Adaptive Optimization Strategies for Surfactant Cyclic Injection in Tight Oil Reservoirs

Summary The surfactant huff ’n’ puff process, or cyclic surfactant injection, is a promising technique for enhancing oil recovery from tight reservoirs. The effectiveness of this process is highly dependent on the design of cycle parameters such as injection rate, surfactant concentration, soaking duration, and production time. This study is the first to apply reinforcement learning (RL) to cyclic surfactant injection, introducing a framework that adaptively optimizes operational parameters across multiple cycles in response to the reservoir’s evolving state to enhance recovery efficiency. For comparison, a cyclewise particle swarm optimization (PSO) optimizes parameters independently for each cycle, while a fixed-strategy PSO is implemented as the baseline. Numerical simulations were conducted for a multifractured horizontal well in tight formations. The model incorporated the coupled effects of interfacial tension (IFT) reduction and wettability alteration through a K-value-based approach. A long short-term memory (LSTM) proxy model was trained using synthetic data generated by a history-matched numerical model. Net present value (NPV) is used as the objective function for all optimization strategies. Results indicate that both adaptive optimization strategies, RL and cyclewise PSO, outperform the fixed-strategy PSO. Cyclewise PSO leads to higher oil recovery in the early cycles but shows a sharp decline in later stages, potentially resulting in early termination. RL-based strategy maintains more stable production over successive cycles, allowing for extended economic operation and a higher overall NPV. Cyclewise PSO may be preferred for development plans aimed at fast returns, while the RL approach is more suited to long-term project goals. Cycle-to-cycle comparisons reveal the main reasons from the underlying physics perspective that explain the differences between RL and cyclewise PSO results. The improved enhanced oil recovery (EOR) performance of the RL strategy over extended cycles is attributed to higher early-cycle surfactant concentrations that enhance oil mobility and areal sweep efficiency, while increased injection volumes in later cycles maintain productivity, sustain the sweep advantage, and further reduce residual oil saturation.