- 學位論文
聚類理論應用於溫溼度感測分析
Analysis of Temperature of Humidity by Application of Cluster Theory
摘要
本論文提出三種聚類法則建立模糊歸屬函數的機制，來取代一般模糊歸屬函數的建立。適應共振理論網路Ⅱ(Adaptive Resonance Theory Network 2, ART2)，它的優點就是能快速及客觀集群；此外並結合模糊C-均質演算法（Fuzzy C-Means, FCM）和自我組織特徵映射網路（Self-Organizing Feature Map, SOFM），重新計算群組中心和範圍，發展出一套全新集群分析模式，進而比較三種聚類法則在相互間的優缺。藉由聚類質心建構模糊三角歸屬函數，並利用輸入區間溫溼度資料產生最終的關聯性法則。 在本研究中我們將模擬多點感測器的佈建，多感測器融合了溫度和濕度的測量數據，將資料建立成聚類資料庫，本論文提出三種聚類演算法去分析資料；而聚類演算法是一種非監督式的資料探勘方法，可以依照資料的分布情況，將性質相似的資料分成若干個群聚，再利用模糊理論與類神經理論執行聚類分析，結合模糊綜合評判解決模糊理論建立規則的缺陷。 我們以人體環境舒適度26度設為目標溫度，並模擬出FCM在區間溫度值有較佳的風速輸出值，ART2在運行演算法過程中速度最快，而SOM風速輸出在低速範圍相對於C-MEAN與ART2不精確，我們以擷取的溫溼度值運行聚類演算法，求得風速輸出變化值與模擬分析。
並列摘要
This thesis presents three kinds of clustering theories to establish fuzzy membership function which replace the traditional fuzzy membership method. The advantage of Adaptive Resonance theory Network 2 (ART2) is adopted to cluster quickly and objectively. The combination of fuzzy C-means and Self-organizing feature map(SOFM) can re-calculate group center and scope, and the develop a new cluster analysis model. The research compares the advantages and disadvantages of such three kind of clustering laws which are utilized to construct the triangle fuzzy membership function. The constructed triangle fuzzy membership function produces the final association rules by input interval data of temperature and humidity. The simulations of multi-sensors have been carried out in the thesis. The multi-sensors provide temperature and humidity measurements to establish a cluster database. The cluster data has been analyzed by three kind of cluster theories which are proposed in this thesis. A cluster method is a unsupervised search method which can group data with similar characteristics into the same cluster. Moreover, we take advantage of the fuzzy theory and artificial neural network to perform clustering analysis. The analysis combines the fuzzy synthesis assessment to improve the defects of setting rules of the fuzzy theory. The object temperature of 26oC is nominated as the most comfort temperature for the human being. Under this object condition, the simulation with FCM can solve out the best wind speed in a studied domain area. ART2 has the best algorithm approaching time than other two methods, i.e. C-MEAN and ART2. SOM has less accuracy than C-MEAN and ATR2 in low wind speed conditions. With the temperatures and humidity data in a specified area, we can apply clustering algorithm to obtain the variations of wind speed outputs and in advance to do more difference simulation analyses.