地下儲槽滲漏監測系統可靠度評估與最佳化設計原則之探討

Abstract

地下儲槽及管線，因材質發生銹蝕或老化的材料受應力扭曲，極易造成破損洩漏。國內的加油站數量龐大，經證實或疑似污染事件層出不窮。雖然歷年來政府頒佈加油站地下儲槽與管線管理的相關法令，但是由環保署調查資料顯示，許多加油站的滲漏監測仍無法符合法令的標準。此外，由於政府管理措施未臻完善，國內許多業者經由監測發現有疑似污染時，大多自行採許取整治措施，政府部門無從介入瞭解實際污染狀況。目前台灣與世界各國加油站監測的方法雖以採行多年，但是除了監測儀器本身的精確度經過驗證之外，諸如土壤氣體監測、地下水監測等監測成效和現場水文地質條件以及設施配置緊密相關的方法，其可靠度均未經過詳細之研究。因此在各國的法令規章中，對於土壤氣體和地下水監測的結果判定均無明確的標準可循。因此，雖然土壤氣體和地下水監測為成本最低且最符合環境監測定義、且全台灣加油站最普遍採用的監測方法，但其監測之有效性卻是監測方法中可靠度最被質疑的。本研究將針對包括土壤氣體與地下水監測的各種不同的滲漏偵測方法利用資料分析和數值模式模擬加以評估。根據大量模擬加油站儲槽與管線配置時，現地水文地質條件參數的合理變化範圍下取得加油站範圍內三維土壤氣體與油品分佈。利用所得結果，可以計算出在加油站範圍內各網格上設置不同深度和井篩長度的土壤氣體監測井和地下水監測井可以得知之油氣濃度和浮油厚度，建立虛擬加油站滲漏土壤氣體濃度與地下水監測浮油厚度資料庫。之後，利用資料探勘與機器學習方式，進行滲漏監測可靠度因子分析以及預測模式，並可據以建立決策支援模式。並利用類神經網路或機器學習方法萃取模擬結果，所得之線性或非線性方程式可供提出加油站監測之最佳化原則。

Although leak detection methods for underground storage tanks and pipelines have been in use for many years, their reliability was never rigorously assessed. There are well established precision standards for the sensors used for monitoring, but appropriateness of the vapor or ground water monitoring wells in which they are installed or applied has not been subject to scrutiny. Furthermore, the criteria for determination of a possible leak are purely empirical. There are a large percentage of gasoline stations in Taiwan that installed soil vapor extraction systems simply to control the concentration of volatile organic chemicals in soil vapor without knowing for sure whether the underground storage tanks and pipelines are leaking or not. The first objective of the study is to assess the reliability of leak detection methods, especially, soil vapor and ground water monitoring under various hydrogeological conditions. The evaluation of each method will be based on the relationship between the time lag from leakage occurs till detection is confirmed. Reliable criteria for determining the occurrence of leak can be established based on the results of the simulations. A large number of simulations with various site layout and hydrogeological conditions will make up the ‘virtual’ databank for data mining. The second objective of the study is to perform a system optimization analysis. Since the soil and ground water monitoring system layout is complex and non-linear, its optimization must rely on advanced methodologies. This study proposes using data mining with machine learning tools such as artificial neural network or support vector machine to analyze the results of a large number of numerical simulations and develop linear or nonlinear equations. The results could be used to perform optimization and decision support of leak detection monitoring system design based on reliability and cost.