Research Of Wellbore Scanning Control Based On Reinforcement Learning

With the rapid development of science and technology represented by information technology,automation equipment is developing in the direction of fast response and high precision.During the operation of multi-axis automation equipment,both tracking errors of uniaxial motion and contour errors of multi-axis will be occurred.These two errors,especially contour errors,directly affect the trajectory of the equipment.Contour error is an important index to measure the machining performance of high precision and high speed multi-axis feed system.How to maintain small contour errors in continuous trajectory motion with large curvature and fast speed is an urgent problem to be solved in current multi-axis motion control system.Contour error control has always been an important research topic in high precision and high speed machining.The cross-coupling control strategy can solve the problem of contour error accuracy caused by the dynamic influence between the axes and has attracted attention widely.The traditional cross-coupled PID controller has simple structure and high robustness.However in practical applications,the structure of the controlled object is relatively complex and highly nonlinear.Under the disturbance or noise,the control accuracy of the traditional PID controller can not meet the requirements of applications.Aiming at the characteristics of PID control,this paper studies and analyses the shortcomings of traditional controller.Then it combines with the advantages of reinforcement learning to propose an improved PID controller,which is applied in cross-coupling control to achieve higher contour error control accuracy.Lastly,it proposes a feasible technical scheme for intelligent control.This paper summarizes and analyses the research status of reinforcement learning,PID control theory and contour error control technology at home and abroad.It studies the improvement of PID controller and the reasonable application of cross-coupling control.The main purpose of this paper is to improve the contour error control accuracy of wellbore scanning device.Firstly,the paper improves the PID controller by combining reinforcement learning algorithm and proposes an adaptive asynchronous advantage Actor-Critic PID controller,which improves the learning rate of the controller by using the CPU multi-threaded parallel training method and enhances the adaptive ability of the controller.The crosscoupling control structure is summarized and analyzed.The improved PID controller is applied to the cross-coupling control,which improves the error control accuracy of the two-axis circular contour.Finally,the improved cross-coupling control is applied to the simulation of the control system of the wellbore scanning experimental device,which reduces the contour error of the multi-axis of the wellbore scanning device and lays a solid foundation for the information processing of geo-steering.