Research Of Insulator Contamination Grades Detection Method Based On Infrared Image And Artificial Intelligence

With the increasingly aggravating of environment pollution and the continuously expanding of power system scale, flashover of contaminated insulators occurs frequently. It seriously affects the security, the stability and the reliability of the power transmission. Developing safe and accurate method for insulator contamination severity detection is a hot issue and a hard task in high voltage and insulation field. it is significant for the resolving of flashover problem. A new method based on infrared image features of contaminated insulators and artificial intelligence for insulator contamination severity detection is first time put forward in this paper. The method is illustrated detailedly from theoretics,algorithm and experiment level. The outstanding virtue of this method is that can achieve an untouching detection. there are many other excellences, such as safety, economics, accurate and facility.A highly sensitive infrared camera is used to get infrared images of contaminated insulators. Then, a wavelet adaptive method based on Bayes estimation is presented for insulator infrared image denoising. An unbiased least-squares estimation rule is adopted to estimate the wavelet coefficients on different orientations of different wavelet scaling spaces adaptively. The mean of the posterior probability distribution is used as the estimations of the wavelet coefficients of different scales. Wavelet inverse transform is used to obtain the denoised image. Experiment results indicate that the denoising method proposed in this paper is good on keeping the information of the original image and excellent on noise removing. The denoising ability of this method is better than the fixed threshold method and the maximum_ minimum threshold method. The eminent performance of this method is proved by its higher SNR(signal- to- noise rate), smaller MSE(Minimizes the mean squared error)The discal surface of the contaminated insulators is the interesting area of the research. The validity of the features extraction is directly depends on whether the discal surface could be well segmented from the whole image or not. According to the different characteristics of the gray intensity histogram of insulator images under different contamination grades, an improved image segmentation method based on the histogram trough of the insulator image is first time presented for the segmentation of the badly contaminated insulators and an OTSU method integrate morphologic method is presented for the segmentation of the slightly contaminated insulators (including clear insulators). Experiment results indicate the segmentation quality is eminent, the discal surfaces of the insulators are intact and the discal surface margins are clear.Ten contamination features are extracted from the contaminated insulator infrared image,viz.the highest , the lowest , the mean temperature of the background, the temperature variance of the background ,the highest ,the lowest ,the mean temperature of the contaminated insulator discal surface , the temperature variance of the insulator discal surface ,the maximum and the mean temperature increase of the discal surface contrasting to the mean temperature of the background .The overmany features will increase the complexity of the data processing. K-L transform is utilized for the first time to decrease the feature dimensions of the contaminated insulators. Three independent principal components which containing information of every original feature are obtained by K-L transform. Experiment results show that K-L transform decrease the dimensions of the original contamination data while preserving the information of the original features, simplify the calculation, increase the distances between classes, and improve the accuracy of data classification.With the humidity and the loading voltage taken into accounted, RBPNN (radial basis probabilistic neural network) classifier and SVM (support vector machine) classifier are designed to check the contamination grades of the contaminated insulators. The mapping relationship between the contamination grades and the contamination features is first time established by the great nonlinear mapping ability of RBPNN and SVM.The fast and accurate recognition of the contamination grades is achieved by the arbitrary approximation capability of RBPNN and its minimum Bayesian risk classification criterion. The excellent processing capability of SVM on nonlinearity, high dimensions and local minimum, especially its mighty processing capability under small samples situation make it achieve the highly accurate detecting of the contamination grades.Experiments results indicate that insulator contamination grades recognition method based on infrared image and artificial intelligence presented in this paper is a safe, effective, accurate and feasible new method for contamination grades detection of insulators.