Abstract:

With the increase of global plastics production, the problem of waste plastics disposal has become increasingly serious. Pyrolysis technology has attracted widespread attention as a method to convert waste plastics into high value-added products. The pyrolysis process of polyethylene (PE) was investigated by combining molecular-level kinetic model with machine learning methods. First, a large-scale pyrolysis dataset was generated for PE raw materials with different molecular weight distributions using molecular level kinetic model. Then, 9 machine learning models were constructed based on the large-scale dataset to evaluate their predictive ability and feature importance, and analyze the key factors affecting the product yields. The results show that reaction time and pyrolysis temperature are the main factors, and the KNN model performs the best in the prediction of gas and liquid phase products. The study also demonstrates that the simulation accuracy and efficiency of the pyrolysis process can be significantly improved by optimizing the machine learning model and expanding the dataset, which provides new ideas and methods for waste plastics resourcing.

Key words: pyrolysis, machine learning, molecular level kinetic model, waste plastics, simulation, waste treatment

摘要：

随着全球塑料产量的增加，废弃塑料处理问题日益严重，热解技术作为一种将废弃塑料转化为高附加值产品的方法，引起了广泛关注。通过分子水平动力学模型与机器学习相结合的方法，研究了聚乙烯（PE）的热解过程。首先，利用分子水平动力学模型针对不同分子量分布的PE原料，生成大规模热解数据集。然后，基于大规模数据集构建了9种机器学习模型，评估其预测能力及特征重要性，分析影响热解产物收率的关键因素。结果表明，反应时间和热解温度是主要影响因素，KNN模型在气、液相产物预测中表现最佳。通过优化机器学习模型和扩大数据集，可以显著提升热解过程的预测准确性和效率，为废弃塑料资源化提供了新的思路和方法。

关键词: 热解, 机器学习, 分子水平动力学模型, 废弃塑料, 模拟, 废物处理