Hybrid Physics-Based Data-Driven Methods… the Future for Petroleum Engineering?

Data-driven methods are widely and successfully used in social sciences, an area where no discernible physics exists, but many observations exist. As a result, there has been a drive to implement such approaches into our industry.

However, in petroleum engineering, we are faced with the opposite scenario: limited observations with ambiguous interpretation and relatively well-known physics and equations. Over the last 50 years, we have seen the establishment of physics-based approaches, with numerical simulation at its core.

We are seeing the emergence of hybrid approaches: data-driven but physics compliant. In this work, we contend that these methods may provide the best of both worlds, and we provide two specific examples of such hybrid methods. First, we introduce a novel method for zonal production allocation, which allows for the inclusion of physics and observation data. We demonstrate through a case study how this approach significantly improves production allocation and therefore reservoir management decisions. Secondly, we discuss the integration of the published remaining oil compliant mapping algorithm with machine learning methods for the purpose of ‘locate-the-remaining-oil’ activities and determining behind casing and infill drilling opportunities. Drilling results from recent projects are examined, and the method’s accuracy is evaluated.