Hybrid machine learning predictions of high voltage polymeric insulator pollution.

This study presents a comprehensive approach for predicting pollution levels on high-voltage insulators in the Eastern Region of Saudi Arabia using five machine learning techniques integrated with the Improved Harris Hawks Optimizer (IHHO). The focus is on accurately estimating two key pollution metrics: Equivalent Salt Deposit Density and Normalized Salt Deposit Density. IHHO was used for hyperparameter optimization to guarantee that all proposed models function adequately. The prediction model is constructed from one year’s worth of environmental data, which includes temperature, humidity, wind speed, solar radiation, elevation, dew point, precipitation, line voltage, and service period. Of all evaluated models, the hybrid IHHO-XGBoost model performed best, with an R2 of 0.993, a MAE of 0.0002, an RMSE of 0.00015, a MedAE of 3.563 × 10⁻5, an EV of 0.992, and an Adj R2 of 0.9923 with tenfold cross-validation. Model validation using Taylor diagram analysis confirmed a high degree of agreement between predicted and actual values. Furthermore, application of the SHAP (SHapley Additive Explanation) technique revealed that the most important predictors were wind speed, temperature, line voltage, and solar radiation. In addition, the results were compared to the results of other benchmark models to improve model explained accuracy and trustworthiness. Not only did the IHHO-XGBoost model best others in accuracy, it equally enhanced understanding of the environmental and operational factors that cause insulator contamination. These predictive capabilities support more effective condition monitoring and maintenance planning, ultimately contributing to improved reliability of electrical grid infrastructure in harsh environments.