基于可见-近红外光谱技术与BP-ANN算法的污水类型鉴定

Identification of the types of waste water based on visible/near-infrared spectroscopy and BP-ANN algorithm

提出了一种基于可见-近红外光谱技术与BP人工神经网络（BP-ANN）算法快速进行污水类型鉴定的新方法。以FieldSpec R3地物光谱仪采集了4种污水样品的光谱数据,共168份,随机将其分成校正集（132份）和检验集（36份）。分别采取全波段（400-2450 nm）与择取波段（400-1800 nm）两种方法建立模型进行分析。光谱经S.Golay平滑和标准归一化（SNV）处理后,以主成分分析法（PCA）降维。将降维所得的前9个主成分数据作为BP-ANN的输入变量,污水类型作为输出变量,建立3层BP-ANN鉴别模型。利用36个未知样对模型进行检验。结果表明：两类模型预测准确率均高达100%,且择取波段模型比全波段模型具有更高的预测精度。说明利用可见-近红外技术结合BP-ANN算法进行污水类型的快速、无污染鉴定是可行的,且波段筛选是优化模型的有效方法之一。

A rapidly and pollution-free method was developed to identify the types of waste water by visible/near-infrared spectroscopy and back-propagation artificial neural network （BP-ANN） algorithm. The spectra data of the total 168 samples were obtained by a FieldSpec R 3 spectrometer. All the samples were divided randomly into two groups, one with the 132 samples used as the calibrated set,and the other with the 36 samples as the validated set,and subsequently were analyzed with the whole wave band（400 - 2450 nm） and the selection wave band（400 - 1800 nm） models, respectively. The spectra data were pretreated by the methods of S. Golay Smoothing and Standard Normal Variable （SNV） ,and the pretreated spectra data were analyzed with Principal Component Analysis （PCA）. The anterior 9 principal components computed by PCA were used as the input variables of BP-ANN model which included one hidden layer, while the values of the types of waste water used as the output variables, and consequently the three layers BP-ANN identification model was built. The 36 unknown samples in the validated set were predicted by the ANN-BP model. The results showed that the recognition rate was 100% in such both models, and the accuracy of selection wave band model was higher than that of the whole wave band model. We suggested that it was feasible to discriminate the types of waste water used by visible / near-infrared spectroscopy and BP-ANN algorithm as a rapid and pollution-free way, and the wave band selection was a validated way to improve the precision of the identification model.