Pipelines are an important part of offshore oil and gas field development facilities and the main means of gathering and transporting offshore oil and gas resources. However, pipelines are subject to deterioration and degradation in the corrosion media. Corrosion risk assessment and prediction is an effective way to avoid leakage of oil and gas field pipelines and facilities, ensure safe operation and save cost. Hence, in this study, a machine learning model with excellent predictive performance was constructed for corrosion rate, to provide an effective mean for processing complex corrosion data and to provide a useful tool for further exploration of submarine pipeline corrosion problems. Meanwhile, a method that can effectively and quickly evaluate the accuracy of corrosion rate prediction model was explored, which can be used as a reference to select the most appropriate and accurate Machine Learning (ML) model based on existing data.
Pipelines have been the main transportation pattern of oil and gas because of their safety and economy,1 which are considered as the lifeline of offshore oil and gas transportation. With the booming development of offshore oil industry, the frequency of pipeline leakage is also increasing.2,3 Corrosion is one of the important factors due to some characteristics such as operating environment, service life and transportation medium, etc.,4 which damages the integrity of the pipeline and damage the normal operation of pipelines.5 Furthermore, leakage accidents caused by pipeline corrosion have occurred all over the world, accounting for 70∼90% of total accidents,6 which has caused huge economy losses and catastrophic environmental damage. Therefore, it is necessary to predict and evaluate the corrosion degree of subsea pipelines with the available corrosion monitoring data to reduce possible accidents resulting from pipeline corrosion.7
In recent years, extensive work has been promoted on pipelines corrosion and a variety of pipeline corrosion prediction models have been established. However, corrosion is a rather complex process, in which there are many factors involved, and there may be relationships among them. As a result, it is difficult for traditional empirical and semi-empirical models to mirror the interactions among various corrosion factors. Besides, they have limitations as they don’t take into account the stochastic nature of the corrosion process.8