Research On Simulation And Intelligent Control Method Of Frequent Gas Kick And Drilling Fluid Leakage Condition

Controlling the bottom hole pressure within the pressure window to prevent abnormal conditions such as gas kick and drilling fluid leakage,is the key to safe and orderly drilling.When drilling into complex pressure formation,the safety pressure window becomes narrow.Moreover,due to the complexity of the downhole,it is impossible to calculate the appropriate wellhead back pressure by using the existing hydraulic model and suppress the abnormal condition through automatic control.Under those conditions,engineers can only manually observe the measurable engineering parameters,such as inlet flow and outlet flow,and adjust the valve opening according to their experience in pressure control.To solve the problems,this thesis studies the intelligent control method of frequent missing conditions,and the main contents are as follows.Firstly the drilling technology and the causes of overflow are briefly introduced,and the overflow model is established according to those information.The model consists of the bottom hole pressure hydraulic model and the overflow leakage rate model.The bottom hole pressure model is used to simulate the change of bottom hole pressure when overflow leakage occurs;the overflow and leakage rate model is used to represent the gas invasion or drilling fluid leakage rate when the bottom hole pressure changes or the formation pressure changes to produce a pressure difference.Secondly,the deep reinforcement learning method is introduced as the intelligent control method.The theory and advantages of reinforcement learning are introduced,and different types of deep reinforcement learning algorithms such as the DQN method based on value function,REINFORCE method based on policy,and Actor-Critic method are compared.After the comparison,the suitable algorithm DQN is selected for the control.Thirdly,the experience data operated by engineers are added into the experience buffer of deep reinforcement learning to train the neural network to learn the experience of engineers and realize the intelligent control of abnormal conditions.In view of the insufficient amount of empirical data of manual operation in the process of managed pressure drilling and the difficulty of convergence of neural network parameters to the optimum,the method of fine-tune is used to update the parameters by using the existing empirical data to correct the deviation in the value function.The robustness of the algorithm is discussed when the environment(i.e.spill model)changes.Finally,the engineering application of the DQN intelligent control method is studied in this thesis.The system objectives and functional requirements are analyzed,and the overall architecture and functional requirements architecture are designed.Further works such as the database design,the communication of equipment,the intelligent control algorithm packaging,algorithm embedding,and the training process design are carried out,and the application steps of the intelligent control algorithm are designed and explained.