Fault-Tolerant Control For De-Oiling Hydrocyclone Systems Based On Game Reinforcement Learning

The research on fault-tolerant control of de-oiling hydrocyclone system is a new hot topic in the field of industrial oil and gas production.Reinforcement learning is one of the effective methods to solve the optimal control problem in recent years.In this thesis,the shortcomings of existing fault tolerant control methods are analyzed.Considering the reliability and stability requirements of the actual de-oiling hydrocyclone system,as well as the practical characteristics of modeling difficulties,an off-policy reinforcement learning fault tolerant control algorithm based on state feedback and output feedback is proposed respectively.The main work of this thesis is as follows:Most of the current researches are focused on the tracking control of de-oiling hydrocyclone system,and few literatures consider the actuator failure.Based on the effects of external disturbances in the actual system deoiling process and valve failures,including loss of effectiveness and bias.In this thesis,the coordinated control problem of the de-oiling hydrocyclone system is modeled as a two-player zero-sum game problem,and a control algorithm based on reinforcement learning is proposed to learn the optimal control solution.On this basis,a model-free fault-tolerant controller is designed to recover the tracking performance degradation caused by the failure of the system actuator.Ensure that the system has good fault tolerance performance and take into account the optimal control performance of the system.Finally,the stability of the closed-loop control system is analyzed and its effectiveness is verified by simulation.It is further considered that the sudden failure of the system may cause the change of the system structure,which leads to the uncertainty of the model parameters,and the uncertainty is time-varying.In order to ensure that there is no need to re-learn when parameters are perturbed,adaptive methods are introduced into fault-tolerant control and adaptive law is designed to estimate the uncertainty of fault signals and time-varying parameters.The online estimation information provided by adaptive law is obtained based on off-policy reinforcement learning algorithm and does not require the knowledge of system dynamics.On this basis,a fault tolerant compensation controller is constructed,and a model-free fault tolerant compensation control scheme is given to track and ensure the stability of the system.Finally,the effectiveness of the designed adaptive law and controller is verified by simulation.Finally,Consider some states of the actual industrial de-oiling hydrocyclone system are unmeasurable,such as pressure drop and change rate of pressure drop.Therefore,this thesis improved the control strategy and proposed the optimal control law of the learning system based on output feedback reinforcement learning algorithm.On this basis,combined with the estimation information of the adaptive law,the fault-tolerant controller was designed.This method only uses the limited historical output data obtained from measurements,and does not require the dynamics knowledge of the system,so it is more suitable for practical industrial de-oiling hydrocyclone systems.