| 孙儒燊,任世学,王伟.基于随机森林算法的制浆造纸碳足迹预测及影响因素分析[J].中国造纸,2026,45(6):142-152 |
二维码（扫一下试试看！） |
| 基于随机森林算法的制浆造纸碳足迹预测及影响因素分析 |
| Prediction of Carbon Footprint in Pulp and Paper Based on Random Forest Algorithm and Analysis of Its Influencing Factors |
| 收稿日期：2025-12-28 修订日期：2026-02-06 |
| DOI：10.11980/j.issn.0254-508X.2026.06.016 |
| 关键词: 碳足迹 制浆造纸 随机森林 XGBoost 生命周期评价 |
| Key Words:carbon footprint pulp and paper random forest XGBoost life cycle assessment |
| 基金项目:黑龙江省自然科学基金项目（LH2019C009）。 |
| 作者 | 单位 | 邮编 | | 孙儒燊* | 1东北林业大学木本油料资源利用全国重点实验室,黑龙江哈尔滨,150040
2东北林业大学材料科学与工程学院,黑龙江哈尔滨,150040 | 150040 | | 任世学* | 1东北林业大学木本油料资源利用全国重点实验室,黑龙江哈尔滨,150040
2东北林业大学材料科学与工程学院,黑龙江哈尔滨,150040 | 150040 | | 王伟* | 3东北林业大学机电工程学院, 黑龙江哈尔滨,150040 | 150040 |
|
| 摘要点击次数: 226 |
| 全文下载次数: 24 |
| 摘要:针对制浆造纸碳排放影响因素复杂、传统生命周期评价（LCA）难以快速支撑过程优化的问题，本研究构建了基于LCA与机器学习的纸产品碳足迹预测框架，以“从摇篮到大门”为系统边界，基于Ecoinvent v3.11和FisherSolve数据库构建数据集，数据集特征空间涵盖原材料、能源结构、工艺参数及废弃物处理等变量，标签为采用IPCC 2013 GWP 100a 方法计算的碳足迹强度。在此基础上，利用随机森林（RF）模型识别关键影响因素，并引入极限梯度提升模型（XGBoost）对RF模型预测残差进行再学习。结果表明，干燥段蒸汽消耗、电网排放因子、化石燃料占比、单位产品电耗和石灰窑煅烧排放是碳足迹主导影响因素；RF-XGBoost堆叠模型显著优于RF单一模型，R²由0.482提升至0.928；均方根误差（RMSE）由104.3 kg CO₂eq/t降至38.9 kg CO₂eq/t。 |
| Abstract:To address the complexity of influencing factors in carbon emissions in pulp and paper industry and the limitation of traditional life cycle assessment (LCA) in rapidly supporting process optimization， this study constructed a carbon footprint prediction framework for paper products based on LCA and machine learning. With a “cradle-to-gate” system boundary， a dataset was developed based on the Ecoinvent v3.11 and FisherSolve databases. The feature space of the dataset covered variables such as raw materials， energy structure， process parameters， and waste treatment， while the target variable was the carbon footprint intensity calculated using the IPCC 2013 GWP 100a method. On this basis， a random forest (RF) model was used to identify key influencing factors， and XGBoost was introduced to relearn the prediction residuals of the RF model. The results showed that drying-section steam consum ption， grid emission factor， fossil fuel ratio， unit electricity consumption， and lime kiln calcination emissions were the dominant factors. The RF-XGBoost stacked model significantly outperformed single model， with an R² increasing from 0.482 to 0.928 and a root mean square error (RMSE) decreasing from 104.3 kg CO₂eq/t to 38.9 kg CO₂eq/t. |
| 查看全文
HTML
查看/发表评论 下载PDF阅读器 |