Abstract:

In the context of the national "dual carbon" initiative, optimizing process parameters to reduce energy consumption in chemical processes is crucial for energy conservation and emission reduction. The catalytic hydration process for producing ethylene glycol is energy-intensive, requiring operational parameter optimization to minimize energy consumption. However, achieving simultaneous multi-parameter optimization solely based on process simulation is challenging due to process complexity. This paper proposed a solution that integrated process simulation-generated data, surrogate model construction, and synchronized optimization of thermal integration. Process simulation was conducted for the annual production of 300000t of ethylene glycol from ethylene oxide. Surrogate model output variables were determined based on utility engineering locations, while sensitivity analysis was used to determine input variables. Sobol random sequences were used to generate sample points, and simulation provided real data considering the mechanistic model. A data-driven approach using neural networks was utilized to construct the surrogate model. Finally, a synchronous algorithm, combining a genetic algorithm and a D-G model, was utilized to optimize the surrogate model with the goal of minimizing the total public utility project cost. The obtained optimal process parameters led to a 4.89% reduction in cost compared to that before optimization. This demonstrated the effectiveness of the method and highlighted its potential for addressing complex full-process synchronous optimization problems in thermal integration.

Key words: ethylene oxide, ethylene glycol, neural networks, genetic algorithm, D-G model, surrogate model, optimization

摘要：

在当前国家“双碳”目标下，通过优化过程参数以减少化工过程能耗是节能减排的有效手段。环氧乙烷催化水合制乙二醇过程耗能高，优化其工艺参数以降低能耗是非常必要的，但由于生产流程复杂，单纯基于流程模拟很难实现多参数的同步最优化。因此本文采用流程模拟生成数据-构建代理模型-同步优化热集成的思路解决该问题。对年产30万吨环氧乙烷制乙二醇工艺进行流程模拟，根据流程中所需公用工程位点确定代理模型输出变量，通过灵敏度分析进一步确定输入变量。由Sobol随机序列生成样本点并通过模拟得到基于机理模型的可靠数据，以数据驱动方式训练神经网络得到代理模型。最后，以总公用工程费用最小为目标，采用遗传算法与D-G模型组成的同步算法对代理模型进行优化，得到最优工艺参数，总公用工程费用较优化前降低4.89%，既证明了方法的有效性，又展示了其在解决复杂全流程同步优化热集成问题方面的应用前景。

关键词: 环氧乙烷, 乙二醇, 神经网络, 遗传算法, D-G模型, 代理模型, 优化