Rate of penetration (ROP) refers to penetration speed in drilling process, which is directly related to drilling cost and time. Therefore, ROP optimization by choosing the best drilling parameters has been an important issue in drilling fields to cut cost and spend less time in drilling. To optimize ROP in an accurate and efficient manner, an accurate ROP prediction should precede. An accurate ROP prediction makes optimizing drilling parameters possible, which leads to cost-effective drilling.
Recently, the development of machine learning brings new insight into various fields including petroleum engineering, where conventional and domain knowledge-based equations were dominant. Particularly, leveraging deep neural networks that are highly capable of representing complex relationship of features has shown promising results in various tasks and fields such as computer vision and language modeling.
In this paper, we present a novel machine learning model for the ROP prediction. We focused on developing a model which is trained with offset well logs data to predict ROP of unseen target wells within the same field. By utilizing the state-of-the-art deep neural networks specialized in sequential data process, we show that the data driven ROP prediction significantly outperforms conventional ROP prediction methods.
In addition, we utilized a stacked generalization ensemble method that integrates several ROP prediction models to produce an optimal ROP prediction model. We also show that our stacked ensemble model can improve prediction performance of a single model.