基于机器学习算法的CO_(2)腐蚀速率预测

Prediction of CO_(2) Corrosion Rate Based on Machine Learning Algorithms

针对传统CO_(2)腐蚀速率预测模型结构简单,泛化性弱,油田应用效果不佳的问题,本文基于机器学习的思想,采用数据挖掘的手段,研究Na^(+)、Cl^(-)、pH值、温度、材质等参数对井筒腐蚀的影响,利用相关系数矩阵确定各因素之间的独立性及对井筒腐蚀速率的影响程度,分别建立了线性回归、支持向量机(SVM)、极端梯度提升(XGboost)算法的CO_(2)腐蚀速率预测模型。以决定系数、平均绝对误差和均方根误差为评价指标,对比分析了不同腐蚀速率预测模型。结果表明,XGboost腐蚀速率预测模型效果最好,线性回归腐蚀速率模型预测效果优于SVM的4种核函数腐蚀速率预测模型。研究结果揭示了基于机器学习的腐蚀速率预测模型有很好的稳定性和泛化能力,预测效果较好,为油田量化CO_(2)腐蚀速率提供了新的方法,在现场中具有一定的推广应用价值。

The traditional CO_(2) corrosion rate prediction model has a simple structure,weak versatility,and poor application effect in oil fields.Based on machine learning,the influences of Na^(+),Cl_(-),pH,temperature and material on wellbore corrosion rate are studied using data mining method,and the independence of each factor and its influence on wellbore corrosion rate are determined using correlation coefficient matrix.The CO_(2) corrosion rate prediction models are established using linear regression,support vector machine and XGbost algorithm respectively.Taking the determination coefficient,mean absolute error and root mean square error as evaluation indexes,the prediction effects of different corrosion rate prediction models were comparatively analyzed.The results show that the XGboost corrosion rate prediction model is the best,and the prediction effect of the linear regression corrosion rate prediction model is better than that of the SVM corrosion rate prediction models with four kinds of kernel functions.The research results reveal that the corrosion rate prediction models based on machine learning have good stability and generalization ability,and their prediction effect is good,which provides a new method for quantifying CO_(2) corrosion rate in oil fields,and has certain popularization and application value in the field.