Maximizing efficiency of deep-reinforcement learning agents in autonomous directional drilling with hyperparameter optimization

摘要: The lack of commercially reliable communication bandwidth and hardware capabilities downhole has hindered autonomous directional drilling methodologies. However, recent advancements in intelligent autonomous technologies developed in other industries have proven to be reliable even with limited amounts of measurements. In this paper, we aim to use advanced artificial intelligence (AI) methods such as deep reinforcement learning (DRL) to improve their performance for geosteering problems. The drilling agent mimics the combined effort of a directional driller on the surface and an operational geologist, executing procedures such as staying on the zone following a predefined wellplan and reducing tortuosity while optimally determining key drilling performance indicators. We present a directional drilling environment based on a robust physics-based simulation engine that can be modeled as a Markov …