比例 (Proportional) ─ 積分 (Integral) ─ 微分 (Derivative) 控制器 (PIDcontroller) 是一個被廣泛使用在工業控制的方法。然而這種控制器面臨最佳化調變的問題。現存的調變法,例如著名的齊格勒-尼科爾斯 方法 (Ziegler-Nichols method),雖然改善了人工調變的不穩定性,但是隨著控制精度需求的提高,這種方式已經不足以應付需求。對於非線性的控制系統,許多研究結合了類神經網路與 PID 控制器。由於 PID的最佳參數通常是未知的,因此許多有名的類神經網路架構並不適合用於這個問題。在本文中,我們設計一種基於雙神經網路的最佳化方法。比較起現存的最佳化法,在傳統化學工廠的實驗中,就是水壓控制與氣壓控制系統的結果顯示我們的方法可以避免局部最小值的問題。 ;Proportional–integral–derivative (PID) controller is wildly adopted in industry controller. Industrial controllers suffer from optimal tuning problem. Existing method driven approach like Ziegler-Nichols method is insufficient with the requirement of high accurate control. Due to the nonlinearities of PID tuning, many studies combine neural network with PID controller. Since the correct/best PID parameter are usually unavailable, makes many popular neural networks are not applicable. In this thesis, we design a duel neuron network based optimizer. Compared to existing optimizers in neural network, the experiment results of water pressure control and of steam relief control in a chemical factory show that our optimizer can avoid local minimum problem.