A method for predicting corrosion rate in grounding grids by fusing parameter features and dual-model weighting

Corrosion rate serves as a critical quantitative indicator for assessing the operational state and formulating maintenance strategies for grounding grids. This paper proposes a predictive framework for grounding grid corrosion rate that integrates soil environmental parameter features, sample anomaly identification, and dual-model dynamic weighting. The framework enhances input representation by introducing derivative features such as resistivity–moisture coupling parameters and aggressive ion concentration ratios. A combined strategy employing isolation forest and a hard threshold prior based on corrosion rate distribution is utilized to stratify samples into normal and high-risk categories. The corresponding support vector regression models are trained for each category. The outputs of these dual models are then adaptively fused by using a smooth weighting strategy based on feature space distance and prediction intervals to enhance predictive robustness. Test results on experimental grounding grid corrosion data demonstrate that the final model achieves a mean squared error, mean absolute percentage error, and correlation coefficient (R) of 0.1642, 4.97%, and 0.9705, respectively. These results indicate superior prediction accuracy and stability compared to baseline models, offering an effective solution for corrosion prediction under small-sample conditions.