應用專家系統於地下水模式自動化參數檢定之研究

Abstract

傳統上地下水模式之參數檢定可分為人工參數檢定及應用優選法之自動化參數檢定。人工參數檢定在推估參數時，過程不但複雜、冗長、且需要花費許多時間，進行參數檢定的人員除了需具備相關地下水知識外，亦須具有參數檢定之經驗與技巧。自動化參數檢定則通常是將地下水模式結合優選法進行參數檢定，雖然可免去人工參數檢定的繁瑣過程，但優選法之參數搜尋過程對使用者而言是難以參與的黑盒，往往只能被動的檢視最後的檢定結果，另外因為地下水模式甚至優選模式本身，往往皆對問題有相當程度的簡化，自動參數檢定所得之結果亦可能產生不符現地條件的情況。 有鑑於此，本研究整合規則式專家系統與地下水模式MODFLOW發展智慧型地下水參數檢定系統，並透過四個檢定案例驗證系統的正確性，其中分別為水力傳導係數(Hydraulic Conductivity, K)檢定案例、比出水量(Specific Yield, Sy)檢定案例及兩個同時檢定水力傳導係數與比出水量之案例。本研究發展之參數檢定系統可在保有自動化參數檢定的快速及方便的同時，亦可有人工參數檢定之彈性及可解釋性，並能進一步累積整合參數檢定之經驗，及具有只需增加相關規則即能擴充系統功能的彈性。

Conventionally, parameters identification of groundwater model can be classified into manual parameters identification and automatic parameters identification using optimization method. Parameter searching in manual parameters identification requires heavily interaction with the modeler. Therefore, the identified parameters value is interpretable by the modeler. However, manual method is a complicated and time-consuming work and requires groundwater modeling practice and parameters identification experiences to performing the task. Optimization-based identification is more efficient and convenient comparing to the manual one. Nevertheless, the parameters search in the optimization approach can not directly interactive with modeler and one can only examine the final results. Moreover, because of the simplification of the optimization model, the parameters value obtained by optimization-based identification may not be feasible in reality. In light of previous discussion, this study integrates a rule-based expert system and a groundwater simulation model, MODFLOW 96, to develop an automatic groundwater parameters identification system. The hydraulic conductivity and specific yield are the parameters to be calibrated in the system. Since the parameter value is automatic searched according the rules that are specified by modeler, it is efficient and the identified parameters value is more interpretable than that by optimized based approach. Beside, since the rules are easy to modify and adding, the system is flexible and can accumulate the expertise experiences. Several hypothesized cases were used to examine the system validity and capability. The result shows a good agreement between the identified and given parameter values and also demonstrates a great potential for extending the system to a fully function and practical field application system.