基于回归算法的渣油加氢装置反应温度预测及系统实现  

System Realization and Reaction Temperature Prediction in Residual Oil Hydrogenation Unit Based on Regression Algorithm

作　　者：卢文君 张金蓉 崔瑞利 张弢 宋俊男 姚远 崔鹏 金玮[1] 侯士超 LU Wenjun;ZHANG Jinrong;CUI Ruili;ZHANG Tao;SONG Junnan;YAO Yuan;CUI Peng;JIN Wei;HOU Shichao(Kunlun Digital Intelligence Technology Co.,Ltd.,CNPC,Beijing 102206,China;Petrochemical Research Institute Company Limited,CNPC,Beijing 102206,China)

机构地区：[1]中国石油天然气集团公司昆仑数智科技有限责任公司,北京102206 [2]中国石油集团石油化工研究院有限公司,北京102206

出　　处：《石油学报（石油加工）》2024年第5期1345-1352,共8页Acta Petrolei Sinica(Petroleum Processing Section)

基　　金：中国石油天然气集团有限公司科学研究与技术开发项目(2021DJ7104)基金资助。

摘　　要：目前对渣油加氢催化剂生命周期预测的技术较少,研究可方便探知催化剂活性状态的解决方案,将有效助力渣油加氢装置的催化剂更换管理。根据影响催化剂活性的多种因素,机器学习建模过程中选择了装置运行数据的运行时间、原料及产品性质等直接变量,并通过计算添加了金属沉积量、总加工负荷等组合变量作为特征,进行算法筛选调整参数,拟合装置运行周期过程的升温规律。训练后模型对测试周期反应温度预测的平均绝对百分比误差0.51%,进而可以通过反应温度的经验阈值得到催化剂预期寿命。考虑到生产数据与实验数据可能存在的分布差异,根据研究成果设计实现的软件系统操作灵活,为专业工程技术人员提供了便捷有效的辅助研究工具。At present,there are few technologies for predicting the life cycle of residual oil hydrogenation catalyst.The research on the solution scheme which can conveniently detect the active state of catalyst will effectively assist the catalyst replacement management of residual oil hydrogenation unit.According to various factors affecting catalyst activity,direct variables such as the running time,raw materials and product properties of device operation data were selected in the process of machine learning modeling,and combined variables such as metal deposition amount and total processing load were added as the features through calculation to carry out algorithm screening and parameter adjustment,and fit the temperature rise law in device running cycle process.After training,mean absolute percentage error of the model for predicting the reaction temperature for each test cycle is 0.51%,and then the catalyst life expectancy can be obtained through the experience threshold of reaction temperature.Considering the possible distribution differences between production and experiment data,the software system designed and implemented according to the research results is flexible in operation,providing convenient and effective auxiliary research tools for professional engineers and technicians.

关 键 词：渣油加氢 催化剂寿命 反应温度预测 机器学习 系统实现 

分 类 号：TE624[石油与天然气工程—油气加工工程]