Intelligent Safety Control Of Electro-hydraulic Servo System Based On Deep Reinforcement Learning

As a typical nonlinear controlled object,the electro-hydraulic servo system is characterized by complex internal structure,variable working environment and high safety requirements.Since the current methods based on analytical models are difficult to accurately describe the dynamic characteristics of electro-hydraulic servo system,it can not improve the control performance and anti-interference ability.As a result,it is necessary to study the nonlinear robust control method of data-driven electro-hydraulic servo system.Therefore,we propose a SAC deep reinforcement learning control method based on optimized sparse rewards to improve the overall control performance of the electro-hydraulic servo system and a safety deep reinforcement learning control method based on a barrier function to achieve the preset safety guarantee goal of the electro-hydraulic servo system,which are based on the VSV electro-hydraulic servo closed-loop control and fault simulation test bench,and the characteristics of the modelfree asynchronous strategy iteration of the SAC algorithm.The main content of this article’s research and innovation are as follows:The first is one that we propose an optimized SAC deep reinforcement learning control method for electro-hydraulic servo system.Firstly,according to the input and output characteristics of the state direct electro-hydraulic servo system,we designed an optimized state space sparse reward method;Then we improve the convergence speed of the control algorithm use by random network distillation for reward shaping and select the appropriate optimized state space ratio,tuning the optimized SAC deep reinforcement learning controller for electro-hydraulic servo system;Finally,the actual electro-hydraulic servo system position control experiment verifies that this method has better control performance and anti-interference performance.The main advantage of this method is that it can solve the problem of designing the reward function in continuous control tasks by optimizing the sparse reward.The data-driven deep reinforcement learning method improves the performance and robustness of the controller.At the same time,it does not rely on accurate analytical models.The second is one that we propose a safety deep reinforcement learning control method to significantly enhance electro-hydraulic servo system.Firstly,according to the pre-built security requirements,we design the control barrier auxiliary reward function;Secondly,we use optimized SAC deep reinforcement learning control method turning controller which satisfies optimality and safety at the same time;Finally,We design an optimized SAC deep reinforcement learning safety controller for the electrohydraulic servo system,and the pressure mutation experiment of the electro-hydraulic servo system verifies the effectiveness of this method.Compared with the traditional method,this method does not need to design a steady-state safety controller independently and achieved preset security goals and maintained good dynamic performance.In summary,this article has achieved innovative results in many aspects such as optimized sparse reward design,optimized SAC deep reinforcement learning control method for electro-hydraulic servo system and safety deep reinforcement learning control method for electro-hydraulic servo system.We systematically established the deep reinforcement learning nonlinear robust control design method and safety control method of the electro-hydraulic servo system,which is of great significance to the application of deep reinforcement learning control in the actual engineering of electrohydraulic servo system.