Intelligent wells with inflow control valves enable flexible management of multiple reservoir intervals. In this paper, we describe model based optimization of inflow control valve settings for a producing North Sea field. A two step, non-invasive, iterative pattern search optimization algorithm is applied. The first step provides global search of the feasible region using a discrete genetic algorithm whereas the second step provides local search around the incumbent solution. To account for reservoir uncertainties, optimization is performed on a diverse set of history matched reservoir model realizations, within an automated framework. The results show a significant improvement in predicted reservoir production performance over the remaining life of the field.
Wells are a crucial element in any field development optimisation effort. A field development will not be optimal unless we drill optimum types and number of wells, at optimum locations in the reservoir. Traditionally, a manual trial-and-error approach is taken for field development optimisation. The manual approach allows the engineer to exercise full control over the process, which can be beneficial when the engineer is very knowledgeable, but this approach is very inefficient and unreliable. Increasingly, stochastic optimisation algorithms are being applied to identify optimal or near optimal well configurations. For typical field development optimisation problems, these stochastic algorithms require many thousand objective function evaluations, in the form of reservoir simulations, and this is a major issue.
In this paper, we demonstrate an alternative approach to field development optimisation. It starts with one or more sets of candidate wells, with associated type, location and trajectory information. Iteratively, elimination of wells is performed based on a simulation derived well valuation metric. The elimination of wells corresponds to a modification of the development well configuration (well count, locations and well types). Therefore, by comparing the field level valuation metrics obtained across iterations, we can determine the optimal development well configuration. Our approach focuses the optimisation search only on the promising areas of the solution space, represented by the initial candidate well sets, or starting well configuration. Determination of locations and trajectories of the initial candidate wells is therefore key. In our work, we determine the locations and trajectories of initial candidate wells using patterns, potential estimates and an Ant System.
Two examples are used to illustrate the effectiveness of our approach. The results show that the approach significantly speeds up field development optimization, allowing simultaneous optimisation of well locations, well count and well types.