Integration of hybrid machine learning model with differential evolution algorithm for underground Pipeline corrosion prediction

Pipeline corrosion compromises pipelines' structural integrity and safety, necessitating reduction to prevent leaks, environmental damage, and potentially catastrophic failures. Although several field and theoretical strategies have been adopted in the literature to determine pipeline corrosion, some limitations remain, such as providing cost-effective methods, less complex procedures, and reliable predictive models. This research seeks to evaluate the ability of a newly developed machine learning (ML) model based on Extreme Learning Machine (ELM), Support Vector Machine Regressor (SVR), Extreme Gradient Boosting (XGB), and Multivariate Adaptive Regression Spline (MARS) and nature-inspired Differential Evolution (DE) optimization for pipelines corrosion prediction. As a secondary research objective, the proposed ML model was integrated with a feature selection (FS) algorithm to identify the essential parameters of corrosion determination. The research used a dataset from an open-source literature review with 259 observations. There are six pipeline parameters, i.e., redox potential (rp), potential of hydrogen (ph), pipe-to-soil potential (pp), soil resistivity (re), pipeline age (t), and pipeline coating type (ct) and six soil parameters, the soil in which the pipeline is buried, i.e., soil textural class (class), water content (wc), bulk density (bd) and dissolved chloride (cc), bicarbonate (bc), and sulfate (sc) ion concentrations. The predicted results indicated that XGB_FS model performed the best with maximum determination coefficient (R² = 0.8407) and minimum root mean square error (RMSE = 0.395). The research evaluated the efficiency of the developed model, the reliability of using such an algorithm in corrosion predictions, and whether it enhances the average performance when combined with other models, such as hybrid ML models.