比例-積分-微分控制器可應用於很多工業自動控制系統,這種控制器的實用性源自於無須鑑定受控系統的動態模型即可調諧控制器參數,而且可以獲得優良的閉迴路系統性能,但是這種控制器的參數調諧方法其實是根據低階線性動態系統的經驗數據建立的,若實際的應用對象偏離低階或線性的假設,則經驗數據所建立的參數調諧方法並不能保證閉迴路系統的控制品質。本論文根據適應型最佳控制理論推導出比例-積分-微分控制器的循序優化法,此循序優化法具有強化學習架構,可以透過學習程序建立評議函數並用以引導優化動作,優化的目標以成本函數表示,使用者可以選擇特定的成本函數來指定閉迴路系統的響應速度和穩定度,本論文並以應用例的電腦模擬結果檢驗控制器循序優化法的效用,線性應用系統採用傳統型比例-積分-微分控制器,非線性應用系統則採用模糊型比例-積分-微分控制器,結果顯示控制器循序優化法可以透過學習程序自動調整控制器的參數,使成本函數逼近最小值。
The PID controller, which consists of proportional, integral and derivative elements, is commonly used in closed-loop control of industrial processes. A nonlinear system can be satisfactory controlled by a PID controller without the need for accurate mathematical model of controlled object. However, the parameters of the conventional PID controller are not often properly tuned for highly nonlinear systems with uncertain parameters. In this paper, a sequential optimal tuning method using adaptive optimal control (AOC) for PID controllers is proposed. With reinforcement learning architecture in the method, the adaptive critic is trained to predict the future system performance and the actor is optimized for the control. The control performance objective can be described in terms of cost function. By defining the cost function to specify desired response and stability of closed-loop system, the demand control performance can be achieved. The effectiveness of the sequential optimal control is verified by several simulation results, where a PID controller and a Takagi–Sugeno (T-S) fuzzy PID controller based on AOC algorithm are used to control a linear and a nonlinear system correspondently. The results show that the proposed method can sequentially optimize the controller behavior by the learning procedure which can automatically adjust the controller parameters and the cost function is explicitly minimized.