人工神经网络法对多组分大气污染物的同时监测

Simultaneous Determination of Multi-component Air Pollution by Artificial Neural Network and FTIR Spectroscopy

用 18 7 8的反向传播人工神经网络 (BP ANN)模型 ,对FTIR光谱图存在着严重混叠干扰的八种有毒易挥发有机化合物 (VOCs)组成的大气污染物进行了同时定量测定 ,得到了各污染物的浓度。所测定的八种VOCs为苯乙酮 ,苯酚 ,三氯甲苯 ,1,3丁二烯 ,氯苯 ,甲醇 ,三氯代乙烷和二氯甲烷。用标准预测误差 (%SEP) ,平均预测误差 (MPE)和平均相对误差 (MRE)来评价其预测能力。结果表明 ,本方法对多组分大气污染物定量分析 ,能够得到较为满意的结果。

An 18-7-8 artificial neural network (ANN) was applied to the simultaneous determination of air pollutant composed of eight air toxic volatile organic compounds (VOCS) whose FTIR spectra overlap each other seriously. The eight VOCs were acetophenone, phenol, benzotrichloride, 1, 3-butadiene, chlorobenzene, methanol, methyl chloroform and methylene chloride. They were mixed together with very low concentrations. The standard error of prediction (%SEP), the mean prediction error (MPE) and the mean relative error (MRE) were utilized to evaluate the prediction ability of the BP-ANN. Results showed that the BP-ANN can be used to obtain satisfactory results when dealing with multi-component analysis of air pollution.