中文摘要：

通过扫描不同NH3-N含量污水的近红外光谱,建立了水样中NH3-N的定量分析模型。考虑到全谱高维数据建模较大的计算负担,采用连续投影算法（SPA）对水样光谱全谱进行特征波长筛选,将筛选后的特征变量采用最小二乘支持向量机（LS-SVM）进行建模;所建立的SPA和LS-SVM分析模型对污水中NH3-N分析的预测均方根误差为3.210 8,相关系数为0.984 4,相对分析误差5.681 2;与全谱LS-SVM模型和全谱部分最小二乘（PLS）模型相比,此处的建模方法将全谱模型的512维数据压缩为28维特征光谱数据（计算量占全谱的5.47%）,但模型分析精度与全谱LS-SVM模型相近,且高于全谱PLS模型;该方法对实现水样NH3-N的快速检测以及低维度变量建模具有指导意义。

英文摘要：

By scanning near infrared spectrum （NIRS） of wastewater with different NH3-N contents, a quantitative analysis model of NH3-N in wastewater was proposed. Taking into consideration the computational burden of full spectral data, successive projection algorithm （SPA） was employed to choose the feature spectral data.The selected feature variables were then used for modeling based on least squares support vector machine （ LS-SVM） .The proposed model resulted in RMSEP = 3.210 8, correlation coefficient = 0.984 4, and RPD = 5.681 2. Compared to the full-spectral LS-SVM model and partial least squares （PLS） model, the proposed SPA and LS- SVM model compressed 512-dimentional full-spectral data into 28-dimentional ones whose computation burden is 5.47% of the full-spectral data.Nevertheless,the precision of the proposed model is similar to the full-spectral LS- SVM model and is better than the full-spectral PLS model.The proposed method has good guidance significances for rapid measurement of NH3-N and low-dimensional variable modeling.