Research On A Fully Closed-loop Intelligent Drilling Guide D Method Based On Reinforcement Learning

Guided drilling is the most costly and most technically intensive part of oilfield exploration and development.The existing drilling guided methods are mainly for semiclosed loop downhole drilling while drilling.It includes surface analysis and decisionmaking and downhole data collection.Through real-time data bidirectional transmission,surface and downhole operations cooperate with each other to perform drilling guides to control the borehole trajectory.However,this method is highly dependent on signal transmission speed and transmission efficiency,and the downhole environment is complex.It is almost difficult to achieve effective data transmission in deep wells and ultra-deep wells far away from the ground.In addition,the analysis and decision-making link on the ground involves complex human expert analysis and fine management,and the labor cost is relatively high.In view of this problem,this thesis proposes a fully closed-loop intelligent drilling guide method based on reinforcement learning.This method can make the drilling guide work completely downhole and automatically adjust the control actions of the drill bit according to the drilling environment to reach the target formation.main tasks as follows:(1)Establish a simulation drilling interaction mechanism based on data generation.This mechanism can dynamically generate simulation data based on a small number of geological data samples and construct a three-dimensional geological environment in real time.By training and testing intelligent drilling methods through multiple interactive methods such as data while drilling,formation profile images,etc.,the actions issued by the drilling methods can be executed and the changes in the drilling environment can be fed back in real time according to the actions.It has the advantages of low latency,low cost and timely feedback.(2)Propose a method of underground fully closed-loop intelligent drilling based on reinforcement learning.The key links such as the drill encounter environment,drill guide action,reward function and update mechanism are defined in detail,and a complete intelligent drill guide method is constructed.This method can process multi-dimensional data while drilling,and evaluate the drilling status and decision of drilling direction online to guide the drill bit to the target formation.According to experiments,as the training progresses,this method can guide the drill bit to drill into the target formation more frequently.(3)Design an adaptive downhole fully closed loop intelligent drilling model.Considering the variety of formation changes and the higher requirements for drilling rate.Based on the intelligent drilling guide method,this thesis proposes an adaptive intelligent drilling guide model.The model can be well matched with the interactive mechanism of simulated drilling.It can process the formation profile image,and further extract the formation features through the adaptive mechanism,so as to give the drilling direction decision.Through comparison experiments and result analysis,compared with the method of downhole fully closed-loop intelligent drilling guidance,the adaptive downhole fully closed-loop intelligent drilling guidance model can obtain a higher drilling rate in the same formation.(4)Develop an intelligent drilling guide simulation system that combines simulated drilling interaction mechanism and intelligent drilling guide model.The system can select different drill guide models and define training parameters independently.Observe the simulated drilling conditions of the drilling guide model in the current formation from multiple panes such as 3D drilling environment,2D formation profile and real-time data display.