Adaptive Gas-Lift Allocation Optimisation Using Model Predictive Control and Genetic Algorithms with Full-Physics Reservoir Simulation

Abstract

This study presents a hierarchical intelligent control framework for real-time gas-lift allocation optimisation in a three-well system based on the SPE9 benchmark reservoir. The approach combines a Genetic Algorithm (GA) as a slow supervisory optimiser with a Model Predictive Controller (MPC) acting as a fast, smooth executor within a MATLAB/Simulink environment. Unlike many existing studies, the framework is validated using a full-physics three-phase black-oil model implemented in MRST, ensuring realistic subsurface behaviour. The GA optimises the allocation of a constrained gas budget of 6.0 MMscfd across the wells at 90-minute intervals using surrogate performance models. At the same time, the MPC tracks these setpoints every 15 minutes with rate-of-change constraints to maintain well stability. To enable fair comparison, three parallel reservoir simulations are executed simultaneously under identical conditions. Results from a 300-minute simulation show that the proposed GA–MPC framework achieves a steady-state field oil production of approximately 2,615 STB/d, representing a 31.3% improvement over both equal-split and proportional allocation strategies. Notably, the optimiser autonomously identifies and shuts off a fully watered-out well, reallocating gas to more productive wells without predefined rules. The GA consistently converges within 4 generations at each optimisation step, indicating robust solutions. The entire simulation completes in approximately 7 minutes of computational time, though this depends on the computer's model, demonstrating the potential for real-time field deployment