The Neural Network Control Of Hysteretic Nonlinear System

The hysteresis is a nonlinear phenomenon with multi-valued mapping memory and undifferential characteristics.The lacking of compensation for the hysteresis effect would cause control problems such as static errors,vibration and even lead to instability of the positioning system where the closed-loop control is implemented.So studing on modeling and control of the hysteresis is meaningful in theory.The smart materials such as piezoceramic shape mamory alloys are widely used in some important area such as microelectronics manufacture,fibre-optical communication because of their high qualities.However,hysteresis that exists in these materials can result in the degradation of system performance.So,it's an important object to study on modeling and control of the hysteresis on the engineering.It is well known that the neural network has the universal approximation capability for the nonlinearities.Also,the derived neural model can be updated online to adapt to the change of environmental condition.In order to compensate the influence caused by hysteresis,the most common approach is to construct the inverse model cascaded with hysteresis.The inverse hysteresis models are developed for the above-mentioned hysteresis respectively.On the other hand,it's difficult to control the systems with hysteresis using the classical control theory or modem control because of the special structure of hysteresis.The Internal model control is a new control strategy based on the mathematic model of the process to design controller.It has gotten widespread attention in control field since being put forward,for its simple design,good control performance and superiority in system analysis.The dissertation is organized as follows:1.For the behaviour of hysteresis,this paper proposed a Hammerstein model which employs a linear dynamic sub-model cascaded with the neural hysteresis sub-model.Also,the corresponding optimization procedures for the model is given.The experimental validations have shown that the Hammerstein model is flexible.2.To slove problem that the inverse preisach model can not be updated online for accommondation of the environmental conditions,an inverse controller using neural networks is proposed.With the introduction of a nonlocal inverse hysteretic operator to represent the change-tendency of the hysteresis inverse,the multi-valued mapping of the hysteresis inverse is decomposed into a one-to-one mapping based on the expanded input space method,and the neural hysteresis-inverse model is established.3.A internal model control scheme and feed-forward PID compensation strategy is employed in this paper.The neural network internal model control scheme is designed by two neural networks.One is for the controlled object model and the other is model controller,Compared with experimental results of feed-forward PID compensation strategy,it is indicated that the internal model control scheme has higher control accuracy.