摘要
汽油辛烷值不仅影响汽油的经济效益,也关乎汽车尾气排放量,因此精确预测成品汽油的辛烷值对国民经济发展和环境保护都有着重要意义。为从高维的工业汽油数据集中准确预测出辛烷值含量,首先提出了一种互信息法回归(Mutual information regression,MIR)-递归嵌入式特征选择(Embedded feature selection,EFS)融合降维算法对数据特征进行评分,筛选出30个特征作为建模的主要变量;然后基于优化后的自适应集成学习随机森林算法建立了辛烷值预测模型;最后基于多种互补判别准则,与现有方法进行了充分仿真计算对比。结果显示,改进的MIR-EFS融合降维算法将数据维度减少了89.65%,训练时间减少了81.43%,预测数据的均方误差(MSE)、均方根误差(RMSE)、最小误差(Min Error)和回归平方和(ESS)分别为0.017、0.13、0.023和0.28。与现有方法相比,MIR-EFS融合降维算法的复杂度性能与计算结果更优,这表明改进的MIR-EFS融合降维算法能够准确地获取成品汽油中辛烷值的含量,为汽油辛烷值预测提供可参考的算法支持。
The gasoline octane value not only affects the economic effectiveness of gasoline, but also relates to automobile exhaust emissions. Therefore, accurate prediction of the gasoline octane value is of great significance to the national economic development and environmental protection. Firstly, in order to accurately predict the octane value from the high-dimension industrial gasoline data set, a mutual information regression(MIR)-embedded feature selection(EFS) fusion dimensionality reduction algorithm was proposed to score the data features, and 30 features were screened out as the main variables for modeling. Secondly, the octane value prediction model was established based on an adaptive ensemble random forest learning algorithm. Lastly, based on multiple complementary discriminant criteria, a detailed simulation calculation was carried out to compare with existing methods. The results show that the improved MIR-EFS fusion dimensionality reduction algorithm reduces the data dimension by 89.65%, the training time by 81.43%, and its mean square error(MSE), root mean square error(RMSE), the minimum error(Min Error) and the regression squares sum(ESS) of the predicted data are 0.017, 0.13, 0.023 and 0.28, respectively. Compared with the existing methods, MIR-EFS fusion dimensionality reduction algorithm shows obvious better performance on the calculation complexity and calculation results,which proves the improved MIR-EFS fusion dimensionality reduction algorithm can accurately predict the octane value of finished gasoline, and provides reference algorithm support for octane value prediction of gasoline.
作者
郇钫策
江驹
余朝军
徐海燕
XUN Fangce;JIANG Ju;YU Chaojun;XU Haiyan(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,Jiangsu,China;College of Economics and Management,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,Jiangsu,China)
出处
《天然气化工—C1化学与化工》
CAS
北大核心
2022年第2期95-102,共8页
Natural Gas Chemical Industry
基金
国家自然科学基金项目(61673209,71971115)。
关键词
石油辛烷值
融合降维算法
高维数据
预测
gasoline octane value
fusion dimensionality reduction algorithm
high-dimensional data
prediction
