PLS-BP法定量分析火灾中的常见有毒有害气体  

Study on the quantitative analysis of common fire toxic gases by PLS-BP model

作　　者：赵建华[1] 高明亮[1] 武秀娟[1] 

机构地区：[1]中国科学技术大学火灾科学国家重点实验室,安徽合肥230026

出　　处：《火灾科学》2010年第3期158-164,共7页Fire Safety Science

基　　金：安徽省自然科学基金资助项目(070415221)

摘　　要：针对火灾和材料燃烧中常见的有毒有害气体,选取低浓度的CO、CO2、NO、NO2、SO2、HCl、HBr、HCN八种典型气体同时进行定量分析。基于傅里叶变换红外光谱技术(FTIR),设计可靠实验系统和实验过程控制,准确得到气体的光谱数据。通过合理去除干扰光谱区间、筛选样本、选择模型参数等,建立BP-PLS回归模型,并对未知样本进行预测,各组分的校正误差均方根RMSEC达到4×10-6以下,预测可决系数R2均达到0.95以上。通过将PLS-BP模型与经典的线性模型经典最小二乘(CLS)和偏最小二乘(PLS)进行比较,PLS-BP模型在非线性拟合能力和预测性能两方面明显高于经典线性模型CLS和PLS。To study the concentration of common toxic gas in fire disaster or other burning materials,8 kinds of classic component with low concentration,including CO,CO2,NO,NO2,SO2,HCl,HBr and HCN,were chosen for the quantitative analysis.Based on FTIR,the experimental system were established by Fourier Transform Infrared Spectrometer,gas compounding equipment and other devices.Accurate spectrum data were obtained by strict control of experimental process.After appropriate spectrum region choosing,standards selecting and parameters setting,Partial Least Square-Back Propagation network（PLS-BP） model was established and calculated for unknown standards.The prediction deciding coefficient（R2）of each component is higher than 0.95.The root-mean standard error of calibration（RMSEC） is lower than 4×10-6.Compared PLS-BP model with classic linear CLS model and PLS model,PLS-BP model is much better in the performance of non-linear fitting and predicting unknowns.

关 键 词：PLS-BP模型 傅里叶变换红外光谱 有毒有害气体 定量分析 

分 类 号：TN247[电子电信—物理电子学] X924.2[环境科学与工程—安全科学]