摘要：

树脂固定床吸附工艺在含重金属废水深度处理中广泛应用，穿透曲线是其设计和优化的重要依据。当前，穿透曲线主要通过拟合柱实验数据获取经验模型参数来模拟，但该方法耗时耗力，且无法量化传质系数与多个影响因素间的关系。通过文献数据挖掘建立数据集，以Thomas模型参数k _Th和q ₀为预测目标，训练并评价了决策树（DT）、随机森林（RF）、梯度提升（GBDT）和自适应增强（AdaBoost）4种机器学习（ML）模型，构建了AdaBoost-Thomas混合模型预测重金属吸附树脂固定床穿透曲线。结果表明，AdaBoost模型在预测k _Th和q ₀中表现优异，其测试集上的R ²分别为0.755和0.832。AdaBoost-Thomas混合模型对重金属吸附树脂固定床穿透曲线具有良好的预测效果，相比传统方法，该模型能够在无柱实验数据支撑下，实现固定床穿透曲线快速、准确预测。

关键词: 机器学习, 树脂, 重金属吸附, 穿透曲线预测, 混合模型

Abstract:

The fixed-bed filled with resin adsorption process is widely used in the advanced treatment of industrial wastewater containing heavy metals, in which breakthrough curve is vital for its design and optimization. At present,the breakthrough curve is mainly simulated by fitting the column experimental data to obtain the parameters of empirical models. However, it is not only time-consuming and labor-intensive, but also unable to quantify the relationship between parameters and multiple factors. In this study, a dataset was established through mining the literature data, and four machine learning(ML) models, including DT, RF, GBDT and AdaBoost, were trained and evaluated with k _Th and q ₀ as prediction targets. AdaBoost-Thomas hybrid model for breakthrough curve prediction of heavy metal adsorption resins was constructed. The results showed that the AdaBoost model exhibited excellent prediction performance in predicting k _Th and q ₀, with R ² of 0.755 and 0.832 on the test set, respectively. The AdaBoost-Thomas model had a great prediction on the fixed-bed breakthrough curve of heavy metal adsorption resins. Compared with traditional methods, the hybrid model could achieve fast and accurate prediction of the fixed-bed breakthrough curve without the support of column experimental data.

Key words: machine learning, resin, heavy metal adsorption, breakthrough curve prediction, hybrid model

中图分类号: 

• X703