【作者】 张欣；

【导师】 吉旭； 周利；

【作者基本信息】 四川大学 ， 化学工艺， 2022， 硕士

【摘要】 近年来,由于原油品质的日益恶化和环保要求的愈加严格,炼油企业对含硫重质原油的加工比例和深度不断增加。加氢裂化和加氢精制是炼油厂内重要的处理原油的工艺,同时也是氢气大量消耗的主要途径。在这样的背景下,氢气资源的需求量不断上涨,氢公用工程费用也成为炼油厂内主要的生产成本。除此之外,在实际生产过程中,原油性质并非始终保持稳定,油田开采深度的变化以及油品配置结构的调整会使其在一定程度上发生波动,进而影响炼厂氢气网络的运行。因此,为了降低生产成本,需要对炼厂氢气资源系统进行集成优化,而具有明确工艺参数的确定性氢气网络优化模型可能不适用于实际操作,氢气需求的不断增加以及原油性质的不确定性要求对氢气网络进行更为稳健的管理。针对于此,本文提出了一种基于质量传递机理的随机规划建模框架,旨在实现氢气网络经济效益和抗扰动能力的同步优化。该框架采用简化模型、机理模型和代理模型集成关键操作单元,可分析原油性质变化对氢气网络运行的影响;利用二阶段随机规划应对可变的原油性质,给出最优设计决策和不同场景下的运行方案,提升氢气系统的经济效益和操作稳健性。主要的工作内容如下:（1）基于简化模型或经验模型将关键的设备单元嵌入氢气网络中,可根据原油性质数据计算加氢反应器入口油品杂质组分含量、加氢精制过程中的耗氢量以及气体生成量等过程变量。过程模型所涉及的部分参数是通过实际炼厂数据计算得到,因此这类模型具有高计算效率的同时,也能够确保一定的模型精度。（2）利用代理模型技术对脱硫单元建模,考虑关键杂质H₂S的脱除,有效地解决杂质气体含量对氢气资源回收利用的限制。模型的构建是基于严格的MDEA脱硫过程,根据实际工程优化任务确定其输入、输出变量及拟合空间;然后采用Sobol采样法进行空间采样,并通过Aspen plus软件仿真模拟,获取数据样本集;最后利用机器学习算法和交叉验证法训练、验证模型,以R²、RMSE及残差图评估模型的精度及可靠性。实例计算表明,所建代理模型的R²值均高于0.98,RMSE值低于0.017,可以很好地拟合脱硫塔的实际生产数据。同时,考虑H₂S的脱除也确保了氢气资源的有效回用。（3）针对原油性质的不确定性,以总年度费用为目标,采用了二阶段随机规划法（SP）对氢气网络建模优化。第一阶段设计氢气网络结构,第二阶段基于随机参数优化网络运行。前者的决策包括了输送管道的配置,设备的配置和设备容量大小的设计。后者的决策涉及了网络运行参数,即供氢量、气体输送量和组分浓度等。最后,本文以国内某大型炼油厂为例验证所提方法的有效性。结果表明,考虑原油性质波动的SP模型具备更高的操作灵活性,并能够针对不同的操作场景给出最佳运行方案,表现出更好的经济性和鲁棒性。更多还原

【Abstract】 In recent years,due to the deteriorating quality of crude oil and the increasingly strict environmental protection requirements,the proportion and depth of sulfur-containing heavy crude oil processing in refineries are increasing.Hydrocracking and hydrotreating are important processes for processing crude oil in refineries,and they are also the main way of hydrogen consumption.Therefore,the demand for hydrogen resources is rising,and the cost of hydrogen utilities has become the main production cost in refineries.In addition,the properties of crude oil are not always stable in the actual production process.The change of oilfield exploitation depth and the adjustment of oil product allocation structure will make it fluctuate to a certain extent,which will affect the operation of refinery hydrogen network.Therefore,integrated optimization of the hydrogen resource system of the refinery is required to reduce production costs.The deterministic hydrogen network optimization model with will-defined process parameters may not be suitable for practical operation.The increasing demand for hydrogen and the uncertainties of crude oil property call for more robust management of hydrogen networks.In view of this,a stochastic programming modeling framework based on mass transfer mechanism was proposed to achieve the simultaneous optimization of economic performance and disturbance rejection ability of hydrogen network in this paper.The framework integrated key operating units with simplified model,mechanism model and agent model to analyze the impact of changes in crude oil properties on the operation of hydrogen network;used two-stage stochastic programming to deal with the variable properties of crude oil,and gave optimal design decisions and operation schemes under different scenarios to improve the economic benefits and operation robustness of the hydrogen system.The main work contents are as follows:（1）Based on simplified model or empirical model,the key equipment units are embedded in the hydrogen network,so that the content of impurities in the inlet oil of the hydrogenation reactor,the hydrogen consumption and the gas production in the hydrotreating process and other process variables can be calculated according to the crude oil property data.Some parameters involved in the process model are calculated from the actual refinery data,so this kind of model not only had high computational efficiency,but also can ensure a certain model accuracy.（2）The desulfurization unit was modeled by surrogate-assisted techniques to consider the removal of H₂S,so that the limitation of impurity gas content on the recovery and utilization of hydrogen resources was solved.The construction of the model was based on the strict MDEA desulfurization process,and the input,output variables and fitting space of the model depended on the actual engineering optimization task;then,Sobol sampling method was used for spatial sampling,and Aspen plus software was used to simulate to obtain the data sample set;finally,the machine learning algorithm and cross-validation method were used to train and verify the model,and the accuracy and reliability of the model were evaluated by R²,RMSE and residual graph.Example calculations showed that the R² values of the agent models were above0.98 and the RMSE values were below 0.017.The surrogate model can fit the actual production data of desulfurization tower well.At the same time,considering the removal of H₂S also ensured the effective reuse of hydrogen resources.（3）Aiming at the uncertainty of the crude oil properties in the refinery,with the total annual cost as the target,the two-stage stochastic programming method（SP）was adopted to model and optimize the hydrogen network.In the first stage,the hydrogen network structure was designed,and in the second stage,the network operation was optimized based on random parameters.The former decision included the configuration of transmission pipeline,the configuration of equipment and the design of equipment capacity.The latter decision involved the network operation parameters,such as hydrogen supply,gas transmission and component concentration.Finally,a large domestic refinery was taken as an example to verify the effectiveness of the proposed method.The results show that the SP model considering the fluctuation of crude oil properties had higher operational flexibility,and can give the best operation scheme for different operation scenarios,showing better economy and robustness.更多还原