- 學位論文
直接類神經網路之適應性控制器在直流馬達及液壓伺服系統之應用
Direct Neural Controller Applied to DC Motor and Electro-Hydraulic Servo System
摘要
本文研究特定學習架構之直接適應性類神經網路控制器之原理及應用，並應用 適應法則於特定學習架構近似輸出層之倒傳遞誤差項，以訓練類神經網路，並且使用雙曲正切函數做為類神經網路活化函數，使類神經網路控制器具有正及負之輸出，將其應用於具有參考模型之控制系統構成直接類神經網路模型追隨適應控制器，若應用於無參考模型之控制系統則構成直接類神經網路自調節適應控制器。 本文提出之直接類神經網路自調節適應控制器應用於高精度之直流馬達速度，位置控制系統及具有外力負載之電液伺服閥控液壓位置控制系統，數學模擬及實驗證明可以經預先訓練得到較佳之神經鍵初使值，再經由線上學習得到更好的調整，可使運動控制系統獲得精確且快速穩定之反應。同時也將線上訓練類神經網路模型追隨控制器應用於電液比例閥控可變排量軸向型柱塞泵斜板偏轉角控制，系統具有快速的學習能力，能強化系統適應性及強健性，在負載變動及不同偏轉角命令下，使斜板偏轉角反應對參考模型有快速之追隨能力。
並列摘要
A direct adaptive neural network controller with specialized learning architecture and its applications are studied in this research. The adaptation law is applied to the direct neural controller for approximating the term of the output layer so that the back propagation iteration can be executed. An arctangent function is applied to be the activation function so that the neural network controller output has negative or positive value. The proposed direct adaptive neural network controllers without reference model are applied to speed and position control of DC motors and position control of Electro-hydraulic servo systems. Simulation shows that a previous training of the neural controller can learn the approximate behavior of the plant and create better initial weights, then followed by on-line trained to fine-tune the network in the operating process. Experiment shows stable and fast responses can be achieved. The same controller with reference model is applied to control the swash plate angle of a variable displacement axial piston pump, which is nonlinear, time variant and with load disturbance. Mathematic simulation and experiment show that the direct adaptive neural network controllers enhance adaptability and robustness of the system and improve the