應用決策樹於核電廠低放射性廢棄物類別預測研究

Using decision tree method to predict low- level radioactive waste classes for nuclear power plant.

萬耀鴻(Yao-Hung Wan)

指導教授 ： 周永燦

中原大學/電機資訊學院/工業與系統工程研究所/碩士(2012年)

https://doi.org/10.6840/cycu201200869

摘要

我國目前有三座核電廠(核一廠、核二廠與核三廠)以及一座正在興建中的龍門電廠。截至2010年7月，各核電廠共產生低放射性廢物桶數約為98,907桶（55加侖），目前暫時存放於各電廠廠內與蘭嶼貯存場。為尋找永久貯存場址，政府於民國95年5月24頒布「低放射性廢棄物最終處置設施場址設置條例」(以下簡稱低放處置條例)。而前置工作尤以低放射性廢棄物分類計算最為繁瑣與複雜。
低放射性廢棄物分類計算須以廢棄物桶內各核種值進行計算，低放廢棄物桶之核種包括加馬核種(關鍵核種)與難測核種; 加馬核種比較容易由一般非破壞技術量測;難測核種則不易量測，大都均需藉由取樣算出比例因數後，再使用公式計算。依據法規中的分類計算公式，低放射性廢棄物桶依輻射強度可分為A、B、C及超C四類，若廢棄物為超C類則無法儲放最終處置場，因此其確認工作則為前置作業中最重要的部份之一。然而在實際的資料中，資料不全或遺失的現場經常發生，若需重新取出、取樣與計算，需耗費大量人力與時間。
有鑑於此，本研究主要目的為運用資料採礦中的決策樹演算法，進行超C類低放射性廢棄物類別的預測。除可作為低放射性廢棄物分類初步判斷，亦可大幅減少所需的人力與時間。為提高精確度，本研究評估了三種不同決策樹模型的計算節點之演算法(Entroy、Bayesian with K2 Prior、Bayesian Dirichlet Equivalent with Unifom Prior)，並透過分類矩陣以及增益圖結果百分比的比較其優劣。本研究結果發現以Bayesian Dirichlet Equivalent with Uniform Prior演算法所建立的模型具較佳的預測能力。除此之外，本研究亦找出影響超C類廢棄物核種順序，亦可作為核電廠核燃料是否有滲漏的參考依據以及了解發生原因，進而精進核電廠人員管理低放廢棄物工作。

關鍵字

資料採礦 ； 超C類廢棄物 ； 低放射性廢棄物 ； 核種活度 ； 決策樹

並列摘要

For the nuclear power plant, one of the main issues is the disposal of the radioactive wastes. To meet the waste acceptance criteria of final disposal repository, waste packages must be characterized for radiological and non-radiological properties. The classifications of the low-radioactive wastes have to use the value of nuclides. For the low-radioactive wastes the nuclides is including the gamma nuclides and difficult to measure nuclides. Many nuclides list in low-radioactive wastes regulation cannot be easily assayed by non-destructive method, it is so-called difficult to measure nuclides. The low- radioactive wastes can be classed to four classes, A, B, C, and above C by using Scaling factor method. When the class is above C, the low radioactive wastes cannot disposal into the permanent disposal site, so it is important to identify the class of low-radioactive wastes. However, missing part of nuclides value or incomplete of the sampling data occurred frequently, and it is cost a lot of money and time for redo the sampling and computing.The aim of this study was to predict the class of the low-radioactive wastes, and found low-radioactive wastes which the class was above C by using the decision tree.The decision tree is one of the technical in data mining. This study compared three different model of the decision tree (Entroy、Bayesian with K2 Prior、Bayesian Dirichlet Equivalent with Unifom Prior). We used classification matrix and the gain figure to evaluated the predict ability of three models, and the BDEU model resulted in greater ability than the other models. For the above C class low-radioactive wastes, we also found the important nuclides and the orders. The result may provide to the staffs of nuclear power plant to evaluate the low-radioactive wastes, and improving the management of the nuclear power plant.

並列關鍵字

low-radioactive wastes ； data mining ； nuclides ； decision tree.

參考文獻

A.Raymond., (1995), “Inventory and characterization of important radionuclides for the safety of storage and disposal.-correlation with keyradionuclides in typical waste streams,” Journal of Radioanalytical and Nuclear Chemistry, Articles,193

Customer Relationship Management, John Wiley &Sons, New York, NY.

Cooper, G. F. and Herskovits, E., (1992), “A Bayesian Method for the Induction of Probabilistic Networks from Data. “Machine Learning, 9, pp. 309-347.

Dzeroski, S., (2006), Towards a general framework for data mining.

Elovici, Y. and Braha, D., (2003), “A decision-theoretic approach to data mining”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 33(1), pp. 42 – 51.