類神經網路應用於混沌系統的系統辨識及直流馬達控制器之設計

Abstract

本論文分成二部份，第一部份研究類神經網路應用於混沌系統的系統辨識 、第二部份研究類神經網路應用於直流馬達控制器之設計。 第一部份，首先我們利用類神經網路做系統辨識，再以此系統辨識為基礎 ，設計新的估測器。此類神經網路是以估測值為輸入，並且加上誤差補償 器，而誤差補償器的功能與線性卡門濾波器的卡門增益相同，可以克服並 聯模式中誤差會因疊代過程而放大的缺點。我們稱此新型的估測器為以卡 門濾波器模式為基礎的類神經網路估測器（NNKF）。估測的結果顯示，只 要系統辨別準確性夠高的話，此估測器有良好的效果。為了幫助我們暸解 混沌現象，我們設計一混沌電路來觀察此奇特現象。 第二部份以DC伺服馬達為研究對象，首先了解DC伺服馬達的物理特性及其 數學模式。根據由量測所得的數據經由系統判別方法找出系統轉移函數， 由轉移函數找出近似的馬達參數，並以此模型為主要的依據，採用PI加反 向動態模式（混合型）之設計方法進行設計。採用混合型設計的方法主要 是利用PI控制器控制暫態響應，反向動態控制器控制穩態響應，可以改善 馬達上升時間及穩態值。藉由PC－MATLAB 從事控制器的分析，設計及模 擬，而後應用於實際控制上，設計之控制器對原系統有良好的改善效果。 This thesis is divided into two parts. Part I is to study the chaotic system identification by neural networks. We purpose a new dynamical predictor model that the inputs of the neural network are fed with prediction values and is updated by an error compensator. The error compensator, whose function is the same as that of the Kalman gain in linear Kalman filter, can overcome the dis- advantages of a Kalman filter predictor. The prediction results show that even the plant is in the chaotic state, the predictor also work well as long as the basic identification model is trained accurately enough. A circuit implementation of the Lorenz system is provided to show the work results. Part II is to study the DC motor control system. First of part II, the mathematical model is generated according to its physical characteristic. We design an experiment to get the transfer function and the parameters in this model by the measurement from the DC motor using least squared method of system identification. We can get transfer function successfully. The part II describes a hybrid(PI and Inverse model) control scheme for DC motor system. The essence of the scheme is to divide the control two stages. For transient response, a PI controller is used, which drives the system output into a pre-defined neighborhood of the steady-state values. The controller then switches to the Inverse model control stage. For steady-state response, an Inverse model is used, which drives the system output into the steady- state value. The experiment results of the control system are pretty good.