Optimized Ensemble Machine Learning Models for Predicting Lifespan of Composite Insulators

Polymeric insulators are a crucial element in power transmission lines, and early, accurate aging prediction is essential to ensure the reliable and secure operation of the power grids. Long service in overhead electrical grids affects the performance of polymeric insulators, leading to changes in their electrical characteristics and unexpected failures. Using ensemble machine learning models represented by Random Forest (RF), Light Gradient Boosting Machine (LGB), Categorical Boosting (CB), Weighted Ensemble (WE), and Extreme Gradient Boosting (XGB), this paper proposes an efficient prediction approach for polymeric insulator lifespan. 360 rows of data of twenty-three aged insulator characteristics for training and testing is used for training and testing to accomplish this goal. An additional 64 data points are used to validate the proposed model. The hyperparameters are tuned using the Improved Grey Wolf Optimization (IGWO) and 10-fold cross-validation techniques. Four other optimization techniques are compared to IGWO for comparison. The IGWO-XGB model demonstrated strong predictive accuracy, achieving an R² value of 99.83% for insulator aging estimation. Shapley Additive Explanations (SHAP)-based feature analysis is implemented to interpret the effects of features on model prediction accuracy. Results of performance indicators revealed that the XGB model delivered the best prediction performance, indicating the proposed model has promising interpretability and can rapidly and accurately predict the critical age of polymeric insulators.