基于随机森林算法的油气回收装置出口浓度预测研究  

Study on the Prediction of Oil and Gas Recovery Device Exhaust Concentration Based on Random Forest Algorithm

作　　者：郑华俊 汪佳奇 李华骐 汪小川 陈志伟 Zheng Huajun;Wang Jiaqi;Li Huaqi;Wang Xiaochuan;Chen Zhiwei(Zhejiang Oil Storage and Transportation Co.,Ltd.,Zhejiang,311227;Zhejiang Key Laboratory of Petrochemical Environmental Pollution Control/National&Local Joint Engineering Research Center of Harbor Oil&Gas Storage and Transportation Technology/School of Petrochemical Engineering&Environment,Zhejiang Ocean University,Zhejiang,316022)

机构地区：[1]浙江油品储运有限公司,浙江311227 [2]浙江海洋大学石油化工与环境学院/临港石油天然气储运技术国家地方联合工程研究中心/浙江省石油化工环境污染控制重点实验室,浙江316022

出　　处：《当代化工研究》2025年第5期45-47,共3页Modern Chemical Research

基　　金：2023年浙江油品储运公司-横向委托项目“油库油气回收系统工艺参数研究与优化”(项目编号:21028008823);2024年浙江海洋大学-启动经费项目“人才引进科研基金”(项目编号:JX6311021223)。

摘　　要：基于油库运行参数,构建数据驱动模型预测油气出口浓度。利用Spearman系数发现压缩机出口温度会对油气出口浓度产生影响,通过随机森林算法得到的特征重要性排序为:吸附罐1温度>入口油气温度>制冷机出口温度>吸附罐2温度>压缩机出口温度>入口油气压力。将上述参数输入随机森林模型中,计算结果表明该模型可以较好的预测油气出口浓度,引入压缩机出口温度能使模型的拟合优度R^(2)值达到0.75以上,出口油气浓度的平均误差为1g/m^(3)。利用机器学习的方法可以预测出口油气浓度,为油库降本增效提供指导。Construction of a data-driven model to predict oil and gas export concentrations based on reservoir operating parameters.Spear-man correlation analysis shows that compressor outlet temperature significantly influences the oil and gas export concentration.Feature importance,derived from the random forest algorithm,ranks as follows:Adsorption tank 1 temperature>entrance oil vapor temperature>condenser outlet tem-perature>adsorption tank 2 temperature>compressor outlet temperature>entrance oil vapor pressure.A random forest model was constructed using these parameters.Results indicate effective prediction of exit oil vapor concentration with an R^(2) value of 0.75 or higher and an average predic-tion error of only 1 g/m^(3).This study offers valuable insights for reducing costs and enhancing efficiency in oil depots through machine learning-based predictions.

关 键 词：油气回收 工艺参数 随机森林算法 出口浓度 预测 

分 类 号：TE3[石油与天然气工程—油气田开发工程]