In this paper we use the concept of experimental design from the System Identification literature and introduce a novel methodology for more accurate characterization and monitoring of waterfloods using linear models. To improve the predictability of the low order linear models they must be trained with sufficiently informative data. To achieve this goal, the input (here injection rate) must contain sufficient variations (without exceeding the technical and operational limits) so that all the relevant dynamics of the system are excited and can be seen in the output (here production rate). In sum, a successful injection scheduling design boils down to introducing predetermined variations on top of the existing injection rates. We will discuss how to design these variations, specifically their amplitude, sampling time, experiment length, frequency of variations and signal type, using priori knowledge of the reservoir dynamics and the production constraints. Our proposed methodology can be applied to currently available linear modeling techniques such as CRM, FIR, subspace models, etc. This study shows that the prediction error of a model that has been trained with a proper injection scheduling design, in a multi injector multi producer oil field, can be reduced significantly regardless of the modeling technique.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.