基于GAN-DRNN的电力变压器剩余寿命预测
李柯夫1，游欣2，徐椰烃3，钱泓江4
(1 成都双流国际机场股份有限公司，四川 成都 610200； 2 国网四川省电力公司超高压分公司，四川 成都 610095；
3 国网四川省电力公司彭州市供电分公司，四川 彭州 611930； 4 四川大学 空天科学与工程学院，四川 成都 610207)
摘 要 ：大型变压器的可靠性关乎电力系统的稳定运行。由于时间、成本的局限性，致使变压器剩余寿命统计数据规模较小，难以充分发挥机器学习算法的最佳预测性能。提出了一种基于生成对抗网络 (GAN) 的数据增强手段，有效解决变压器剩余寿命样本稀疏问题；构建了具有良好预测性能的动态递归神经网络 (DRNN) 模型，并验证了其高效性。试验结果表明，经GAN作用的增强数据集能有效激发DRNN模型的预测性能，其预测精度最大提高了7.16%，预测结果均在 2.0 倍误差分散带以内，实现了小样本 情形下变压器剩余寿命的高精度预测，较大程度上压缩了变压器剩余寿命预测的时间和成本。
关键词 ： 变压器 ；生成对抗网络 ；数据增强 ；动态递归神经网络 ；剩余寿命预测
中图分类号 ：TM401 ；TM407 文献标识码 ：B 文章编号 ：1007-3175(2025)12-0036-06
Prediction of Remaining Life Power Transformers Based on GAN-DRNN
LI Ke-fu1 , YOU Xin2 , XU Ye-ting3 , QIAN Hong-jiang4
（1 Chengdu Shuangliu International Airport Co., Ltd, Chengdu 610200, China; 2 State Grid Sichuan Electric Power Company Extra-High Voltage Branch Company, Chengdu 610095, China;
3 State Grid Sichuan Electric Power Company Pengzhou Power Supply Branch Company, Pengzhou 611930, China;
4 School of Aerospace Science and Engineering, Sichuan University, Chengdu 610207, China）
Abstract: The reliability of large transformers is closely associated with the stable operation of the power system. Due to the limitations of time and cost, the scale of statistics on the remaining service life of transformers is relatively small, it is difficult to fully leverage the best predictive performance of machine learning algorithms. In order to solve the above problems, this paper proposes a data augmentation means based on generative adversarial network (GAN), which effectively solves the problem of sparse samples of the remaining life of transformers, constructs a dynamic recurrent neural network (DRNN) model with good prediction performance and its high efficiency is verified. The experimental results show that the enhanced dataset treated with GAN can effectively stimulate the prediction performance of the DRNN model. Its prediction accuracy has been increased by up to 7.16%, and the prediction results are all within the error dispersion band of 2.0 times. It has achieved high-precision prediction of the remaining life of transformers in the case of small samples, and has largely compressed the time and cost of predicting the remaining life of transformers.
Key words: transformer; generative adversarial network; data augmentation; dynamic recurrent neural network; prediction of remaining life
参考文献
[1] SAYAKA Kamei, SHARAREH Taghipour.A comparison study of centralized and decentralized federated learning approaches utilizing the transformer architecture for estimating remaining useful life[J].Reliability Engineering & System Safety， 2022，233 ：109130.
[2] ABDUL W, MUHAMMAD Y, JOHN G B, et al.SelfAttention Transformer-Based Architecture forRemaining Useful Life Estimation of Complex Machines[J].Procedia Computer Science，2023， 217 ：456-464.
[3] 栗然，刘会兰，卢云，等 . 基于交叉熵理论的配电变压器寿命组合预测方法 [J]. 电力系统保护与控制， 2014， 42(2) ：97-103.
[4] 于长海 . 基于数据驱动的电力变压器剩余使用寿命预测方法研究 [D]. 保定 ：华北电力大学， 2018.
[5] 林喆，兰生，张宇航 . 基于广义回归神经网络的油纸绝缘变压器的寿命预测 [J]. 高压电器，2015， 51(2) ：125-130.
[6] RAHUL Soni, BHINAL Mehta.Diagnosis and prognosis of incipient faults and insulation status for asset management of power transformer using fuzzy logic controller & fuzzy clustering means[J].Electric Power Systems Research，2023， 220 ：109256.
