Reinforcement learning for hydrocarbon well optimisation

This problem involves not only optimization in partially unknown environments, but also adaptive actions depending on environmental feedback. E.g., the tasks of identifying optimal placements of pumps is convoluted as activity of one pump influences the performance of other pumps. This brings the need for adaptive strategies, such as reinforcement learning, which will be challenging due to the complex correlations between various reservoir  parameters. Recent progress in reinforcement learning involving deep models which was successful in other domains such as game play (AlphaGo) provide promise. In this project, we will further enhance the underlying learning models with near-term quantum architectures, which will allow us to tackle more complex correlations better, and provide the needed speed-ups to enable these automated procedures to make correct and safe decisions on-the-fly in real environments.