王金咏,王新元,魏文光,张凤山,黄鹏,周景蓬,万兵,牛国强,刘鸿斌.基于集成深度学习的造纸废水出水指标预测模型研究[J].中国造纸学报,2025,(2):173-182 本文二维码信息
二维码(扫一下试试看!)
基于集成深度学习的造纸废水出水指标预测模型研究
Study on Intelligent Prediction Model of Papermaking Wastewater Effluent Quality Index Based on Ensemble Deep Learning
投稿时间:2024-08-07  修订日期:2024-10-21
DOI:10.11981/j.issn.1000-6842.2025.02.173
中文关键词:  造纸废水过程处理  数据降维  长短期记忆网络  集成学习  软测量模型
Key Words:papermaking wastewater treatment process  data dimensionality reduction  long short-term memory network  ensemble learning  soft sensor mode
基金项目:山东省自然科学基金(ZR2021MF135);江苏省高等学校自然科学研究重大项目(22KJA530003)。
作者单位邮编
王金咏* 南京林业大学江苏南京210037 210037
王新元 南京林业大学江苏南京210037 210037
魏文光 山东华泰纸业股份有限公司山东东营257335
山东黄三角生物技术产业研究院有限公司山东东营257399 
257399
张凤山 山东华泰纸业股份有限公司山东东营257335
山东黄三角生物技术产业研究院有限公司山东东营257399 
257399
黄鹏 山东华泰纸业股份有限公司山东东营257335 257335
周景蓬 山东华泰纸业股份有限公司山东东营257335 257335
万兵 山东华泰纸业股份有限公司山东东营257335 257335
牛国强 华南师范大学广东广州510006 510006
刘鸿斌* 南京林业大学江苏南京210037
山东华泰纸业股份有限公司山东东营257335 
257335
摘要点击次数: 0
全文下载次数: 0
中文摘要:
      为克服单一模型的局限性、提高模型鲁棒性,针对小型造纸厂单一工段的废水处理数据集,首先利用核主成分分析(KPCA)降维技术,有效提取数据关键特征,再采用装袋集成(Bagging)算法集成多个可有效建模废水时间序列特征的长短期记忆网络(LSTM)学习器,建立KPCA-Bagging-LSTM造纸废水出水指标预测模型。结果表明,KPCA-Bagging-LSTM模型的决定系数(R2)达0.76,显著优于其他方法;均方根误差(RMSE)和平均绝对百分比误差(MAPE)分别为3.55 mg/L和4.01%,表明该模型具有更低的预测误差和更高的精度。本研究通过特征降维和集成学习提升了KPCA-Bagging-LSTM模型的性能,为造纸废水COD等出水指标预测提供了有效的解决方案。
Abstract:
      To address the limitations of single model and improve overall robustness, this study focused on a wastewater treatment dataset from a single processing stage of a small-scale papermaking plant, took kernel principal component analysis (KPCA) for dimensionality reduction to effectively extract key features from the data, firstly. Then, integrated multiple long-short term memory (LSTM) learners using the Bagging ensemble strategy, and constructed a KPCA-Bagging-LSTM prediction model, where LSTM was capable of modeling the temporal characteristics of wastewater data. The results showed that the KPCA-Bagging-LSTM model achieved a coefficient of determination (R²) of 0.76, significantly outperforming other methods. The Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) were 3.55 mg/L and 4.01%, respectively, indicating lower prediction error and higher accuracy. By combining feature dimensionality reduction with ensemble learning, the proposed KPCA-Bagging-LSTM model enhanced prediction performance and provided an effective solution for forecasting COD and other effluent indicators in papermaking wastewater treatment.
查看全文  查看/发表评论  下载PDF阅读器  HTML

分享按钮