PID控制器是目前產業界應用最多的控制器，但其控制器參數調整不易，大多依賴專家調整，非常不便。本文提出一個自調式類神經PID控制架構，應用倒傳遞類神經網路理論，對於系統模型參數未知的情況下，使用兩個類神經網路分別進行系統鑑別與PID控制器參數調整。由電腦模擬結果可知，本控制架構能在很短的時間內調整出極佳的控制器參數。最後將此控制架構實際應用於超音波馬達位置控制上，實驗結果則顯示，本控制架構確實可以在實際控制應用上實現，其調整結果亦相當快速良好。 The PID controller has been used widely as a major control method in industrial applications. However, it is difficult to tune the PID gains during the controller development, and can only be carried out by expert with control knowledge and experience. This thesis presents a self-tuning PID controller based on the neural network theories. There are two multilayer neural networks within the self-tuning PID controller, one for system identification for unknown controlled systems, and the other for the PID gains determination. Back-propagation method is adopted to perform both the neural networks training. The results of computer simulation show that the neural based PID control scheme can tune suitable PID gains within a short period. In addition, the controller was implemented to the position control of an ultrasonic motor. The experimental results have shown that the control scheme is also practically successful.
