The flammable Limits Model for Organic Compounds Using Quantitative Structure Property Relationship Approach

• Main Authors: Chao-pei Lai, 賴昭蓓
• Other Authors: Chan-Cheng Chen
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/62797711020512157966

碩士 === 國立高雄第一科技大學 === 環境與安全衛生工程研究所 === 103 === Flammable Limits(FL) is an important index for assessing the fire and explosion hazards of vapor of flammable liquid or flammable gas.The European Union’s legislation REACH requires assessment flammability characteristics of all chemicals that are produced in or imported into EU.Because complete evaluation of these properties by experimental tests is dangerous and time-consuming,the prediction of FL is an important study.REACH encourages the use of cost-effective methods like QSAR approach ,so we use QSAR approach to built models to predict the Up Flammable Limit(UFL) and Low flammable Limit(LFL) of organic compounds. The LFL data is 458 organic compounds, which are experimental values from the DIPPR 2012 database.These 458 compounds are randomly divided into a training set of 366 compounds and a test set of 92 compounds.Then, using structure of compounds to calculate the 3196 molecular descriptors.The stepwise regression method is applied to choose molecular descriptors that are highly correlated with LFL.The model regression parameters are build with training set.Finally,Using test set verify predictive performance of model.This model has four molecular descriptors,the R2 value to be of 0.8830, the average absolute error(aae) to be of 0.405 vol% and predict the test data with Q2 =0.8576, the aae to be of 0.375 vol%. The UFL data is 389 organic compounds, which are experimental values from the DIPPR 2012 database.First, we use 389 organic compounds to built MLR model with six molecular descriptors. This model has six molecular descriptors,the R2 value to be of 0.7073, the average absolute error(aae) to be of 3.574 vol% and predict the test data with Q2 =0.6458, the aae to be of 4.620 vol%.However, we found 389 compounds which greater than 40 vol% only 13 compounds. And these compounds impact model too large. Therefore, we use 376 organic compounds to built MLR model with five molecular descriptors and nine molecular descriptors. Which nine molecular descriptors MLR model R2 value to be of 0.7027, the aae to be of 2.620 vol% and predict the test data with Q2 = 0.5423, the aae to be of 3.033 vol%.Moreover, we want to compare with the previous literatures. We use 345 organic compounds to built MLR model with four molecular descriptors,five molecular descriptors and seven molecular descriptors. Which seven molecular descriptors MLR model R2 value to be of 0.7474, the aae to be of 1.640 vol% and predict the test data with Q2 =0.6820, the aae to be of 2.014 vol%.