[7] AIZPURUA J I, STEWART B G, MCARTHUR S D J， et al.Improved power transformer condition monitoring under uncertainty through soft computing and probabilistic health index[J]. Applied Soft Computing，2019，85 ：105530.
[8] COSTA J G, LEITE A M, DA SILVA G A H, et al. Probabilistic evaluation of distribution power transformers reliability indices considering load transfers and mobile unit substations[J].Electric Power Systems Research， 2020， 187 ：106501.
[9] NAIMUL I, RIAZ K, SAJAL K D, et al. Power transformer health condition evaluation: A deep generative model aided intelligent framework[J].Electric Power Systems Research， 2023， 218 ：109201.
[10] 赵洪山， 常杰英， 曲岳晗， 等 . 基于二元非线性 Wiener 随机过程的变压器油纸绝缘剩余寿命预测方法 [J]. 电 工技术学报，2023，38(15) ：4040-4049.
[11] BOKANG Agripa Tlhabologo, RAVI Samikannu, MODISA Mosalaosi.Experimental characterisation and life data analysis of superior mixed liquid dielectrics for power transformer application[J]. Scientific African， 2022， 18 ：e01424.
[12] RAHUL Soni, BHINAL Mehta.Evaluation of power transformer health analysis by internal fault criticalities to prevent premature failure using statistical data analytics approach[J]. Engineering Failure Analysis， 2022， 136 ：106213.
[13] 刘文泽，张俊，邓焱 . 基于深度置信网络和多维信息融合的变压器故障诊断方法 [J]. 电力工程技术， 2019，38(6) ：16-23.
[14] VIMAL Angela Thiviyanathan, PIN Jern Ker, YANG Sing Leong, et al.Power transformer insulation system:A review on the reactions， fault detection， challenges and future prospects[J]. Alexandria Engineering Journal， 2022， 61(10) ： 7697-7713.
[15] 林朝晖，张彼德 .利用糠醛含量进行变压器寿命预测 [J]. 变压器， 2009， 46(7) ：35-38.
[16] 蔡红飞 . 电力变压器剩余技术寿命的综合评估 [D]. 郑 州 ：郑州大学，2012.
[17] ABBASI Ali Reza.Fault detection and diagnosis in power transformers: A comprehensive review and classification of publications and methods[J]. Electric Power Systems Research， 2022， 209 ： 107990.
[18] 李振柱， 谢志成， 熊卫红， 等 . 考虑绝缘剩余寿命的变压器健康状态评估方法 [J]. 电力自动化设备， 2016，36(8) ：137-142.
[19] CHEN Qiyu, CUI Zhesi, LIU Gang, et al.Deep convolutional generative adversarial networks for modeling complex hydrological structures in Monte-Carlo simulation[J].Journal of Hydrology， 2022， 610 ：127970.
[20] VASUDEVAN V, HUANG C, SIMIELE E, et al.Combining Monte Carlo with Deep Learning:Predicting High-Resolution, Low-Noise Dose Distributions Using a Generative Adversarial Network for Fast and Precise Monte Carlo Simulations[J]. International Journal of Radiation Oncology, Biology, Physics，2020，108(3) ：S44-S45.
[21] OLIVEIRA F M F, DANIEL G, LIMOUSIN O.Artificial Gamma Ray Spectra Simulation Using Generative Adversarial Networks(GANs) and Supervised Generative Networks(SGNs)[J].Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment，2023，1047 ：167795.
[22] YANG Jiashu, CHEN Jianbing, JENSEN Hector. Structural design optimization under dynamicreliability constraints based on the probability density evolution method and highly-efficient sensitivity analysis[J].Probabilistic Engineering Mechanics，2022，68 ：103205.
[23] 钱泓江，董秀成，张征宇，等 . 基于Elman的荧光油膜厚度近效测量 [J]. 航空动力学报，2023，38(3) ： 558-568.
[24] SIMA Wenxia, ZHANG Han, YANG Ming, et al. Diagnosis of small-sample measured electromagnetic transients in power system using DRN-LSTM and data augmentation[J].International Journal of Electrical Power&Energy Systems， 2022， 137 ：107820.
