Sistema de apoio à decisão no desenvolvimento de um campo de petróleo com injeção de água usando aprendizagem por reforço

Abstract

For the development of an oil field the operational solution is quite complex, due to the high amounts of variables involved in the process, such as: well spacing, well numbers, fluid injection system for supplementary recovery, among others. In this context, this research presents the implementation and the application of an intelligent decision support system searching for alternatives for the development of an oil field, submitted to the water injection, using Reinforcement Learning (RL) based on Q-Learning algorithm. The environment considered was a homogeneous, semi synthetic reservoir simulation model, with some similar data from a reservoir located in the Brazilian Northeast. The analyzed model corresponds to a reservoir of dimensions 400m x 400m x 26m size, permeabilities between 40mD and 400mD, porosity of 23%. The learning algorithm was applied, along with the mathematical simulator STARS (Steam Thermal and Advanced Processes Reservoir Simulator) from CMG (Computer Modeling Group), the optimal policy (decrease or not change the initial water injection rate), in a timely manner, to provide as reward the highest Net Present Value (NPV) in a production horizon of 240 months. The results show that the optimum water injection approach was obtained using the implementation of RL allowed significant increases in NPV and the Oil Recovery Factor at the end of the project, as well as improving profitability, reducing costs of injection, treatment and disposal of the produced water, thus increasing the time feasibility of the project.