石油学报(石油加工) ›› 2020, Vol. 36 ›› Issue (5): 988-994.doi: 10.3969/j.issn.1001-8719.2020.05.012

• 研究报告 • 上一篇    下一篇

延迟焦化开工线腐蚀机理及温度模型预测

任佳,王西刚,赵梦恩,金浩哲   

  1. 浙江理工大学 机械与自动控制学院,浙江 杭州 310018

  • 收稿日期:2019-06-10 修回日期:2019-09-29 出版日期:2020-09-25 发布日期:2020-11-27
  • 通讯作者: 金浩哲,男,教授,博士,从事石油化工煤化工高风险装置流动腐蚀预测及安全保障技术方面的研究,Tel:0571-86843348;E-mail:haozhe2007@163.com E-mail:haozhejin@zstu.edu.cn
  • 作者简介:第一作者:任佳,女,副教授,博士,从事机器学习、工业过程软测量、故障分类与预测等相关领域的研究,E-mail:jren@zstu.edu.cn
  • 基金资助:

    国家自然科学基金项目(51876194)、国家重点研发计划项目(2017YFF0210406)和浙江省公益技术研究项目(LGG20F030007)资助

Corrosion Mechanism and Temperature Model Prediction of Delayed Coking Start-Up Pipeline

REN Jia, WANG Xigang, ZHAO Meng’en, JIN Haozhe   

  1. Faculty of Mechanical Engineering & Automation, Zhejiang Sci-Tech University, Hangzhou 310018, China
  • Received:2019-06-10 Revised:2019-09-29 Online:2020-09-25 Published:2020-11-27

摘要: 以某延迟焦化装置为研究对象,明确了延迟焦化装置开工线发生的低温湿硫化氢(H2S)腐蚀问题并分析了腐蚀机理,在此基础上确立了开工线温度作为腐蚀失效的表征参量;由于开工线温度直接测量有延迟、成本高,且延迟焦化生产中各环境变量具有较强的非线性、时变性和复杂性,在高斯过程回归(GPR)模型的基础上,提出了基于引力搜索算法(GSA)优化的复合核函数高斯过程回归(GSA-CKGPR)模型,实现了开工线温度的软测量。通过对实际延迟焦化过程数据的训练预测分析,表明该预测模型相比于单核GPR模型、支持向量回归机(SVR)模型以及其他启发式优化算法具有更好的预测精度和稳定性,相对GPR模型均方根误差降低了47.3%,有利于延迟焦化开工线温度的精准预测,可为该装置的工艺操作参数优化及安全稳定运行提供理论支撑。

关键词: 延迟焦化, H2S腐蚀, 开工线温度预测, 高斯过程回归, 万有引力搜索

Abstract: Low-temperature wet hydrogen sulfide corrosion on the start-up pipeline of a delayed coking unit was identified and corrosion mechanism was discussed. Based on the above results, start-up pipeline temperature was chosen as the corrosion failure indication parameter. Direct start-up pipeline temperature measurement is expensive and always time-delayed, and environmental variables in the delayed coking process is characterized with strong nonlinearity, time-variation and complexity. Therefore, based on the traditional Gaussian process regression (GPR) model, a composite kernel function Gaussian regression model based on gravity search algorithm (GSA) optimization (GSA-CKGPR) was proposed to indirectly measure the start-up pipeline temperature. Through prediction analysis of the delayed coking process data, compared with the support vector regression machine (SVR) model, single-kernel GPR model and other heuristic optimization algorithms, the proposed GSA-CKGPR model is more accurate and stable. Compared with the traditional GPR model, the root mean square error can reduce 47.3%. The proposed model can accurately predict the temperature of delayed coking start-up pipeline and also can be used as a tool to support process optimization, and improve safety and reliability of the unit.

Key words: delayed coking, hydrogen sulfide corrosion, start-up pipeline temperature prediction, Gaussian process regression, gravity search algorithm