Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/47261
標題: 水庫及水力電廠的安全監測技術整合研究暨示範例之建構-子計畫:監測資料之補遺演算法(I)
Algorithm to the Solutions for Missing Data in Monitoring for Structure Safety (I)
作者: 郭其珍
關鍵字: 土木水利工程類
應用研究
摘要: 水庫壩體之安全與維護,一般係將監測數據依時間繪成曲線,曲線中顯示觀測值之長期、短期或異常之突變,再依曲線之變化趨勢,與過去觀測資料、理論分析及自然現象之預期趨勢相比較,依其穩定性、相關性、合理性、一致性、突變性及對稱性加以研判評估,並與警戒值或危險值相比較後,以發現潛在問題,做預防災變之處置。於此過程,首要依賴監測資料之收集與評析。監測系統涉及儀器之觀測、資料之記錄、儲存、傳送、處理、繪圖、分析、比較等項,任何偏離均影響整個監測之精確性。然而,於實務操作上,不利環境條件如漏水、濕氣、落石、風等及監測之通路狀況,儀器條件如功能老化,故障或其它因素,人為條件如疏失或訓練不足,其影響,皆可能中止監測資料的收集,或導致資料失真。因而,如何將這些遺失或不正確的資料庫適當處理以資運用,成為重要的課題。傳統上,缺失資料的處理的方式,諸如;ListwiseDeletion,Mean Imputation,Closest Fit,Maximum Likelihood,Expectation Maximization,Multiple Imputation 與Neural Network approach,於實際的運作上,各有其局限。因此,為尋求適當的補遺方式,所提之研究案,擬以人工智慧的角度,融合particle swarmoptimization 方法與Naïve Bayes classifier觀念,做進一步之研討,進而推展出合宜的演算法。
For the integrity of dam structure, monitoring the structural response to the environmentalchange has to be cared. The plot of monitored data versus time read the implications for severalpossibilities. The unexpected result or odd information is an alert to the necessity of furtherexamination to the risk of dam structure. The base for the risk analysis depends on the collecteddata. Practically, the monitoring system is not always dependable because of the inevitable crashof mechanical running, also the human error in management. All of these lead to the data missing.For the requirement of risk evaluation, the incomplete data set has to be processed in advance ofdeduction. For fulfilling this necessity, methods of listwise deletion, mean imputation, closestfit, maximum likelihood, expectation maximization, multiple imputation and neural networkapproach, are commonly applied. Notwithstanding the validity of these approaches in use isin arguments because of different cognitions. Thus a project is proposed herein to seeking avalid approach with less argument for dealing with missing data. The proposal is designed tolink particle swarm optimization technique and Nave Bayes classifier concept for buildingup the available model searching for missing data.
URI: http://hdl.handle.net/11455/47261
其他識別: NSC99-2625-M005-011
Appears in Collections:土木工程學系所

文件中的檔案:

取得全文請前往華藝線上圖書館



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.