A Novel Non-contact Pollution Detective Method For Outdoor Insulation Of Overhead Line Based On Hyperspectral Technique

The distribution of energy resources and load centers in China is reversed.This objective reality determines the inevitability of large-scale cross-regional power flow and long-distance transmission in China.UHV DC transmission has outstanding advantages such as large transmission power,long transmission distance,small line loss,less land occupation,and strong controllability.It is the best choice for large-capacity and long-distance power transmission.Due to the effect of the directional electric field,the pollution deposition on the DC insulator surface is intensified.At the same time,because the DC current has no zerocrossing point,the DC arc burns stably,and the creepage distance utilization rate of the DC insulator is low under the same pollution conditions.Pollution is one of the main factors affecting the safe operation of the UHV DC projects.Accurately assessing the contamination degree of the external insulation,arranging the cleaning cycle according to the actual situation and formulating prevention measures in advance are the prerequisites to ensure the safe operation of the UHV DC transmission lines.However,the accurate detection and evaluation of the pollution degree of the external insulation has always been an "old,big and difficult" problem in academia and industry.This paper explores the feasibility of the hyperspectral imaging technology in the non-contact detection of external insulation contamination using the hyperspectral imaging technology which has the characteristics of high spectral resolution,continuous spectral lines,and map unification,and expands new ideas for novel methods of external insulation contamination detection.It has important theoretical research significance and engineering practical value.This paper focuses on three key indicators which are the most concerned in the detection and evaluation of external insulation pollution,i.e.pollution level,pollution salt density and pollution distribution.The hyperspectral line data of the samples reveals the response characteristics of the external insulation pollution to the hyperspectral line in the visible/nearinfrared range.And the variation rule of the spectral reflectance of the artificially-coated insulating sheet samples with the pollution level is found.The influence of three sample set division methods i.e.KS(Kennard-Stone)method,RS(Random Sampling)method and sequential method on the model performance are compared and analyzed from the points of view of statistics and the model recognition results.An optimal sample set suitable for qualitative identification of pollution levels and quantitative prediction approach for pollution salt density are proposed.The approach reduces the model performance degradation due to the defects of the sample set.The processing effects of various preprocessing methods on the hyperspectral data are compared and analyzed.A suitable hyperspectral prediction method from the perspective of improving model recognition and prediction performance is proposed.The processing method effectively reduces the impact of the spectral data noise and other factors on the model performance.At the same time,it is found that the data preprocessing method has a significant impact on the model prediction performance,while the modeling method has minor impact.Different characteristic band selection methods are studied.For the impact on the pollution detection model performance,the best band selection method is proposed,and the key sensitive spectral characteristic bands that can replace the full-band spectral data are selected,which effectively reduces the dimensionality of the hyperspectral data and improves the accuracy of the detection model and the calculation speed.By combining the sample set division method,spectral data preprocessing method and band selection method,a qualitative identification model of pollution level and a quantitative prediction model of pollution salt density are obtained.Finally,the hyperspectral spectral line characteristics of naturally contaminated insulators were studied,and it was found that the spectral reflectance of the naturally contaminated insulators varies with the pollution level and the equivalent salt density.The feasibility and effectiveness of the sample set division method based on the hyperspectral data of the artificially contaminated samples,the hyperspectral data preprocessing method,and the feature band selection method in the detection of natural contaminated insulators are verified.The pollution degree prediction of the single pixel point of the natural pollution insulator image is realized using the hyperspectral imaging technology which can obtain the continuous spectral line data of a single pixel point.The visualization of the pollution distribution of the natural pollution insulator is realized using the pseudo color map,and the non-contact detection of the pollution level and equivalent salt deposit density of the external insulation surface is realized.