A Large Scale Network Model to Obtain Interwell Formation Characteristics

Abstract

Limited data availability and poor data quality make it difficult to characterize many reservoirs. For reservoirs that have undergone waterflooding, production and injection data are a reliable source of information from which injector-to-producer connections can be inferred. In this research, we use well locations, injection and production rate data and well fractional flow values to develop a reservoir-scale network model. A Voronoi mesh divides the reservoir into a number of node volumes each of which contains a well. Bonds connect each of the nodes with conductance values that must be inferred from the rate data. An inverse problem is formulated where the mean-squared difference between the modeled and actual production data are minimized and the conductance values between each node are the unknowns. A derivative free optimization algorithm is utilized to minimize the objective function. Knowing the conductivity of each of the bonds, a two phase problem is formulated and solved in this work to obtain the fractional flow at node interfaces and at each producer. The set of cross-sectional areas open to flow and time-of-flight Dykstra-Parsons coefficients that minimize the simulated and actual producer fractional flow values is the solution to this problem. This work is primarily for secondary and tertiary floods with limited geological data. The solution parameters are directly proportional to formation properties. In addition, they help to evaluate the degree of sweep between wells. This approach has been successfully tested for different synthetic permeability distribution cases and field injection scenarios. The main advantages of the proposed method are: It can model changes in flow pattern caused by adding new wells or shutting in producers. It uses conventional history matching methods to solve a simplified inverse problem using only production and injection data. It uses a small number of nodes and converges to a better posed solution than statistical approaches. Convergence to a solution for higher frequency data only decreases the speed of the method slightly. The shape of the fractional flow curve is determined by the time-of-flight Dykstra-Parsons coefficients and cross-sectional areas open to flow which can be related to reservoir properties.