Research on intelligent regulation of layered water injection based on reinforcement learning SAC algorithm

• Published in: Scientific reports Vol. 15; no. 1; pp. 33044 - 17
• Main Authors: Hu, Jinzhao, Jia, Deli, Liu, Shichu, Wang, Wenchang, Ren, Fushen
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 26.09.2025; Nature Publishing Group; Nature Portfolio
• Subjects: 639/166/988; 639/4077/4082/4095; 639/705/117; Algorithms; Artificial intelligence; Automatic allocation algorithm; Control algorithms; Control systems; Deep learning; Efficiency; Energy consumption; Flow characteristics; Flow velocity; Humanities and Social Sciences; Injection; Intelligent optimization; Layered water injection; multidisciplinary; Neural networks; Oil and gas fields; Oil reservoirs; Reinforcement learning; Robotics; Robots; SAC algorithm; Scheduling; Science; Science (multidisciplinary); Water flooding; Working conditions; Intelligent optimization; SAC algorithm; Layered water injection; Automatic allocation algorithm; Reinforcement learning
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-025-11521-w

Water injection is a key technology to maintain oil reservoir pressure and guarantee high and stable oilfield production. In this paper, we propose an intelligent control method based on reinforcement learning for layered water injection columns, which achieves accurate control by constructing a layered water injection model and algorithm. A water injection column flow model incorporating the structure of the water injection device, column pressure loss, and flow characteristics is established, and the intelligent control system is designed by combining it with the SAC algorithm. The training environment and simulation platform are built using PyTorch, and the performance of SAC, PPO, and DDPG algorithms are compared and verified in different water injection sections. The experimental results show that the average regulation error of the SAC algorithm under the 5% error threshold is only 5%, significantly better than PPO’s 37%; the qualified rate of regulation reaches 81%, which is much higher than PPO’s 45% and DDPG’s 60%; the average number of adjustment steps is 42% fewer than PPO’s and 28% fewer than DDPG’s; SAC algorithm exhibits more substantial stability and adaptability under complex working conditions, and its regulation accuracy, qualified rate and response speed are better than that of PPO and DDPG. This study provides theoretical support for intelligent layered water injection technology, which has significant reference value for improving the development efficiency of oilfields. Our code will be available online at: https://github.com/HJZ-hub/SACIntelligentLayeredWaterInjection .