Research On Active Control Strategy Of Metal Rubber Isolator Based On Reinforcement Learning

In recent years,the deep-sea development of offshore oil and gas has increased the complexity of loads that ships must endure,making vibration control of diesel engines,the primary power source for ships,particularly important.Passive isolation is a common method for vibration control,but the isolation effect of passive isolation systems is affected by changes in vibration frequency and amplitude and is not very stable.In contrast,although active isolation systems can enhance the suppression effect of double-layer isolation on low-frequency and wideband random vibration,they require parameter design and adjustment in advance and have a relatively weak ability to adapt to complex loads and uncertain environmental changes.Reinforcement learning control algorithms use autonomous interactive methods to learn control strategies,avoiding the need for additional domain knowledge and can adapt to changes in dynamic external loads and system parameters,providing a feasible way to solve the problem of the impact of complex loads and system parameter uncertainty on isolation systems.Therefore,this paper aims to explore the control effect of reinforcement learning active isolation control strategies in solving the problems of external load changes and uncertainty in isolation system parameters,and to provide a reference for active control of isolation systems.The main contents of the study are as follows:Taking the vibration isolation of marine oil storage and transportation ship diesel engines as the background,metal rubber is designed as the isolation unit,and a metal rubber doublelayer isolation model and control system state space equation are established on this basis.Combined with dynamic loading experiments of metal rubber isolators,the effects of changes in isolation parameters and their uncertainty caused by loading amplitude and frequency on isolation performance are studied.Combining Q-learning policy in reinforcement learning with active control theory of double-layer isolation,an active isolation control method based on Qlearning reinforcement learning algorithm is proposed.Through simulation analysis,the effectiveness of the Q-learning algorithm is verified,and the influence of key parameters in the algorithm on isolation effect is analyzed.In view of the limitation that the Q-learning algorithm’s discrete values are not suitable for continuous variables,the DDPG reinforcement learning algorithm is selected to further optimize the control strategy,in order to achieve active isolation in continuous state space and action space.Through simulation comparison of robust control,fuzzy PID control,PID control,and reinforcement learning algorithm control of vibration amplitude and acceleration root mean square,this paper verifies the superiority of the reinforcement learning algorithm based on active isolation in the face of load changes and uncertainty in isolation system parameters.The metal rubber double-layer isolation method based on the reinforcement learning control strategy proposed in this paper reduces the impact of external excitation changes and system parameter uncertainty on active isolation systems and provides a feasible choice for active strategies of uncertain isolation systems under complex loads.Through the study,it was found that reinforcement learning active control strategies can alleviate the impact of complex marine loads on ship diesel engine vibration.