The operation of oil and gas fields requires that a multi-level decision hierarchy is used to address field optimization at different time scales. These decisions affect production volumes and costs, not just in their short-term outcomes but also over the life of the field. A typical production operations workflow such as production forecasting or production optimization uses dynamic models with varying degree of complexity and detail, to suit the immediate needs of the workflow requiring the reservoir model. For instance, conventional reservoir simulators use a numerical (full-physics) model derived from first principles combined with a history match exercise based on measured field values. However, these full-physics models are not always conducive to use in all workflows responsible to achieve real-time production optimization (RTPO) objectives.

In order to make predictions suitable for short-term decisions, proxy-models (e.g. neural networks, response surface) have been proposed to mimic the reservoir simulator. Although, simple in form, such proxy-model based methods have some practical drawbacks like they are purely data driven, do not capture phenomena that are not observed in the training data set and are not directly validated or updated based on the plant measurements.

In this work we present an overview of the multi-scale RTPO approach with a discussion of modeling issues related to real-time decision making in the time scale of hours to days to weeks. A short-range, reduced-order model structure is developed, and the model parameters are obtained from production history data. Since the model structure is motivated by the decomposition of a full-physics model, it is expected to be feasible to extrapolate outside the range of history data. The reduced-order model also increases the computational efficiency and effectiveness in carrying out the simulation objectives. The paper presents a number of case studies to illustrate the benefits of the proposed methodology and its application in typical workflows for closed-loop reservoir management (e.g. forecasting, optimization, reservoir monitoring etc.).