群集分析於空間破壞機制之資料探勘

本文介紹了群集分析，應用在固體岩材受力作用，於不同的應力路徑下，探討其破壞機制的資料探勘 (Data Mining)。研究中，以群集分析方法中的K-均值法 (K-Means)，與自組織映射圖網路 (SOM) 演算法建立分析模式。同時考慮兩種不同尺度之固體岩材破壞資料：大尺度為斷層錯動之三角點位水平變化，範圍為花東縱谷部分，長約159公里，寬約80公里，為空間資料型態；小尺度為脆性岩材之微震裂源演化，範圍為試驗材料本體部分，長約150公厘，寬約150公厘，為空間與時間的資料型態。前者進行海岸山脈三角點位「位移方向與位移量」之分類，與「斷層位置」空間範圍之推估。在已知群集數目下，將海岸山脈的三角點位與位移量，以本研究方法分類成北、中與南三段，其位移方向由南段西北方向轉至北段東北方向，位移量介於1.5(m)~5.5(m)之間。經比對專家經驗判釋，海岸山脈「位移方向、位移量與斷層位置」的成果，兩個方法在位移方向與位移量皆與其接近。而在斷層位置之推估上，K-Means 演算法所推估之斷層位置優於SOM 演算法，確可輔助人為判斷，且K-Means 演算法在執行時間上優於SOM 演算法。後者進行延-脆性破壞演化分類，與「彈-塑性區界面」空間範圍之推估。在未知群集數目下，將微震裂源分類成延性與脆性兩種破壞模式(演化)，與彈性、塑性與脆性三個力學演化分區(空間分佈)。結合理論解析解與專家經驗組合的成果，作為比對值，本研究兩個方法中，以SOM 演算法之分類結果較與其相近，且在判斷「彈-塑性區界面」範圍的時間上，已從最久半天的分析時間，減少至需30分鐘到1小時的時間。從而達到客觀與減少時間之目的。

關鍵字

資料探勘；空間資料探勘；群集分析； K-均值法；自組織映射圖網路；破壞機制

並列摘要

This study applies cluster analysis to solid-rock materials by different stress pathways, and explores data mining of fracture mechanism. Two cluster analysis methods, K-Means and Self-Organizing Map (SOM) were utilized to establish an analytical model. Two different scales of solid-rock material destruction data were utilized: large-scale data is the triangulation changes of fault dislocation, using the East Rift Valley section of Taiwan (length 159 km, width 80 km) as the experiment area, which represented a spatial data type; small-scale data are the evolution of microseismic sources, using Quasi-brittle rock as the test material (length 150mm, width 150mm), which represented a spatial and time data type. In the former experiment, the ‘direction’ and ‘amount’ of displacement of the East Rift Valley section of Taiwan was classified into three part: north, center, south, and figure out approximately location of the fault in the spatial coverage was estimated. The ‘direction’ of displacement was from northwest of south part to northeast of north part, and the ‘amount’ of displacement was from 1.5(m) to 5.5(m). After we compared the estimation results from professional experience, the estimation results using K-means and SOM were close to them, the fault location estimation results using K-Means were better than SOM, we conclude K-Means can assist artificial estimation and it was faster than SOM. The latter experiment test material is Quasi-brittle rock, ductility (ductile) and brittleness (brittle) destruction models by ‘occurring location’ and ‘occurring time’ of the microseismic source was classified into three part of mechanics space: elasticity, plasticity, brittleness. After we compared the estimation results from professional experience, the estimation results of SOM were more approximate to them, and the time spending on determine the range of ‘elasticity- plasticity interface zone’ was reduced 30min to 1hour, more efficient than estimation results.

並列關鍵字

Data Mining ； Spatial Data Mining ； Cluster Analysis ； K-Means ； SOM ； Fracture Mechanism

參考文獻

[34] 潘麒全，可修正的二分群集法，碩士論文，中原大學資訊管理學系，2003

[32] 林鳳書，結合群聚計數之搜尋演算法於容錯系統之研究，碩士論文，中原大學機械工程學系，2003

[33] 金昆翰，開發同步化聲光非破壞檢測於類岩材料受雙刀式貫切破壞之研析，碩士論文，國立台北科技大學土木與防災工程學系，2008

[1] Bi, S., Xue, S., Xu, Y., Pei, A., “Spatial data mining in settlement archaeological databases based on vector features”, Fuzzy Systems and Knowledge Discovery, Volume 4, Article number 4666398, pp 277-281, 2008

[3] Birant, D. and Kut, A., “ST-DBSCAN: An algorithm for clustering spatial-temporal data”, Data and Knowledge Engineering, Volume 60, Issue 1, pp 208-221 , 2007

國際替代計量

群集分析於空間破壞機制之資料探勘

全文下載

主題瀏覽