Intelligent Control For Several Constrained Systems

The task environment of nonlinear systems such as robot,is complex and changeable,and the control requirements are gradually increasing,which makes the system have to face many control problems.Constraint problem is a typical example.The restriction of the system is mainly reflected in two aspects,one is the constraints imposed by the external task environment on the output system state,and the other is the input or output constraints imposed by the internal hardware conditions of the system.In addition,the system uncertainty is also a common problem,which makes the model-based controller cannot work normally and accurately.With the rapid development and wide application of the control technology,researchers have conducted extensive and in-depth work on the problems of constraints and uncertainties.Since there are many types of actual controlled systems in practical applications,such as aircraft,ground mobile robots,robotic arms,etc.,different types of systems correspond to kinematics and dynamics models with different structures and parameters,which indicates that the research on the specific systems has application restrictions and cannot be applied to other systems.In the face of increasingly complex operating environments and increasing task requirements,there are also higher requirements for the learning ability of systems.However,the traditional classical control theory cannot meet these requirements.In view of the good nonlinear fitting ability of the neural network,combined with traditional adaptive control technology,adaptive neural network control is often adopted by researchers to solve the uncertainty and partial nonlinear problems of the system.Although the conventional adaptive neural network control can make the system complete the corresponding task through learning,it cannot solve the problem of optimal control.With the development of artificial intelligence technology and control theory,intelligentization has become an important development direction of control.The control system not only needs to have learning ability,but also needs to have certain optimization ability.Because adaptive dynamic programming and reinforcement learning have strong learning and optimization ability,they are widely adopted to solve the optimal control problem when the agent interacts with the environment to achieve specific goals.The essence of applying the principles and ideas of adaptive dynamic programming and reinforcement learning to the control field is an optimal control method for continuous state and action space.In reinforcement learning,the ideological framework of actor-critic learning plays a vital role.However,the reinforcement learning algorithm in machine learning is weak in interpretability.Combining the ideological framework of actor-critic learning with adaptive neural network control technology has become an interesting and meaningful research topic.This topic is not only beneficial to handle the problem of weak interpretability of the actorcritic learning but also to improve the learning and optimization abilities of the system.For the control system,stand-alone control is the cornerstone,and collaborative control is the development trend.The research of stand-alone control and collaborative control is equally important.In this paper,starting from the constraint problem of the system,we systematically study the intelligent control from the certain system to the uncertain system.The main research contributions of this paper are shown as follows:(1)An improved critic learning based optimal control algorithm is proposed for a class of input-constrained nonlinear systems.Considering the environmental conditions and requirements of practical applications,prescribed constraints are imposed on the system output states to guarantee the control performance and normal operation of the system.An error transformation function is adopted to cope with the prescribed constraints and generate an equivalent unconstrained error for the convenience of the intelligent control design.In order to improve the learning ability and optimize the control performance,critic learning is introduced to the control design based on the transformed equivalent unconstrained system.By introducing an auxiliary Lyapunov function,the system dependence on the initial admissible stabilizing control is removed.The stability of the closed-loop control system is proven by the Lyapunov stability theory.(2)An improved critic learning based optimal control algorithm is proposed for input-andoutput constrained nonlinear systems.A transformation function is introduced to transform the constrained states into unconstrained states.Based on the transformed states,according to the one-to-one relationship of the monotonically increasing function,the system cost function is redesigned based on the input constraints.The input-and-output constrained optimal control is obtained by solving the redesigned cost function.In addition,through the auxiliary Lyapunov function and the stored historical data of the system,the updating law of the critic neural network is improved,and the dependence on the initial admissible stabilizing control and external persistence of excitation condition is removed,which expands the scope of application of the critic learning.Furthermore,the system stability of the closed-loop control system is also proven according to the Lyapunov direct method.(3)An adaptive neural network control algorithm is proposed for nonlinear systems subject to input dead-zone,output constraint and uncertainties.For the actual system,the model uncertainty caused by the slight change of the inherent parameters of the structure is unavoidable,and the dead-zone nonlinearity often occurs in the input of the system.In addition,due to environmental space or control requirements,the output state needs to be within a certain range.Neural network technology is adopted to approximate the uncertainty of the system and the unknown term of the input dead-zone nonlinearity.The integral barrier Lyapunov function is adopted to handle the output constraint problem,and ensures the system stability.Combined with the adaptive control technology,the updating law of the neural network is designed.Furthermore,the stability of the closed-loop system is analyzed and achieved according to the Lyapunov stability theory.(4)An adaptive control algorithm with actor-critic design is proposed for nonlinear systems with input constraint and uncertainties.The actor-critic learning idea framework in reinforcement learning is introduced into the controller design of the input-constrained nonlinear system.The critic network aims to approximate a cost function,which is used to judge the control performance.The actor network is used to deal with the system uncertainty and the input nonlinearity,and then generates control input into the actuators based on the critic results.In the actor network,the problems of partial variable unmeasurability and input saturation constraint are handled by high gain observer and auxiliary variable design,respectively.In addition,based on the Lyapunov stability theory,the stability of the closed-loop intelligent system is proved.(5)Actor-critic learning based adaptive coordinated control is proposed for uncertain systems with prescribed performance and unknown input backlash-like hysteresis.The actor-critic learning idea framework in reinforcement learning is introduced into the coordinated control design.A dynamic model for the coordinated control system with unknown backlash-like hysteresis is established.The critic network is used to approximate the cost function for judging the task performance of the system,while the actor network guarantees the prescribed performance through the barrier Lyapunov function,and solves the unknown term of the input backlash-like hysteresis through neural network technology.The control input is also generated into the actuators based on the critic results.Based on the Lyapunov stability theory,the stability of the system under the cooperative control algorithm is guaranteed.