Study On Speed Control Strategy For Gas Turbine Engine Based On Reinforcement Learning

As "the pearl in the crown of equipment manufacturing industry",the development level of the gas turbine is the comprehensive embodiment of the military and industrial strength of various countries.The gas turbine has complex characteristics such as nonlinear and time-varying,and stable rotational speed is an important symbol of its safe operation.As artificial intelligence theory and the rise of big data technology,combined with a data-driven intelligent control method has become a hot research direction,because of the traditional problem of failure and self-tuning control parameters,the artificial intelligence theory combined with the traditional control theory,to overcome the past difficult problem of the complex object of control parameters self-tuning,In this paper,a control strategy combining q-learning reinforcement learning intelligent algorithm with traditional control technology is proposed,which is applied to improve the speed stability control operation and safety performance of gas turbine and the system self-healing ability when control parameters fail.In this paper,a kind of three-axis gas turbine is taken as the research object.Firstly,modeling is carried out according to the working principle of the gas turbine.Because of the problem that the inherent characteristics of gas turbines such as nonlinear,strong coupling,and time variability lead to the speed instability due to the failure of control parameters under various working conditions,And the influence of dynamic response demand,technical state decline,external interference and other factors on gas turbine working state under different sea conditions.The control strategy based on reinforcement learning theory is designed.The q-learning table learning algorithm in reinforcement learning theory is absorbed and improved,and the intelligent control algorithm based on PI control is designed.The PI control parameters and the actual speed in gas turbine operation are taken as the environmental state of the input agent,and the adjustment of PI control parameters is taken as the action output by the algorithm.The artificial debugging process is replaced by the intelligent algorithm to realize the application of artificial intelligence theory to engineering practice.The algorithm designed in this paper can judge the trigger in the gas turbine operation process,train itself,take the training results as the standard to update the control action library,and manage the control parameters during the gas turbine operation to ensure the stable operation of the gas turbine in various emergencies.After verifying the intelligent algorithm of the software layer,the hardware-in-the-loop test of a speed control strategy based on reinforcement learning is completed according to the fast control prototype and automatic code generation toolbox.As can be seen from the simulation results and experimental data,when the gas turbine is unstable,the speed stabilization can be completed in a short time through the automatic optimization of the control parameters by the enhanced learning algorithm,and the speed error can be controlled within 2rpm,thus realizing the self-rescue when the speed is unstable.