期刊
RELIABILITY ENGINEERING & SYSTEM SAFETY
卷 219, 期 -, 页码 -出版社
ELSEVIER SCI LTD
DOI: 10.1016/j.ress.2021.108231
关键词
Corrosion rate prediction; Subsea oil pipeline; KPCA; BRANN; Machine learning
资金
- National Natural Science Foundation of China [52004195]
- China Postdoctoral Science Foundation [2020M673355]
- Fundamental Research Funds for the Central Universities [20CX02315A]
- Opening Fund of National Engineering Laboratory of Offshore Geophysical and Exploration Equipment
This paper presents a novel data-driven model for corrosion degradation prediction of offshore oil pipelines, which integrates KPCA and BRANN techniques to improve the robustness and accuracy of the prediction.
Corrosion is an important reason for the structural degradation of offshore oil pipelines, which may cause serious economic loss and environmental pollution. Nowadays the digitalized devices make a number of monitoring data become available. The prediction of corrosion degradation based on monitoring data becomes an efficient tool to prevent corrosion failure of offshore oil pipelines. This paper integrates KPCA and BRANN techniques to develop a novel data-driven model for corrosion degradation prediction of offshore oil pipelines. The model can eliminate the redundant information from the original monitoring data and improve the robustness by regularization constraints. KPCA is applied to reduce the dimension of the factors affecting pipeline corrosion, and the extracted principal components of corrosion variables are inputted in BRANN to build a corrosion degradation prediction model. The data with dimension reduction are divided into training set and validation set. The model is compared with BRANN alone and KPCA-LMANN model, which indicates KPCA-BRANN model presents superiority in the robustness and prediction accuracy (MSE = 0.46%; R-2=0.99). The proposed model can be used as an online prediction module of digitized process safety system, and support the reliability assessment and maintenance planning of corroded subsea pipelines.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据