Capitalizing on Human – Technology Interaction for Shifting from Automated to Augmented History Matching: A Classical Demonstration to Front End Study for a Complex Thermal Field in Petroleum Development Oman

Abstract This paper presents a classical demonstration of human-machine interaction to design and deploy a history matching workflow for a complex EOR brown field development. The field in the Sultanate of Oman (with viscosity from 400cP - 600,000cP) under study is unique with hardly any analogue and is developed thermally. Thermal simulations for such complex fields at fullfield scale become computationally intensive, especially when integrating multiple finer geological realisations with suite of dynamic data, e.g., continuous data from observation wells, time-lapse logs etc. in addition to production, injection and pressure histories. This limits our ability to holistically sample uncertainty space by running multiple scenarios and hence, restricts ability to create robust history match (HM) and forecast solutions. This paper presents an integrated modelling workflow which overcomes such limitations for complex simulations-based field studies. It helps in model maturation and facilitates accelerated field (re-) development decisions. A working reference case simulation model along with holistic uncertainty matrix are first created. Afterwards, approach splits into two parallel but interacting work-streams for faster solutions. In first stream, model scenarios are created in ad-hoc fashion and various physically viable combinations of uncertainty parameters are tried to yield HM. In second work-stream, numerical and data analytics capabilities of simulator are utilized, wherein uncertainty matrix is used for sensitivity analyses and creation of a Response Surface Model. Then HM engine is set-up by defining appropriate objective functions at global and local scale and selecting numerical optimizer. Global optima are found in solution space, and then local minima are searched for the corresponding global minima. The alternate HM scenarios from first work stream supplements the local minima obtained from the second. These simulations are used to create proxy functions based on polynomial regression or neural network. Subsequently, PDFs are generated, and posterior distribution of uncertainty variables are obtained leading to uncertainty reduction workflow. The workflow yielded successful outcomes for the field. The Assisted or "Augmented" HM methodology yielded not just one unique HM, rather multiple (~10) non-unique HM solutions underlying discrete geological scenarios. HM realisations were forecasted with established ranges around system design/control parameters to yield spectrum of UR predictions. This assisted in identifying reservoir true potentials, gaps/bottlenecks and therefore, optimizing field development and initiating WRFM projects in addition to providing robust forecasts for asset's business planning and reserves. The methodology impacts business bottom-line by significantly reducing the time taken in front-end studies and accelerating delivery of field development/re-development plans for complex fields. It proved a key enabler in maturing significant undeveloped volumes in the field. Advancements in technology, e.g., cloud computing and AI can boost it further by launching thousands of smart runs on cloud and improving quality of proxies.