基于交叉验证梯度提升决策树的管道腐蚀速率预测

Prediction of Pipeline Corrosion Rate Based on Cross-Validation Gradient Boosting Decision Tree

基于集成学习的思想,在梯度提升决策树算法的基础上建立管道腐蚀速率预测模型,并使用网格搜索与交叉验证方法进行超参数寻优。利用某输油管道的腐蚀实测数据对模型进行验证,并与广泛使用的BP神经网络与支持向量机模型的预测结果作比较。结果表明:梯度提升决策树模型预测结果的平均绝对百分误差为2.25%,低于BP神经网络的6.03%和支持向量机的7.99%,说明梯度提升决策树模型具有更高的预测精度和更优的泛化能力,并且该模型具有可解释性强的优点,可为将来的管道腐蚀速率预测提供一种更加实用的新方法。

According to the idea of ensemble learning, a prediction model of pipeline corrosion rate was established based on the gradient boosting decision tree algorithm, and its hyper-parameters were optimized by grid search and cross-validation methods. The corrosion test data of a certain oil pipeline was utilized to verify the model and was compared with the results predicted by widely used BP neural network and support vector machine model. The results show that the average absolute percentage error of the gradient boosting decision tree model was 2.25%, which was lower than the 6.03% of the BP neural network and the 7.99% of the support vector machine. It indicated that the gradient boosting decision tree model had higher prediction accuracy and better generalization ability than the BP neural network and the support vector machine. In addition, this model had fine interpretability and could provide a new more practical method for the prediction of pipeline corrosion rate in the future.