Haoran Sui, Kangning Wu, Haoyue Jiang, Zichao Yang, Zhuolin Zhang, Benhong Ouyang, Jianying Li
To efficiently and economically assess the insulating condition of widely used cross-linked polyethylene (XLPE) power cables, a non-destructive method for quantitatively predicting breakdown properties based on near-infrared (NIR) spectroscopy is proposed in this work. The NIR spectra of 24 XLPE samples with different aging conditions was established for calibration set. 40 key wavelengths near 1715, 1730, 1750, and 1763 nm strongly associated with breakdown strength were selected among 301 wavelengths by competitive adaptive reweighted sampling (CARS). The selected wavelengths are proved to be related with crystallinity which is positively correlated with breakdown strength. Further results of surface potential decay (SPD) reveal that breakdown strength is modulated by deep trap energy level which is positively correlated with the intensities at 1730 and 1763 nm. On that basis, a back-propagation (BP) neural network model with an R2 value of 0.96 and a relative RMSE of 3.20% was constructed using these wavelengths, achieving better predicting abilities than the models without CARS. It also achieves only 1.34% error in predicting the breakdown strength of XLPE samples from cables that have been in service for 6 different times. This work provides an accurate and efficient method for evaluating insulating condition of power cables.