Streamline approaches for integrating production history with geologic information in reservoir models

Reservoir management often requires reservoir models. Geostatistics is a prevailing technique to generate reservoir models that honor known, but sparsely sampled, geology. However, these models are not constrained by production data. History-matching plays an important role in modifying reservoir models to reproduce the production history. However, most history-matching approaches manipulate parameters at the gridblock level, and hence, demand intensive computation. Reservoir models obtained by many history-matching methods do not reflect the known geology. This study addresses these problems by developing a streamline-based history-matching method and two approaches integrating history-matching with geology.; For the streamline-based history-matching method, we first establish a relationship between the production data and streamline properties, such as the time-of-flight and effective permeability. Based on that relation, a system of equations is constructed to minimize the error of a model with respect to the observed production history. The modifications of model parameters are performed at two levels: first, modifications are computed for streamlines; then, these modifications are mapped into grid-blocks.; In the first approach for integrating geostatistical and history-matching techniques, an optimization technique called the Gauss-Markov-random-function (GMrf) is combined with the streamline-based history-matching method. We perform history-matching, aimed at matching the production data alone, and then use GMrf to restore the geology possibly damaged during history-matching. The second approach ranks streamline-time-of-flight distributions among reservoir models. We establish the correlation between various production data and their corresponding streamline properties, and consequently formulate a fast ranking-algorithm as follows: first, a history-matched reservoir model is obtained. Next, streamline geometry and times-of-flight are computed for the history-matched model and the models to be ranked. Finally, the errors of each model are computed with respect to the history-matched model; models with minimal errors are selected.; This thesis presents the theoretical and numerical development of the three methods. Algorithms are explained in detail, followed by synthetic examples. Their advantages, limitations, and computational efficiency are examined. All three methods are robust and computationally efficient. The two approaches for integrating geostatistics and history-matching generate reservoir models that honor both known geology and production data.