygen. Traditional methods for calculating aging rates are limited to discrete values of aging factors, which restricts their applicability in real-world scenarios. Proposed method extends the calculation of transformer aging rate using ML by training data which is achieved by the discrete and limited values obtained in experimental works in laboratories. For this purpose, ML models including Gaussian Process Regression (GPR), Support Vector Machines (SVM), Neural Networks (NN), Fine Tree (FT), Linear Regression (LR), Kernel and Ensemble are utilized to expand aging rate calculation. Values of metrics indicate that GPR had the best accuracy (RMSE = 0.055, R2 = 1, MSE = 0.003, MAE = 0.028 and MAPE = 7 %). On the contrary, LR had the worst accuracy considering the values RMSE = 3.462, R2 = 0.57, MSE = 11.986, MAE = 2.36 and MAPE = 2196 %. Taking prediction speed into account, NN had the higher value of 5800 (obs/sec), while GPR had the value of 2600 (obs/sec). Moreover, training time was lower in FT and LR (30.3 s and 34.8 s respectively) compared to other models with higher accuracy.