模糊鑑別器於Burn In系統之設計

建立一個模糊系統主要是依靠專家經驗來提供設計方法。如果一個受控系統只知道它的輸入輸出資料時，如何在沒有專家經驗下來建立適當的模糊鑑別器是很大的挑戰。本論文會探討設計一個模糊鑑別器的方法，藉著學習法則來不斷地學習輸入輸出的資料的行為，使其儘可能趨近於欲鑑別的系統。在論文的第一部份，我們發展了一種適用於模糊空間切割的分群演算法，它可以有效的探勘所處理資料之群聚分佈狀態，分析欲鑑別系統輸入輸出資料的群聚關係，而所得到之結果則可用作模糊系統粗略式的結構鑑別。得到初步的模糊系統之後，便可以系統之輸入輸出資料做為訓練目標，進一步學習以細調模糊系統的參數，使之能夠更精確符合受鑑系統的行為。在論文的第二部份，我們介紹一種基於競爭式學習的模糊系統建模方法。我們利用SOFM 神經網路在低維度矩陣空間內輸出拓樸網路來得到高維度輸入資料的推理法則，並產生有意義的規則資料庫以重塑受鑑別系統。而為了更精確逼近受鑑系統的特性，本論文將萃取出的模糊規則進一步結合遞迴式最小平方法進行參數鑑別的設計程序，來達到微調的效果。論文的最後一個部份則是將模糊鑑別器的設計方法應用在一Burn-In測試系統的恆溫調節。在此Burn-In測試系統中，我們需要控制加熱器與散熱風扇讓測試溫度穩定在所設定範圍，才能達到對每一個待測物Burn-In的效果；傳統作法中使用PI控制器所實現的溫度控制系統需要耗費很多時間來調整控制參數，在上升時間與超越量等性能也不易滿足測試的需求。本論文使用模糊鑑別器的設計方法建立模糊控制器以操控風扇及加熱器之運作，實際結果發現除了減少參數調整的試誤時間外，對系統也有較快上升時間及較小的超越量。

關鍵字

模糊系統；群聚分析演算法:奔應系統

並列摘要

To establish a fuzzy system is to rely on the experience of an expert to provide the design methods. For a controlled system, if we only know its input-output data, it would be a challenge to establish an appropriate fuzzy system without expert experience. The first task of the dissertation is to develop the fuzzy space clustering algorithm. It can effectively explore the cluster distribution of the processed data and analyze the clustering of the input-output data of the identified system. We can use the input-output data of the identified system as a training target to tune the parameters of the fuzzy systems more precisely in line with the behavior of the identified system. The second task of the dissertation is to introduce a competitive learning fuzzy system modeling approach.We use the topological network sent out from SOFM to get the meaningful fuzzy rules. We can use the input-output data as the training samples to further fine-tune the parameter of fuzzy inference system so that it can more accurately match the behavior of the system to be identified. The last task of the dissertation is to apply the fuzzy indenitification design to approach a thermostatically controlled Burn-In test system. In this Burn-In test system, we need to control the heater and fan to keep the temperature stabilized in the set range. We use the propsed fuzzy identification method to build a fuzzy controller to manipulate the operaion of the fan and heater. The actual results show the method can reduce the time of trial-and-error for the parameter adjustment. The system also has faster rising time and smaller overshoot.

並列關鍵字

Fuzzy Systems ； Clustering Algorithms ； SOFM ； Burn-In System

參考文獻

[10] R. R. Yager and D. P. Filev, Essentials of Fuzzy Modeling and Control, John Wiley, New Jersey, USA, 1994.

[11] R. R. Yager and D. P. Filev, “Approximate clustering via the mountain method,” IEEE Transactions on Systems, Man & Cybernetics, vol. 24, no. 8, pp. 1279-1284, Augest 1994.

[47] C. C. Chen, Design of Fuzzy Systems Based on Partitioning Input Spaces, Ph. D. Dissertation, University of Tamkang, 2000.

[1] Y. J. Chen and C. C. Teng, “Rule combination in a fuzzy neural network,” Fuzzy Sets and Systems, vol. 82, no. 2, pp. 161-166, September 1996.

[2] E. T. Kim, M. K. Park, S. W. Ji, and M. N. Park, “A new approach to fuzzy modeling,” IEEE Transactions on Fuzzy Systems, vol. 5, no. 3, pp. 328-337, Augest 1997.

國際替代計量

模糊鑑別器於Burn In系統之設計

全文下載

主題瀏覽