Predictive Modeling of Outdoor Glass Insulator Surface Contamination and Ageing Using Hybrid Machine Learning Classifiers

Surface contamination on insulators is a significant issue in high voltage transmission lines, often leading to electrical flashovers and subsequent power disruptions. Frequent flashovers may compromise the power supply reliability and the integrity of the electrical equipment. The leakage current analysis serves as a critical diagnostic approach for assessing surface contamination, abnormal electric stress, and operational or health status of insulators. However, the present leakage current diagnostic techniques often neglect insulator ageing and overlook the impact of relative humidity on leakage current indices, reducing their accuracy. In fact, the leakage current signal analysis is very data-intensive, which highlights the need for an effective data reduction method. In this study, twelve glass insulators grouped into three categories namely new, moderately aged, and heavily aged, were pre-contaminated to varying degrees and then tested under high voltage in a purposely made climatic chamber with controlled humidity. This work had successfully identified the best predictive models using multi model classifier technique. For the insulator ageing condition predictive model, the proposed multi model classifier (Hybrid model-4; ANN; NFS) gives the best overall performance at 97.6%. For the insulator surface contamination predictive model, the proposed multi model classifier (Hybrid model-5; ANN; NFS; ANN) gives the best overall performance at 99.0%. Although PCA reduced accuracy in most ageing prediction models, it significantly improved processing time, reducing training duration. Hence, this advantage can be used as a motivation to incorporate the PCA technique in future classifying models with improved overall performance.