基于红外光谱利用机器学习快速测定生活垃圾中的水分  

Employing Machine Learning to Rapidly Determinate the Moisture in Municipal Solid Waste Based on Infrared Spectroscopy Technique

作　　者：宋立杰 魏然 张瑞娜 赵由才[2,3] 周涛 Song Lijie;Wei Ran;Zhang Ruina;Zhao Youcai;Zhou Tao(Shanghai Environmental Sanitation Engineering Design Institute Co.,Ltd.,Shanghai 200232;State Key Laboratory of Pollution Control and Resource Reuse,Tongji University,Shanghai 200092;Shanghai Institute of Pollution Control and Ecological Security,Shanghai 200092,China)

机构地区：[1]上海环境卫生工程设计院有限公司,上海200232 [2]同济大学环境科学与工程学院,上海200092 [3]上海污染控制与生态安全研究院,上海200092

出　　处：《广东化工》2021年第18期166-168,共3页Guangdong Chemical Industry

基　　金：生活垃圾高效脱水及快速检测技术(No.18DZ1202604)。

摘　　要：本文选取棉花为生活垃圾典型代表物,利用红外光谱探讨不同含水率的反射波峰分布情况。应用了线性回归和BP神经网络模型,构建映射反射峰与含水率之间的关系。同时,也探讨BP神经网络映射反射峰与含水率的最优参数。实验结果表明,随着棉花含水率的增加,其反射峰呈下降的趋势;BP神经网络模型(R^(2)=98.49%)比线性回归模型(R^(2)=74.72%)能更加有效反映反射峰与含水率之间的关系。BP神经网络最佳识别参数为学习率为0.001,迭代次数为150和隐藏层个数为4326个,在此条件模型获得预测值与试验测定含水率的相关性为99.11%。该实验结果将为后期测定多组分掺杂和实际生活垃圾中的水分快速测定奠定基础。The aim of this study is to establish the model to map the relationship between reflection peak and moisture in cotton as MSW represented employing the linear regression(RL)and machine learning,namely back propagation neural network(BPNN)based on infrared spectroscopy.Meanwhile,the optimized parameters for BPNN to predict the moisture in cotton are also discussed.The results indicate that the reflection peak shows decreased trend as the amount of moisture increased.The performance of BPNN(R^(2)=98.49%)is better than RL(R^(2)=74.72%)to map the relationship between reflection peak and moisture.The values of 0.001,150 and 4326 for learning rate,epoch and the number of neurons best fit BPNN to establish the correlation with predicted and experienced moisture(R^(2)=99.11%).This result could provide the reference for the latter rapidly determinate the moisture in multi-componence mixture and actual MSW.

关 键 词：快速测定水份 红外光谱 BP神经网络 生活垃圾 

分 类 号：TQ[化学工程]