State Key Laboratory of Clean Energy Utilization, Zhejiang University
College of Chemistry and Materials Engineering, Zhejiang A&F University
Institute of Energy and Power Engineering, Zhejiang University of Technology
College of Mechanical Engineering, Hangzhou Dianzi University
Jinhua Ningneng Thermal Power Co., Ltd.

通过实验方法测定多源有机固废的热化学转化特性通常是耗时且劳力密集的过程,借助机器学习的方法可以挖掘不同原料特性与基本热化学转化特性之间的关联机制,并快速进行预测分析。构建了一个综合数据库,其中包含38种工业有机固废基本特性及热解特性信息。通过描述性统计分析、相关性分析以及主成分分析(PCA),深入探究了数据库的总体规律。进一步采用随机森林(RF)、梯度提升树(GBDT)和极限梯度提升树(XGBoost)算法对有机固废高位热值(HHV)、快速热解产物分布和不同气氛下热失重曲线进行预测,其中对HHV、热解产物分布和热失重曲线预测的R2分别在0.835~0.866、0.701~0.875和0.976~0.980范围内。最后,基于树模型的平均杂质减少(MDI)和SHapleyAdditiveexPlanations(SHAP)方法对建模结果进行可解释性分析,筛选出在模型决策过程中起关键作用的特征,并揭示了原料基本特性与HHV、热解产物分布及热解特性之间的关联,旨在为实际有机固废的智能管理与高效处置提供一定的指导。

Experimental determination of thermochemical conversion characteristics of multi-source or⁃ganic solid wastes is a time-consuming and labor-intensive process. By leveraging machine learningmethods, the correlation mechanism between different feedstock properties and thermochemical charac⁃teristics can be explored to enable fast and accurate prediction. A comprehensive dataset was construc⁃ted based on the fundamental properties and pyrolysis characteristics of 38 types of industrial organicsolid waste. Descriptive statistical analysis, correlation analysis, and principal component analysis(PCA) were employed to uncover patterns within the dataset. Subsequently, the random forest (RF),gradient boosting decision tree ( GBDT), and extreme gradient boosting ( XGBoost) algorithms wereutilized to predict the high heating value (HHV) of organic solid waste, the distribution of fast pyrolysisproducts, and the thermogravimetric curves under various atmospheres. The R values achieved forHHV, product distribution, and thermogravimetric curves ranged from 0.835 to 0.866, 0.701 to 0.875,and 0.976 to 0.980, respectively. Additionally, the Mean Decrease Impurity (MDI) and SHapley Ad⁃ditive exPlanations (SHAP) methods were applied to analyze the model′s performance and identify keyfeatures influencing the model′s decision-making process. This allowed for explaining the relationshipbetween feedstock properties and HHV. It also enabled explaining the connection between product dis⁃tribution and pyrolysis characteristics. This study aims to offer valuable insights into the intelligent man⁃agement and efficient disposal of organic solid waste.

浙江省“领雁”研发攻关计划资助项目(2022C03092)

张子杭,邢博,马中青,等.多源有机固废热解特性研究与预测分析[J].能源环境保护,2024,38(5):135-146.

ZHANG Zihang, XING Bo, MA Zhongqing, et al. Research and predictive analysis of pyrolysis characteristics ofmulti-source organic solid wastes[J]. Energy Environmental Protection, 2024, 38(5): 135-146.