【摘要】 应用遥感傅里叶变换红外光谱,采用主成分提取-线性判别分析(PCA-LDA)技术,对丙酮、二氯甲烷、甲苯、苯、氯仿和甲醇等六组分的任意混合体系进行定性鉴别。被选用的这6种大气有毒有机化合物的红外光谱图相互间存在着严重的混叠,并和反向传播人工神经网络(BP-ANN)的预测结果进行了比较。PCA-LDA的鉴别判对率达92.2%,识别率94.4%,误判率7.8%;BP-ANN分别为91.1%、95.6%和8.9%。结果表明PCA处理克服了LDA对多变量数据预测的局限性,预测性能和BP-ANN相当。鉴于BP-ANN计算耗时和繁琐,PCA-LDA模型被确定为建立VOCs预警模型最适当的方法。更多还原

【Abstract】 The system, which contained one to six components including acetone, methylene chloride, toluene, benzene, chloroform and methanol, was analyzed qualitatively with the combination of principal component analasis-linear discriminate analysis(PCA-LDA) and remote sensing FTIR technique. There are FTIR spectra overlap ped seriously each other for the six air toxic organic compounds selected, The prediction results of PCA-LDA and BP-ANN were compared. The ratio of correct recognition ratio, recognition ratio and error recognition ratio of PCA-LDA were 92.2%, 94.4% and 7.8% respectively. The corresponding values of BP-ANN were 91.1%, 95.6% and 8.9%, respectively. The results demonstrated that limitations of LDA were overcome with PCA and then the performance of LDA was improved by PCA. The prediction performance of PCA-LDA was comparable to BP-ANN. Considering time-consuming and fussy operation of BP-ANN, PCA-LDA was determined as the suitable method for the alarm on volatile organic compounds in the atmosphere.更多还原