Reinforcement Learning Breaks Ground in Fracturing Applications

An industry first reinforcement learning (RL) agent that directly interacts with the subsurface during a fracturing operation was developed and deployed across several offshore fracpack operations. This involved training an RL agent to yield optimal, consistent decisions during fracpack execution that mimicked a fracturing expert’s decision making process based on decades of experience over hundreds of fracpack treatments. The agent effectively removed inherent psychological biases during execution by analysing real time data and basing the decisions specific to fracturing behaviour of rock based on surface parameters like pump rate and fluid quality. No offshore fracturing treatment is similar and having a consistent advisory system that relies on real time data enables more efficient outcomes. These RL concepts are scalable to various other expert disciplines that relies on experience and muscle memory, thereby allowing a wider range of people to train, utilize and develop capabilities in complicated engineering environments.