Application of Gaussian Process Regression on the Quantitative Analysis of the Aging Condition of Insulating Paper by Near-Infrared Spectroscopy
LI Yuan1, ZHANG Wen-bo1, CHEN Xiao-lin2, 3, LI Han1, ZHANG Guan-jun1
1. State Key Laboratory of Electrical Insulation and Power Equipment, Xi’an Jiaotong University, Xi’an 710049, China
2. Electric Power Research Institute of Hainan Power Grid Co., Ltd., Haikou 570125, China
3. Key Laboratory of Physical and Chemical Analysis for Electric Power of Hainan Province, Haikou 570125, China
Abstract:As the aging condition of the insulating papers determines the remaining lifetime of the oil-immersed transformers, a fast and effective aging assessment method for insulating paper is of great significance. As it is known, the degree of polymerization (DP) is the most direct parameter to characterize the aging condition of insulating papers. However, the traditional detection method or so-called viscometry is time-consuming and destructive. Near-infrared spectroscopy (NIRS) technology, as a non-destructive detection method can rapidly determine the samples’ components and contents. Until now, it has been successfully applied in many fields and will hopefully be employed as an alternative method to viscometry. However, the current spectral quantitative analysis method is still not accurate enough to predict the DP of insulating paper samples. In this paper, we introduce Gaussian process regression (GPR) to predict DP of insulating papers accurately. Firstly, the NIRS database of insulating papers under different aging conditions is established, and in this procedure, the raw spectra are preprocessed by the Savitzky-Golay method to improve the signal ratio to noise. Then GPR models with various kernels are established, and the prediction accuracy and stability of the different models are comparatively studied. The results show that the GPR model with Exp kernel is of poor generalization performance, and the models with Matern32, Matern52 and RQ kernels are highly sensitive to the model parameters. Finally, the SE kernel is selected as the optimal kernel function of the GPR model. The DP prediction results of the SE kernel GPR model are compared with traditional PLS, SVR and BPNN models, and the results show that our established GPR model has the lowest RMSE (65.5 and 70.6) and highest correlation coefficient r (0.94 and 0.93), both for the training set and testing set. The RMSE of the GPR model is lower than PLS, SVR and BPNN models by 54.1%, 58.8% and 12.9% respectively. It is indicated that the established GPR model can be a powerful tool for the aging assessment of insulating papers by the NIRS technique.
李 元,张文博,陈晓琳,李 含,张冠军. 高斯过程回归在近红外光谱定量分析绝缘纸老化状态中的应用[J]. 光谱学与光谱分析, 2022, 42(10): 3073-3078.
LI Yuan, ZHANG Wen-bo, CHEN Xiao-lin, LI Han, ZHANG Guan-jun. Application of Gaussian Process Regression on the Quantitative Analysis of the Aging Condition of Insulating Paper by Near-Infrared Spectroscopy. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2022, 42(10): 3073-3078.
[1] ZOU Jing-xin,CHEN Wei-gen,WAN Fu,et al(邹经鑫,陈伟根,万 福,等). Transactions of China Electrotechnical Society(电工技术学报),2018,33(5):1133.
[2] Wang M, Vandermaar A J, Strivastave K D. IEEE Electrical Insulation Magazine, 2002, 18(6): 12.
[3] Saha T K, Purkait P. IEEE Transactions on Power Delivery, 2008, 23(1): 10.
[4] Baird P J,Herman H,Stevens G C,et al. IEEE Transactions on Dielectrics & Electrical Insulation,2006,13(02):309.
[5] YANG Ding-kun,CHEN Wei-gen,WAN Fu,et al(杨定坤,陈伟根,万 福,等). Proceedings of the CSEE(中国电机工程学报),2021,41(13): 4710.
[6] LI Guang-mao,QIAO Sheng-ya,ZHU Chen,et al(李光茂,乔胜亚,朱 晨,等). High Voltage Engineering(高电压技术),2021,47(6): 2007.
[7] Wang S,Liu S,Yuan Y,et al. Infrared Physics & Technology,2020,106:103276.
[8] JIANG You-lie,ZHU Shi-ping,TANG Chao,et al(蒋友列,祝诗平,唐 超,等). Spectroscopy and Spectral Analysis(光谱学与光谱分析),2020,40(11): 3515.
[9] ZHOU Li-jun,LI Xian-lang,DUAN Zong-chao,et al(周利军,李先浪,段宗超,等). Proceedings of the CSEE(中国电机工程学报),2014,34(21): 3514.
[10] Schulz E,Speekenbrink M,Krause A,et al. Journal of Mathematical Psychology,2018,85: 1.
[11] Kong D,Chen Y,Li N,et al. Mechanical Systems and Signal Processing,2018,104: 556.
[12] Wang B and Chen T. Chemometrics and Intelligent Laboratory Systems,2015,142: 159.
[13] Tao D,Wang Z,Li G,et al. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy,2019,208:7.
[14] Liu Y I,Sun L,Ran Z,et al. Journal of Food Protection,2019,82(10):1655.