Bridging the performance gap between passive and autonomous inflow control devices with a hybrid dynamic optimization technique integrating machine learning and global sensitivity analysis

Wells equipped with flow control devices across their completion intervals have become a proven field development option for geologically complex and/or viscous oil reservoirs. Such wells increase oil recovery, reduce water and gas production, minimize the need for well workover operations, and subsequently lower the wells' carbon footprint. The uncontrolled types of inflow control devices include early-generation passive inflow control devices (ICDs) and later-generation autonomous inflow control devices (AICDs). The superior performance of AICDs over ICDs in managing water and gas production, as well as enhancing the overall well and reservoir performance has been demonstrated in multiple research and case studies. This superiority stems from the AICDs' ability to self-adjust and increase their flow resistance when undesired fluids (i.e., water and/or gas) flow through them. While ICDs lack this self-adjusting feature, they are more affordable and more readily available on the market. This study aims to reduce the performance gap between passive and autonomous inflow control devices by developing a hybrid dynamic optimization technique. This approach integrates a metaheuristic algorithm, machine learning, global sensitivity analysis, and correlation measures to facilitate the optimization problem by identifying the high-impact control variables. Next, the proposed workflow finds the necessary adjustments to the original well completion design by modifying the high-impact control variables during the optimization process. This results in a modified well completion design that is less influenced by the type of inflow control device (passive or autonomous), thereby bridging the performance gap between these two completion types. The study employs a benchmark 'Egg field' model, featuring two multilateral wells (MLWs) producing under a water flooding recovery mechanism. Two different completion designs, utilizing either ICDs or AICDs, are optimized using standard optimization (SO) and the proposed hybrid dynamic optimization techniques. The standard optimization, which employs a standalone Particle Swarm Optimization (PSO) algorithm, highlights, as expected, the superiority of the AICD-based completion, yielding an approximately 13% increase in the net present value (NPV) over the ICD-based completion. However, when applying the hybrid optimization (HO) technique, this difference is significantly reduced to 3.4%. This indicates the potential for the hybrid optimization technique to make ICD-based completions more competitive and economically favourable compared to their AICD-based counterparts.