模糊可微分小腦模型控制器之設計與應用研究

Abstract

本論文提出一個模糊可微分小腦模型控制器（FDCMAC），它是結合模糊邏輯控制器（FLC）與可微分小腦模型控制器（DCMAC）之學習控制架構。模糊邏輯控制器採用模糊知識庫來描述一個系統的控制邏輯，在實際控制上模糊邏輯控制器比起一般傳統控制方法擁有更好的強健性與適應性。但是，模糊邏輯控制器的缺點是模糊知識庫需採嘗試錯誤法來建立且有穩態誤差，無法保證達到精確控制。可微分小腦模型控制器是一種應用查表方式的類神經計算技術，對於非線性函數具有快速的學習收斂速度和良好的區域性類化能力。藉由可微分小腦模型控制器的加入，可以改善模糊邏輯控制器的缺點，縮短以嘗試錯誤法來設計模糊知識庫的時間，並進而提昇控制系統的效能。經由模擬結果證實，在簡單的模糊邏輯控制器設計方式下，本控制器可以明顯降低系統的追蹤誤差，並有效地提昇控制精確度。最後，將本論文所提之控制架構實際應用於線性壓電陶瓷馬達（LPCM）位置控制，結果證實具有良好之控制性能和強健性。

This thesis proposed a fuzzy differentiable cerebellar model articilation controller (FDCMAC). Its main method is to combine fuzzy logical controller (FLC) and differentiable cerebellar model articulation controller (DCMAC). FLC usually uses a fuzzy knowledge base to characterize its control logic for a given system to control. As compared with conventional controllers such as PID controller, FLC can provid better robustness and adaptation in practical control. Its fuzzy knowledge base is created by trial and error. It has steady state error, so it may not guarantee precise control. DCMAC is a table look-up neuron-computing technique. It performs well in terms of its fast learning speed and local generalization capability for approximating nonlinear function. Compared with the FLC, this new controller shortens the design process of fuzzy knowledge base by less trial and error, and improves performance of the control system. According to simulated results, this controller can significantly reduce the tracking error and effectively elevate the accuracy in control process. At last, the experiment results for linear piezoelectric ceramic motor (LPCM) drive system with proposed controller has performed to demonstrate a high performance and robust control system.