Development of a Predictive Model for Carbon Dioxide Corrosion Rate and Severity Based on Machine Learning Algorithms

Area of Science:

• Materials Science
• Chemical Engineering
• Data Science

Background:

• Carbon dioxide (CO2) corrosion is a critical challenge in oil and gas pipelines, impacting equipment integrity and safety.
• Traditional corrosion prediction models struggle with complex interactions, leading to inaccuracies and poor adaptability.
• Accurate prediction is vital for material selection and maintenance strategies in the petroleum industry.

Purpose of the Study:

• To evaluate the effectiveness of machine learning algorithms for predicting CO2 corrosion rates and severity.
• To compare the performance of six different machine learning models against conventional methods.
• To identify the most suitable algorithm for enhanced corrosion management.

Main Methods:

• Systematic data organization, correlation analysis, and feature examination.
• Development and comparison of prediction models using Random Forest (RF), K-Nearest Neighbors (KNN), Gradient Boosting Decision Tree (GBDT), Support Vector Machine (SVM), XGBoost, and LightGBM.
• Evaluation of model performance using R-squared values for prediction and accuracy/F1-score for classification.

Main Results:

• Support Vector Machine (SVM) showed the lowest performance.
• Random Forest (RF) achieved the highest prediction accuracy with R-squared values of 0.92 (training) and 0.88 (test).
• RF demonstrated superior performance in classifying corrosion severity, with 99% accuracy and an F1-score of 0.99.

Conclusions:

• Machine learning algorithms, particularly Random Forest, offer a powerful and accurate approach to CO2 corrosion prediction and classification.
• These advanced methods provide innovative solutions for improving predictive accuracy and operational efficiency in corrosion management.
• The study underscores the potential of AI in addressing complex industrial challenges like pipeline corrosion.