<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=Windows-1252">
<style type="text/css" style="display:none;"> P {margin-top:0;margin-bottom:0;} </style>
</head>
<body dir="ltr">
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<b style="color: inherit; font-family: inherit; font-size: inherit; font-style: inherit; font-variant-ligatures: inherit; font-variant-caps: inherit; background-color: ;">Automated Well Construction Decision Making through Deep Reinforcement Learning</b><br>
</div>
<div id="Signature">
<div>
<div>
<div style="font-family:Calibri,Arial,Helvetica,sans-serif; font-size:12pt; color:rgb(0,0,0)">
<div><br>
</div>
<div>Dr. Qiuhua Liu, Senior Artificial Intelligence Engineer, Schlumberger<br>
</div>
<div><br>
</div>
<div>Thursday, Nov. 7th, 1 pm<br>
</div>
<div>Room 120J, Technology Annex (T1)<br>
</div>
<div><br>
</div>
<div><br>
</div>
<div>Abstract:<br>
</div>
<div>In today’s Oil & Gas industry, safe, efficient and consistent performance is a key requirement. For example, in directional drilling, determining the correct control sequence to optimize trajectory accuracy and operational efficiency is critical to optimize
performance. Deep reinforcement learning (DRL) provides a general framework for training computer-based agents to make optimal decisions customized for particular environments.<br>
</div>
<div><br>
</div>
<div>In this work, we prototyped an agent for motor directional drilling using the DRL framework. The agent perceives the state from the environment, and then decides the best action to take (to slide or to rotate) to achieve the maximum total rewards. The
environment is affected by the agent’s actions and returns corresponding rewards to the agent. The rewards can be positive (such as drilling to target) or negative (such as deviation from the planned trajectory, increased cost of drilling, and penalties associated
with action switching). Deep neural networks are employed to decide the action sequence and to maximize the total expected future rewards.<br>
</div>
<div><br>
</div>
<div>To train the agent, we created an artificial layered earth model environment, with variable depth ranges, and layer properties such as BHA directional responses. The drilling responses are simulated by a drilling simulator running in this environment.
The agent is trained with the simulator to drill thousands of plans and learns to optimize the steering decisions to provide accurate, efficient, and consistent results.<br>
</div>
<div><br>
</div>
<div><br>
</div>
<div>Bio:<br>
</div>
<span>Qiuhua Liu is a Senior Artificial Intelligence engineer at Katy Drilling Software Center of Schlumberger, working on drilling automation. She has been working with Schlumberger for 10 years . Her current work is to apply machine learning and automated
planning techniques for automated drilling and event detection. She has a B.S. degree in physics from Peking University, Beijing, China, and a Ph.D degree in machine learning from Duke University.</span><br>
</div>
<div id="Signature">
<div class="BodyFragment"></div>
</div>
</div>
</div>
</div>
</body>
</html>