(Plant Engineering. (n.d.). [Waterflooding]. Retrieved from https://plant-engineering.tistory.com/267)
Conventional fields are ripe with data! In contrast to shale fields, conventional reservoirs are also normally well connected. Can historical data – well production rates, well injection rates, and their measured bottom hole pressures – all recorded in production history software be used to obtain insights about your reservoir? Without using any geological model, can you rely on just this data to determine effective and ineffective water injection wells? It turns out there are many ways to do that. I want to elaborate on one such surveillance module workflow in HawkEye called Capacitance Resistance Models.
Capacitance resistance models were first proposed in 1943 – obviously when computing was not a rage! I am referring to a time when engineers used reduced physics models for optimization. By reduced physics models, I mean models that are likely ODEs (Ordinary differential equations) with closed-form analytical solutions (for the uninitiated – solutions that do not require any recourse to numerical methods). Figure below shows an image of the first publication where this idea was introduced.
If you have been a grad student, how could you not be mesmerized by analytical solutions – Buckley Leverett solution (first-order PDE) or these CRM models (ODEs)? I was mesmerized too and enjoyed working on some of these problems. One might be tempted to ask – in today’s day and age with accessibility to numerical computations – why bother? I hear you. But it turns out analytical solutions can help with quick decision making and how solutions change if you vary parameters. They help build intuition on solutions – isn’t that something that drew us to engineering in the first place?
Several impressive extensions of Capacitance Resistance models were done at The University of Texas at Austin. Larry Lake (my Ph.D. adviser) was instrumental in popularizing these methods (one of his talks). One of the main criticisms has been its adaptability to field data – often characterized by missing data (were the wells closed or did we just not measure data?) as well as the challenge of finding specific production and injection schedules to apply these models. Some companies do use these models as excel sheets. But that is not useful, especially when you have too many wells resulting in challenges to nonlinear optimization.
At Resermine, our love for analytical solutions has inspired us to make seamless CRM workflows. While our browser-based platform automates data preparation and time period selection, qualitative analysis on this segmented data is a crucial first step. In particular, Pearson’s correlation coefficient and Spearman’s rank correlation coefficient helps qualitatively segregate injector-producer pairs. This integration of qualitative analysis in CRM models has helped obtain nonlinear optimization solutions, especially when you have many several wells – sample this EAGE IOR conference publication. Such integration is the backbone of creating solutions that scale and are applicable to different reservoirs.
It turns out machine learning algorithms can also help with qualitative analysis. Additionally, results from CRM can be combined with reservoir simulations to make predictions from machine learning models more reliable. Our expertise is such hybrid solutions available on our HawkEye product. More on that later.
It is about time we leverage production and injection data to have a continuous water injection strategy. Figure below shows the benefits of doing a continuous CRM optimization) taken from our EAGE IOR conference publication).