利用類神經網路進行線上自我調整PID控制器之設計

On-Line Self-Tuning PID Controller Design Based on Neural Network Models

黃天志(Tien-Chih Huang)

指導教授 ： 陳榮輝

中原大學/工學院/化學工程研究所/碩士(2002年)

https://doi.org/10.6840/cycu200200525

摘要

化工程序中多為複雜且常常呈現時變非線性的型態，而且對於外在影響有極高的敏感度，在正常操作狀態下稍有一些變動則產物將會有很大的變異性，所以在作控制器設計時必須格外重視即時系統的掌控特性。本研究即在發展出一套利用瞬間線性化類神經網路的模式，以改善傳統PID控制器於非線性系統的控制策略。在模式建立方面，採用Levenberg-Marquardt演算法來學習類神經網路ARX模式，以代表系統輸入輸出之行為。當其應用於控制迴路下時，基於對控制策略簡化及加速運算的實際需求，使用瞬間線性化類神經網路的技術抽取出線性化模式，以提供控制參數的即時運算。在控制器設計部分，運用GMV(General Minimum Variance)最小化目標函數的概念，及存在限制條件下找尋即時最適的PID控制器調諧參數。此控制策略與一般的gain scheduling同樣擁有各操作區段的線性化動態預測模式，但根據非線性類神經網路結構所推算出的控制參數擁有無限區段的結果，所以在系統動態擷取上具有更高的可信度。為了應用在實際的操作過程，本研究也提出了動量過濾、更新控制參數的參考準則及控制作動步伐的修正法，期能在實際運用上有大幅的改善。故此研究有兩大特點：(1)簡化複雜的計算模式，(2)將具有非線性學習特性的類神經網路融入控制器的設計。最後將此所發展的控制架構應用於非線性差分方程式、pH中和反應槽與批次反應器之模擬控制，以驗證此方法的效用。

關鍵字

瞬間線性化 ； 線上自我調整 ； 類神經網路 ； 最適化計算 ； PID控制器

並列摘要

The inherent time-varying nonlinearity and complexity usually exist in chemical processes. The design process would be significantly deviated from the normal operating condition when only a slight disturbance occurs. Accordingly, the design of control structure should be properly adapted based on the instantaneous state. In this research, an improved conventional PID control scheme using linearization through a specified neural network is developed to control nonlinear processes. An input-driven output neural network ARX model trained by Levenberg-Marquardt algorithm is introduced in the model design. The linearization of the neural network model will be proposed to extract the linear model for updating the controller parameters. In the scheme of the optimal tuning PID controller, the concept of general minimum variance is presented and a constrained criterion is also considered. Like gain scheduling, the control system of the proposed method at each time interval is chosen from a set of predefined linearized dynamic model. Unlike gain scheduling control, the control parameters based on the neural network model have an infinite scheduling resolution when using a neural network for updating parameters. To apply the proposed method to most of the practical application problems, several variations of the proposed method, including the momentum filter, the updating criterion and the step size of the control action, are presented to provide significant improvement and make the proposed algorithm more practical. The proposed method has two advantages. First, less computation of linear adaptive control scheme is used. Second, the nonlinear characteristics of neural networks can be incorporated into the control design. To demonstrate the potential applications of the proposed strategies, several problems, including batch reactors and pH neutralization processes, are applied.

並列關鍵字

On-line self-tuning ； Optimization ； Linearization ； Neural Networks ； PID controller

參考文獻

1. Ahmed, M. S. and Tasadduq, I. A., “Neural-net Controller for Nonlinear Plants：Design Approach Linearisation,” IEE Proc-Control Theory Appl., 5, 315 (1994).

2. Ali, E., “Control of Nonlinear Chemical Processes Using Adaptive Proportional-Integral Algorithms,” Ind. Eng. Chem. Res., 39, 1980 (2000).

3. Ali, E., “pH Control Using PI Control Algorithm with Automatic Tuning Method,” Trans. IChemE., 79, 611 (2001).

4. Arthur, J., “A Nonlinear PI(D) Controller,” Canadian J. Chem. Eng., 67, 485 (1989).

5. Astrom, K. J., “Theory and Applications of Adaptive Control – A Survey,” Automatica, 19, 471 (1983).