To improve performance of waterfloods with minimal capital investment is important as the crude price is low. Adjusting the well controls to achieve a more efficiency sweep pattern is more economic than side tracking or infill drilling. This paper presents a methodology designed to guide well controls and maximize the recovery of remained oil in large and mature waterfloods by modeling and optimizing the inter-well connectivities. The workflow includes three steps: modeling, uncertainty quantification (UQ), and production optimization. Firstly, the reservoir is modeled as a connected network characterized by the strength and efficiency of each injector-producer connection. The concept is similar to the flux pattern derived by streamlines (Thiele and Batycky, 2006). But the presented approach does not use streamlines, and instead simulates tracer concentration between each well pair to quantify the strength of energy support from injectors to producers. The technique is a generalized form of the work by Shahvali. et al (2012). The efficiency of connections measures the oil contribution of each connection, which identifies the water cycling. It is history matched by a data-driven technique. In the UQ step, the method estimates the possible range of efficiency due to the non-uniqueness of the history matching solution. The efficiency of the connection carrying less flux in the entire history tends to be more uncertain. We quantify the uncertainty by evaluate the upper and lower bound of the efficiency subject to similarly good history matching. The formulation of the maximization/minimization was inspired by the work of Van Essen, et al. (2010), but the optimization algorithm differs and is a non-linear constrained partem search method. For production optimization, a nonlinear optimization problem is formulated based on the connectivity model to find well controls strengthening efficient connections and weakening inefficient connections. The optimization algorithm takes advantage of the linearity of the network model to achieve faster performance than pattern search. Here UQ regulates the risk of the recommended well control strategy. The methodology was tested based on a full simulation model of a real field with 200+ wells, which was regarded as the true reservoir in this study. We trained our network model for 3 years then started to optimize the waterflooding strategy for six months. The results demonstrated that the optimized strategy maintained oil production and reduced water production by 50% without adding new well, while the historical operation satisfied the oil target by drilling tens of new wells and scarifying water-cut.
展开▼