基于SSA-RELM的S Zorb装置在线产品预测及多目标操作优化分析

根据某炼油厂S Zorb装置的生产工艺和操作规范，用24个操作变量与精制汽油主产品的流量和硫含量的实际生产数据进行了相关性分析，压缩为10个操作变量后建立了基于Aspen Plus的生产过程机理模型；经随机抽样检验和灵敏度分析后，以原料进料流量和硫含量、加热炉进口温度、加氢石脑油进料流量、热分压力、热分温度、干气出装置温度、冷分温度为输入，精制汽油的流量、硫含量和氮含量为输出运行机理模型，拓展了装置的在线产品预测数据集；在此拓展数据集上，采用基于麻雀搜索算法的正则化极限学习机(SSA-RELM)建立了装置的在线产品预测数据驱动模型；最后以进料分区，将精制汽油流量、硫含量和氮含量为优化目标，给出了6个分区的在线操作最优化方案。

S Zorb Device Online Product Prediction and Multi-Objective Operation Optimization Analysis Based on SSA-RELM

To achieve online product prediction and multi-objective operation optimization of S Zorb unit in a refinery, twenty four operation variables in the process were used as inputs, and corelated them with flowrate and sulfur content of gasoline product. Then, ten operation variables with high correlation coefficient were selected. Based on the above work, a production process mechanism model on Aspen Plus software platform was built. Furthermore, through random sampling inspection based on actual production data and sensitivity analysis with Aspen Plus, eight operation variables (i.e., feed flowrate and its sulfur content, furnace inlet temperature, hydrogenated naphtha feed flowrate, hot separator pressure and temperature, dry gas outlet temperature and cold separator temperature) were taken as final inputs, and flowrate of gasoline product and its sulfur and nitrogen contents as outputs, mechanism model on Aspen Plus was performed to expand the online product prediction database of S Zorb unit. On the extended database, sparrow search algorithm based on regularized extreme learning machine (SSA-RELM) was employed to establish online product prediction data-driven model. Finally, flowrate of gasoline product and its sulfur and nitrogen contents were optimized based on six feed partitions, and multi objective online operation optimization schemes were provided.