随机森林算法在石油馏分临界性质预测中的应用  

Prediction of Critical Properties of Petroleum Fractions by Random Forest Method

作　　者：宋新阳 刘子媛[2] 王从岗[2] 

机构地区：[1]青岛经济技术开发区第一中学,山东青岛266510 [2]中国石油大学(华东),山东青岛266580

出　　处：《广州化工》2016年第20期71-73,共3页GuangZhou Chemical Industry

摘　　要：在实测原油馏分临界性质及基础物性数据基础上,采用随机森林方法预测其临界性质。预测临界温度的随机森林模型训练的平均相对偏差在0.6%左右,测试的平均相对偏差在1.4%左右。预测结果精度较高,应用范围广。预测临界压力的随机森林模型训练的平均相对偏差一般在3%左右,测试的平均相对偏差一般在6%左右。考察了随机森林模型参数的影响,其中Ntree取500或800时预测结果能够满足要求;对于输入参数数目k≥4时,Mtry取k-1预测精度较高,当k较小时Mtry取k的预测精度较高。Based on the experimental data of critical properties and basic physical properties of petroleum fractions,random forest method was used to predict the critical properties. The average relative deviation for critical temperature prediction was about 0. 6% for the train result,and 1. 4% for the test. The average relative deviation for critical pressure was about 3% for the train result,and 6% for the test. The results showed that random forest method had relatively high veracity and a wide range of application. The parameters of random forest model were examined. The appropriate value of Ntree was 500 or 800. For models which had more than 4 input parameters,the appropriate value of Mtry was the number of input parameters minus 1,but for models which had smaller parameters,the appropriate value of Mtry was the number of input parameters.

关 键 词：随机森林 石油馏分 临界性质 预测 

分 类 号：TE622[石油与天然气工程—油气加工工程]