Analysis of rate of penetration (ROP) prediction in drilling using physics-based and data-driven models

作者：Hegde Chiranth^*; Daigle Hugh; Millwater Harry; Gray Ken

来源：Journal of Petroleum Science and Engineering, 2017, 159: 295-306.

DOI：10.1016/j.petrol.2017.09.020

• ROP
• Data-driven
• Machine learning
• Drilling
• Data analytics

Modeling the rate of penetration of the drill bit is essential for optimizing drilling operations. This paper evaluates two different approaches to ROP prediction: physics-based and data-driven modeling approach. Three physics-based models or traditional models have been compared to data-driven models. Data-driven models are built using machine learning algorithms, using surface measured input features - weight-on-bit, RPM, and flow rate - to predict ROP. Both models are used to predict ROP; models are compared with each other based on accuracy and goodness of fit (R-2). Based on the results from these simulations, it was concluded that data-driven models are more accurate and provide a better fit than traditional models. Data-driven models performed better with a mean error of 12% and improve the R-2 of ROP prediction from 0.12 to 0.84. The authors have formulated a method to calculate the uncertainty (confidence interval) of ROP predictions, which can be useful in engineering based drilling decisions.

• 出版日期2017-11