Machine Learning as New Approach for Dogleg Severity Prediction
Abstract Conventionally offset well studies are performed by individuals where the results depend very much on visual perception, interpretation, and experience. In the specific cases for predicting the dogleg severity (DLS) output, the offset well study will take time proportionate to the volume of input, with the results being averaged out and contain high tolerances. In specific projects, these tolerances are larger than accepted, encouraging the service provider to utilize conservative solutions such as rotary steerable system (RSS) with high DLS capability in order to reduce the residual risks. These solutions can often be more costly in terms of maintenance and may add unnecessary tortuosity to the hole leading to issues during execution. This paper explores the concept of using machine learning (ML) to perform offset well study and defining key parameters affecting the DLS output. This concept consolidates the vast volumes of data that have been acquired while drilling and defines the relationship of each parameter to the final output of DLS. The first analysis reviewed five offset wells and found a multivariable correlation between applied drilling parameters to the DLS output. This correlation was then applied in 6 boreholes (3 multilateral wells), observing consistent DLS output increase by 50% using the same technology and optimal drilling parameters. The second analysis uses the same process to determine a planning DLS limit in a curve section over different formations. This paper demonstrates the potential of ML in offset well studies and beyond to predict behavior and define the relationship in a big data environment.