利用網路平台共同設計模糊控制器時控制規則之選擇

Abstract

Web 2.0網路平台為一開放式的分享空間，可藉由平台分享知識、想法、共同開發軟體程式與共同創作音樂等，本論文探討於網路平台共同設計模糊控制器時一些可能面臨的問題與挑戰。本論文選擇討論模糊控制器，乃因須於網路上共同設計的系統應較具難度，而模糊控制器特別適用於非線性、時變或數學模型不完全的系統，有其優勢。 以全球線上百科全書Wikipedia為例，本論文推演與設想一個共同設計控制器網路平台會需要哪些元件與架構，分別由平台之系統架構、設計原則與頁面設計三方面進行討論，並比較Wikipedia平台與控制器設計平台兩者之異同，以了解擬設計之平台之可實現性。具體而言，本論文假設輸入平台的控制建議為模糊控制規則，以控制問題上傳的角度審視兩個主要課題： (1) 如何有效率的判別個別控制規則建議替換入原規則庫後，其響應相較原模糊控制器將變好或變壞。由於模糊控制器經常被運用於無數學模型的受控廠，欲於執行模擬前由規則間的變化判別其響應優劣，是相當具有挑戰性的問題。 (2) 得知個別響應後，如何組合較原控制器響應更好的控制器，或滿足規格的控制器。不同前件部之規則如何互相搭配，使其控制性能優於原本未替換前之模糊控制器性能，而不相互影響降低控制器之品質為一重要的 課題。 本論文測試四個受控廠，而選定的響應指標分別為最大超越量與上升時間，並經由下列五步驟制定六個準則以解決上述問題：一、觀察響應值隨著歸屬函數的變化。二、藉由步驟一中的變化擬定解決方針。三、尋求其他工具輔助判斷。四、選定受控廠、控制規格、模糊控制器及假想的控制規則建議，制定一套判斷準則。五、經試誤法測試不同的受控廠與控制規格以調整判斷準則。針對平台上雜亂無序的控制規則建議，依準則1分類排序，可清楚地了解設計提供者針對那些控制規則提供控制規則建議。而針對同前件部的控制規則，準則2建議優先測試的控制規則建議。準則3為針對響應値隨著歸屬函數變化為不變型之判定依據。準則4與準則5可根據原模糊控制器與替換特定規則後之模糊控制器間語意梯度的變化，判別替換後知響應優劣。準則6針對問題(2)，提供一可行的組合機制以使模糊控制器之控制性能優於原本平台上的模糊控制器。 最終，本論文以倒單擺驗證設計準則之可行性。假設於平台上設計提供者提供的規則共647條，根據準則1至5，其中需要替換入原規則庫執行模擬者共196條，其餘451條欲判別其響應優劣，正確者共389條，正確率約莫86.3% 。 關鍵字：網路平台、共同設計、模糊控制器、模糊控制規則

Web 2.0 internet platform is an open space for sharing. Nowadays, sharing knowledge and ideas via the internet platform is extremely popular. For example, collaborative software programming and music composition through the internet platform are common. Inspired by this sharing possibility, this thesis proposes the concept of collaborative fuzzy controller design via the internet platform and studies related problem. Since the control problems uploaded to the proposed platform are more likely difficult ones, and fuzzy controllers are particularly suitable for nonlinear, time-varying and uncertain systems, collaborative design of fuzzy controllers has its niche. The internet platform for collaboratively designing fuzzy controllers to be discussed in this thesis has similar components and structure to the online encyclopedia “Wikipedia”. The construction of such a platform is considered from three different perspectives: system structure, design principle, and web page. To understand the feasibility of implementing such a platform, similarities to and differences from the “Wikipedia” are investigated first. It is assumed that in response to the control problems uploaded, fuzzy control rules are sought for a given fuzzy controller framework, and the platform needs to accomplish the following tasks. i. To judge if the performance is better or worse when an existing rule is substituted by a new control rule. Fuzzy controllers are frequently adopted for plants without explicit mathematical model, so it is challenging to judge if control rules would enhance the performance before new control rules are replaced into the original rule base. ii. To select control rules to compose a new fuzzy rule base with better performance than that of the original fuzzy controller. To do the above to four selected plants, this thesis proposes six criteria for control problem providers to discriminate and utilize the control advices received on the platform. All the six criteria are developed by testing four plants through the following five steps: (i) observing the change of performance index value with respect to the membership function changes; (ii) developing the solution policy based on results from step (i); (iii) searching other tools to assist judging the performance of control rules; (iv) selecting plants, control specifications, fuzzy controllers, and control rule advices to create a set of judging criteria; and (v) testing different plants and specifications to adjust the criteria by the trial and error method. The performance indices of the above five steps are maximum overshoot and rising time. According to criterion 1, one can classify all the control rule advices into several groups. Users who need to engage in control design can understand what control advices are available on the internet platform. Criterion 2 suggests which control rule advice with the same antecedent should be substituted into original rule base to run simulation first. This thesis classifies the change of performance index value with respect to the membership function changes into three types. Criterion 3 is developed for invariant type. According to criterion 4 and criterion 5, one can judge if the performance will become better or not by observing the linguistic gradient changes of original fuzzy controller and fuzzy controller whose rules are replaced with control rule advices on the platform. Criterion 6 provides a feasible combination mechanism of control rule for designing a fuzzy controller whose performance is better than the original one. Finally, an inverted pendulum is adopted as the plant to verify the feasibility of the proposed criteria. Assume that there are 647 control rule advices proposed on the platform, and 196 control rules need to be substituted into the original rule base to run simulations. According to the criteria proposed in this thesis, performances of the remaining 451 control rules are judged without replacing the rules into original rule base. The accuracy of judging result is about 86.3%. Keywords:sinternet platform, collaborative design, fuzzy controller, fuzzy control rule.