An insulator pollution degree detection method based on crisscross optimization algorithm with blending ensemble learning

Pollution degree not only affects the lifespan of insulators, but also plays an important role in the operation of transmission lines. Current detection methods are usually conducted offline, which cannot provide a timely reference for monitoring and maintenance. To address this issue, an insulator online detection system is designed in this paper. Considering the problems of optimization difficulty, over-fitting and low accuracy, a novel hybrid model is proposed to classify the insulator pollution degree based on crisscross optimization algorithm (CSO) and blending ensemble learning. First, two key features, i.e., humidity and leakage current, are extracted. Second, since the hyperparameters of the random forest directly influence the prediction accuracy, the CSO algorithm is employed to dynamically optimize these key parameters. On this basis, a blending ensemble learning algorithm is applied to establish the hybrid model by integrating random forest and XGBoost. Experimental results show that the proposed hybrid model can effectively classify the pollution degree of insulators.