查詢結果分析
來源資料
頁籤選單縮合
題 名 | 自調諧神經網路在非線性程序系統線上控制的應用 |
---|---|
作 者 | 張克仁; 張偉德; | 書刊名 | 高雄工商專校學報 |
卷 期 | 25 1995.12[民84.12] |
頁 次 | 頁61-82 |
分類號 | 446.014 |
關鍵詞 | 自調諧神經網路; 非線性系統程序; 控制演算法; |
語 文 | 中文(Chinese) |
中文摘要 | 從傳統的控制理論發展來看,大部份的研究都是針對線性系統。若想做到非線性 系統的控制。首先必須基於非線性系統在某一個狀態下,對此系統做線性化 (linearization) 的工作,有了線性化的模式後,再利用線性系統理論來設計控制器,進而 分析整個系統的性能 (performance) 與穩定性,(stability),這是一般控制非線性系統的 步驟。不過線性化的模式是基於系統某個狀態下所獲得的,所以控制的有效區間只能在此附 近,當超過某一界線時,因線性化的模式與非線性系統本身相差太多,所設計出來的控制器 可能無法有效地操控系統,此乃利用線性化技巧的缺點。本文以自調諧神經網路為基礎,基 於直接式的控制架構,來探討對非線性程序系統的控制情形。首先我們將自調諧神經網路當 成控制器,串接在非線性程序系統的前級,網路的輸出即為控制力,因為自調諧神經網路有 可調飽和準位與斜率的參數,因此具有下列幾個優點: 1. 飽和準位的修正會依設定點及非 線性程序系統自動做調整,以提供適當的控制力輸出,在此並不用考慮到尺度調整的問題, 使得在控制過程中更為直接與方便。 2. 因為有可調斜率大小的參數,所以在控制進行時, 非線性飽和函數的斜率會往最佳的方向去調整,使得網路輸出不至於飽和掉,造成整個系統 的停滯。 |
英文摘要 | In this study, we consider the issue of applying neural network to nonlinear process control. To achieve this goal, a new type neural network which allows the shape tuning of the nonlinear saturated function (activation function) as well as the weights tuning is proposed. An associated autotuning algorithm is derived based on steepest descent method for adjusting the weights between neurons, the saturation level and slope of the nonlinear saturated function simultaneously. Furthermore, the prescaling and postscaling of the input -output training data are not necessary, which are usually unavoided for commonly used constant nonlinear function shape multilayer perceptron with back propagation algorithm. Due to its good features and advantages, we adapt the new type neural network as process controller in the replacement of conventional ones for on line operation. Based on conventional feedback control architecture, an autotuning algorithm is derived to adaptively abjust the parameters of the controller in order to provide appropriate input to a process to reach some desired outputs. The necessary and sufficient condition that guaranteed the convergence of the autotuning method is presented. the excellent performance of the new type neural controller was demonstrated by controlling a nonlinear CSTR. Based on the simulation results and dus to its simple structure, efficient autotuning algorithm and excellent performance, it is convinced that the proposed scheme has high potential for practical on line control of complex nonlinear chemical processes. |
本系統中英文摘要資訊取自各篇刊載內容。