Predicting the Octane Number of Reformer Products and Optimizing Cata1ytic Reforming Unit

[[abstract]]觸媒重組工場所生產的重組油是汽油摻配的主要原料。由於重組油的價格遠較甲基四丁基醚（MTBE）為低，提高重組油的辛烷值可取代部分MTBE摻配量，同時可大幅降低成本。況且，美國加州已於今（2003）年元月禁止MTBE的使用，許多國家也在考慮禁用；因此，減少MTBE的用量不僅降的成本，而且還順應世界環保主流趨勢。如欲達成此目標，石油公司必須在不進行硬體改善的條件下，設法提高油品的辛烷值。 本論文探討如何改善觸媒重組工廠的操作情況，以提昇產品的辛烷值。首先收集、分析過去數年的操作數據，從數據中找出主要的操作變數，如反應器的溫度與壓力、原料的品質、反應物停留時間等，然後再應用多變量分析與類神經網路等方法，以建立辛烷值與操作變數的關係。 研究結果顯示，應用EXCEL軟體中的LINEST功能函數可建立一個低標準誤差（0.4）、高相關係數（0.926）的經驗方程式。此經驗方程式可供操作人員調整主要變數，使生產最適化。以一個日產8000公秉的95汽油摻配工場而論，重組泊的辛烷值每提高一點時，每天可節省45.6公秉的MTBE與48萬元成本。應用類神經網路可建立一個非參數型的關係，可以作為線上虛擬分析儀，以取代一般煉油廠所使用的價格昂貴且易失常的機械式分析儀。 本論文顯示應用類神經與多變量迴歸分析等先進數學與統計方法，可以準確的描述產品品質與操作參數的關係，如果能將此類方法應用於工場的操作中，將可大幅提昇生產效率與降低成本。[[abstract]]The petroleum products from the catalytic reforming units are major ingredients in the gasoline blending. Since the reforming products are cheaper than MTBE, a major chemical to boost the octane number of gasoline, an increase in the octane number of the reforming products will definitely replace a fraction of the MTBE blended and save a significant amount of money. Furthermore, California, USA prohibited the use of MTBE in 2003 and many nations may follow in the near future. Therefore, reducing the use of MTBE will not only reduce the cost of gasoline, but also follow the main stream in the oil business. To achieve this goal, one has to produce higher octane number products without major retrofitting. This thesis investigates how to improve the operation of the catalytic reforming units to obtain higher octane number of the products. Historical operating data were first collected and analyzed. Major input operating variables such as reactor temperature and pressure, feedstock quality, retention time, etc were identified. Multi-variant Regression Analysis（MRA）and Artificial Neural Network（ANN）were then used to correlate the output variable of octane number and the input variables of the operating parameters. The purpose is to establish both parametric and nonparametric relationships among octane numbers of the products and the operating variables. The results showed that a regression equation could be obtained by using the LINEST function of the EXCEL software with a standard deviation of 0.40 and a correlation coefficient of 0.926. This indicated that the empirical equation was reliable and could be used to improve the octane number of the reformer products. For a gasoline blending unit with 8000 kiloliters/day production, 45.6kiloliters of MTBE and 481,308NT dollars would be saved daily, if the octane number of the reformer products mixed in the blending unit was increased by one. A nonparametric relation between the octane number and the operating variables was also established using ANN with two hidden levels and an average deviation of 0.26.The nonparametric relationship between octane number and the operating variables could be used as an online virtual analyzer for the replacement of the expensive and less reliable mechanical analyzer currently employed. This thesis showed that modem mathematical and statistical methods such as Artificial Neural Network and Multi-variant Regression Analysis could be used to accurately describe the relationship among the product properties and the operating variables such as reactor temperature, pressure, and feedstock qualities. Significant amounts of money would be saved if these techniques were applied to the production units