近红外光谱和极限学习机分析反硝化除磷中胞内聚合物

Study on intracellular polymers using near infrared spectroscopy and extreme learning machine in denitrifying phosphorus removal process

为利用近红外光谱技术同时实现反硝化除磷工艺中胞内聚-β-羟基丁酸酯(PHB)、多聚磷酸盐(Poly-P)、糖原(Gly)含量的快速分析,采用多元散射校正预处理法和极限学习机算法建立PHB、Poly-P、Gly含量分析的校正模型(ELM模型).结果表明:多元散射校正预处理法可以有效消除散射对原始近红外光谱数据的影响.采用极限学习机算法对预处理后的光谱数据建立PHB、Poly-P、Gly的定量分析模型,优选出的PHB、Poly-P、Gly的ELM模型主成分数分别为6、6、7,隐含层节点数分别为18、12、17.模型对PHB、Poly-P、Gly含量的校正相关系数(rc)分别为0.9835、0.9499、0.9589,校正均方根误差(RMSECV)分别为0.0541、0.0579、0.0489.模型对PHB、Poly-P、Gly含量的预测相关系数(rp)分别为0.9683、0.9288、0.9488,预测均方根误差(RMSEP)分别为0.0668、0.0776、0.0501,模型对PHB、Poly-P、Gly含量有较好的预测效果.用近红外光谱技术结合极限学习机算法建立ELM模型为反硝化除磷工艺中PHB、Poly-P、Gly的同时快速定量分析提供了较为简便的方法.

In order to realize rapid determination of intracellular poly-β-hydroxybutyrate(PHB),polyphosphate(Poly-P)and glycogen(Gly)in denitrifying phosphorus removal process with near infrared spectroscopy,the calibration models(ELM models)of PHB,Poly-P,Gly were established by multiple scatter correction preprocessing and extreme learning machine algorithm.The preprocessing results showed that the multiple scattering correction can eliminate the scatteringeffects on the raw near infrared spectral data of PHB,Poly-P and Gly.The ELM models of PHB,Poly-P and Gly were established with preprocessed spectral data by extreme learning machine.The principal component numbers of ELM models of PHB,Poly-P and Gly were respectively6,6and7,with the nodes number of hidden layer being18,12and17respectively.The ELM models of PHB,Poly-P and Gly showed that the correlation coefficients(rc)were respectively0.9835,0.9499,0.9589,with the root mean square errors of cross validation(RMSECV)being0.0541,0.0579,0.0489respectively.The prediction results of ELM models of PHB,Poly-P and Gly indicated that the correlation coefficient(rp)were respectively0.9683,0.9288,0.9488,with the root mean square errors of prediction(RMSEP)being0.0668,0.0776,0.0501.It showed that ELM models of PHB,Poly-P and Gly had better prediction performance for the contents of PHB,Poly-P and Gly.This study provides a convenient method for rapid determination of PHB,Poly-P and Gly in denitrifying phosphorus removal process with near infrared spectroscopy and extreme learning machine.