关键词: 煤制乙二醇, 数据驱动建模, 最小二乘支持向量机, 能耗, 质量

Abstract: With the rapid development of polyester industry, the increasing demand of ethylene glycol (EG) is in conflict with the shortage of supply in China. Large project of EG production from coal has been receiving more and more attention. In the production of coal-to-ethylene glycol, the optimization of distillation operations represents a vital means to achieve energy saving and consumption reduction, as well as quality enhancement and efficiency improvement. The foundation of optimization lies in the establishment of precise models for the process. However, due to the complex reactions, strong system coupling, and non-linearity inherent in the distillation process, it is difficult to accurately construct models using traditional mechanistic methods. Therefore, this study uses the distillation process of coal-to-ethylene glycol as the research subject, employing a refined least squares support vector machine (LSSVM) algorithm to accurately construct energy consumption and product quality models for the ethylene glycol distillation process. In this process, the actual industrial data from the coal-to-ethylene glycol distillation process was used as the benchmark, the mutual information method was employed to extract the main feature parameters, and variable screening and data pre-processing were conducted. Subsequently, by introducing local target sets and using the UMDA algorithm for iterative optimization, the optimal hyperparameters were derived. After determining the optimal hyperparameters, the improved LSSVM algorithm was used to model the data samples and further compared this model with other purity and energy consumption models established by different algorithms. This comparison confirmed the high efficiency and accuracy of the improved LSSVM algorithm based on UMDA proposed in this work. In summary, compared with traditional support vector machine methods, the LOS-LSSVM model based on the UMDA optimisation process has a clear advantage in data fitting, accurately reflecting the actual situation of the distillation process and effectively improving the operational efficiency of ethylene glycol production.

Key words: coal-to-ethylene glycol, data driven modeling, least squares support vector machine, energy consumption, quality