複數模糊推理與粒子群體智慧計算之函數近似研究;Study on Complex Fuzzy Inference and PSO for the Problem of Function Approximation

NCUIR > School of Management at National Central University > Department of Information Management > Research Project > Item 987654321/57729

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/57729

Title:	複數模糊推理與粒子群體智慧計算之函數近似研究;Study on Complex Fuzzy Inference and PSO for the Problem of Function Approximation
Authors:	李俊賢
Contributors:	中央大學資訊管理學系
Keywords:	資訊科學--軟體;複數模糊集;複數模糊推理;粒子群計算;群體智慧;機器學習;函數近似;系統鑑別;預測 complex fuzzy set;complex fuzzy inference;particle swarm optimization;swarm intelligence;function approximation;system identification;prediction
Date:	2009-09-01
Issue Date:	2012-10-01 15:42:09 (UTC+8)
Publisher:	行政院國家科學委員會
Abstract:	此研究計畫使用複數模糊推理(complex fuzzy inference)及粒子群體智慧計算 (particle swarm optimization, PSO)進行函數近似的研究。函數近似研究是許多科學研究與應用的基礎。在有限的資源:包含有限的資訊數据取樣、有限的知識、有限的計算能力、有限的時間條件下建立一個可用的函數模型以利後續的決策判斷、動態行為分析及預測、原(雛)型系統設計和系統調控。眾多應用多直接的或間接的與函數近似的研究相關。二十一世紀初的人工智慧在各領域的科學研究與應用更普遍而深入。模糊集的觀念更進一步擴展為複數模糊集(complex fuzzy set)，這是一個新的計算模式(computing paradigm)的開始，有別於傳統模糊集所完成的模糊推理系統，對智慧型系統的研究與應用有重大影響，但目前尚未有合適的完整系統設計出來。複數模糊集表示不明確的資訊或觀念就更進一步從經驗導向(experience oriented)進入經驗抽象化導向(abstraction oriented)的表達。複數模糊集定義在單位複數圓盤(unit disk in the complex plane)之複數區間複數模糊集,其歸屬程度可用複數歸屬函數(h) r (h)exp(j (h)) s s s    表示。自從基因演算被提出，開啟了群體智慧式機器學習的曙光，智慧型系統機器學習的研究進入自然群體演化的時代。群體智慧(swarm intelligence)的相關機器學習方法的啟發靈感來自於造物主所創造的自然界許多生物族群之覓食、求生或繁衍的集體行為，例如鳥群、魚群等。在PSO 的方法中，每個個體稱為粒子(particle)，一整群粒子稱為群族(swarm)。在最佳化搜尋策略上，一整群粒子同時平行式的進行，而且彼此分享且交換搜尋資訊，比較不容易掉入局部最小(local minimum)的區域。在智慧型系統的學習進行過程中，每一個粒子的位置代表一個潛在可能解(potential solution)。本研究計劃提出PSO-複數模糊系統之函數近似研究，系統鑑別及預測之應用，在人工智慧的科學研究領域發展創新。 ; A particle-swarm-optimization-based complex fuzzy inference approach is proposed in the research project, in which complex fuzzy inference system using complex fuzzy sets and the machine learning method using particle-swarm-optimization (PSO) for the adaptive inference system are used to study the problem of function approximation. Function approximation is a critically and fundamentally scientific research problem. The need of function approximation is found in many research areas and applications, such as system modeling and identification, time series prediction, control, signal processing, image processing, network security, and many others. The target of function approximation is to make use of a selected function (or a system in some form, such as the proposed PSO-based complex fuzzy inference system), under condition of limited computing resources and time, to build a sufficiently useful model-based relationship between sampled input-output data pairs, which may be obtained from the desired function or system (which may be unknown to us, and yet whose behavioral information can be observed), so that it can be used later for the purpose of decision-making, system analysis, prediction, prototype design, and performance control. The concept of complex fuzzy sets is originated from the extension of fuzzy sets. The membership of complex fuzzy sets are continuously complex-valued, within the unit disk of the complex plane, in the form of (h) r (h)exp( j (h)) s s s    , where rs(h) is the amplitude of the complex function and  s(h) is the phase. This is a new extended concept, which may leads to new computing paradigm for artificial intelligence research. The idea of PSO machine learning was inspired by the foraging behavior of bird flock or fish school. Basically, the whole group of individuals is called a “swarm” and an individual in the swarm is called a “particle”. Each particle is viewed as a potential solution of the domain problem. The PSO-based approach of swarm intelligence is adopted in the research project. The target of the research project is to study the adaptive ability of complex fuzzy inference system for function approximation problem, using multiple-swarm-based PSO approach, and then apply the studied result to the application of system identification and prediction. ; 研究期間 9808 ~ 9907
Relation:	財團法人國家實驗研究院科技政策研究與資訊中心
Appears in Collections:	[Department of Information Management] Research Project

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	331	View/Open

社群 sharing

Loading...