基于随机森林的汽油精制过程中辛烷值损失模型

Octane loss model in gasoline refining process based on random forest

目前,随着汽车尾气排放污染日趋严重,汽油质量标准日益严格,中国大力发展以催化裂化为核心的重油轻质化工艺技术,对汽油进行精制处理,实现汽油清洁化。在实现汽油清洁化的过程中,会不可避免地降低辛烷值(RON),亦会同时出现较大损失值单位,无疑给企业增加了生产成本,减少了收益。为此,本文通过建立基于随机森林的汽油精制过程中RON损失预测模型,对RON及其指标进行预测。首先,命名建模变量并计算矩阵相关性,利用随机森林法对降低RON损失模型所涉及的158个变量进行二次降维,提取前30个主要变量;其次,基于随机森林法对样本数据进行划分,建立损失预测模型并对模型进行验证,得到预测值与真实值曲线对比图,保证所建模型合理化;最后,运用遗传算法对主要变量进行优化,力求将RON损失值降幅控制在15%以上,以此确保损失预测模型真实有效。

At present,with the increasingly serious automobile exhaust pollution and increasingly strict gasoline quality standards,heavy oil lightening process technology with catalytic cracking as the core,refining gasoline,and realizing gasoline cleaning are vigorously developed.In the process of realizing gasoline cleaning,the octane number(RON)will inevitably be reduced,and a large loss value unit will also appear at the same time,which will undoubtedly increase the production cost of the enterprise and reduce the income.To alleviate this problem,this paper predicts RON and its indicators by establishing a random forest-based prediction model for RON loss during gasoline refining.First,modeling variables are named,the matrix correlation is calculated,and the random forest method is used to perform secondary dimensionality reduction for the 158 variables involved in the RON loss reduction model to extract the first 30 main variables.Secondly,the sample data is divided based on the random forest method,the loss prediction model is established and verified,and the curve comparison between the predicted value and the actual value is obtained to ensure the rationalization of the model.Finally,the genetic algorithm is used to optimize the main variables and attempt to control the loss of RON to more than 15%to ensure that the loss prediction model is true and effective.