近十年來，模糊類神經網路控制器已成功應用至各種不同的控制問題上。但是，模糊類神經網路只使用明確的歸屬度而無法包含語言不確定性，且使用前饋式網路架構只能具有靜態的響應。為了克服其缺點，本論文提出具有擾動項模糊類神經網路及具有迴授項模糊類神經網路兩種網路架構。藉由調整擾動項的振幅與頻率來克服人們對數值描述感覺的不確定性，以及利用回授項捕捉動態響應及訊息儲存的能力。並進一步，本論文提出適應性模糊類神經滑動模式控制器與適應性模糊類神經二階滑動模式控制器，運用上述所介紹的兩種新型模糊類神經網路架構來線上學習系統動態方程式。同時，本論文設計了一個平滑補償器來克服模糊類神經網路的學習近似誤差所造成的影響。最後，利用混沌動態系統及倒單擺擺動系統來測試所提出的控制方法，經由模擬結果驗證其可以獲得良好的控制結果。 In recent years, fuzzy neural network (FNN) has been developed. But the FNN have two major drawbacks, one is their application domain is limited to the static problem due to their feedforward network structure, and the other is their unable to directly handle the rule uncertainties due to the membership function is a crisp number. To attack this problem, this paper proposes a perturbed fuzzy neural network (PFNN) and recurrent fuzzy neural network (RFNN). Meanwhile, a fuzzy neural network sliding-mode control (FNSMC) system and a fuzzy neural network second-order sliding-mode control (FNSSMC) system are proposed. Since the RFNN has an internal feedback loop, it can capture the dynamic response with an external feedback. On the other hand, the PFNN uses a perturbed membership function to handle the information uncertainties when it is hard to exactly determine the grade of the value of a basis function. To cope with the approximator error, a smooth compensator is proposed to reduce chattering in the control input. Finally, a chaotic system and an inverted pendulum are applied to example studies. The simulation results show that the proposed two control methods can achieve favorable control performance.
