迭代最小二乘法气体光谱自动水汽差减算法

傅里叶红外光谱法具有测量速度快、信噪比高、检测范围广等优势，在针对污染源废气排放的快速检测及长时间在线监测中具有巨大的发展潜力。水汽是红外光谱污染气体检测中的主要干扰物，影响NO_X，SO₂等重要污染物的检测，差减水汽背景谱消除光谱中水汽干扰可提高这些污染物的检测精度，具有重要意义。气体光谱中水汽吸收峰由于受到水分子团簇、仪器线型函数等影响，通过数值方法对其计算的误差较大；为此，水汽背景谱一般需采用同一台光谱仪实测获得。主要有两种方法： 第一种是通过反复调节水汽/氮气混合气中的水汽浓度，使水汽背景谱中的水汽吸收峰与污染气体光谱中水汽吸收峰相同，此方法耗时较长，且受环境条件制约很难在现场检测中使用；第二种方法是预先测量不同浓度的水汽光谱，在检测污染气体时选取两幅与污染气体光谱中水汽吸收峰最为接近且将其夹在中间的水汽光谱作为参考谱，使用这两幅参考谱线性拟合与污染气体光谱中水汽吸收峰相同的水汽背景谱，此方法可获得高度近似的水汽背景谱，但当前缺乏相关自动算法妨碍了其在快速自动消除水汽干扰方面的应用。为此，提出一种选取水汽参考谱及拟合水汽背景谱的自动算法，用于自动差减消除水汽干扰。在参考谱选取中，使用污染气体光谱依次减去浓度由低至高的水汽光谱，依据差减后光谱中水汽吸收峰所在波数的吸光度正负性来选取参考谱。在水汽背景谱计算中，基于迭代最小二乘法逐步剔除光谱中受污染物吸收峰干扰的波数，采用剩余波数上的数据拟合水汽背景谱，使其与污染气体光谱中水汽吸收峰相一致。使用水汽背景谱对污染气体光谱进行差减即可消除污染气体光谱中的水汽干扰。对含有NO₂的污染气体光谱进行了差减消除水汽干扰实验，结果表明所提出的自动算法可快速准确消除水汽干扰；NO₂在消除水汽干扰后可由其位于1 629 cm-1的强吸收峰检测，相比消除水汽干扰前使用不受水汽干扰的位于2 917 cm-1的弱吸收峰检测，其检出限得到了大幅提高。

Automatic Algorithm for Water Vapor Compensation of Gas Spectra Through Iterative Least Square Method

Fourier transform infrared spectroscopy (FTIR) enables the simultaneous measurement of multiple pollutants at highspeed and is thus a powerful technique that can be used for the rapid detection and online monitoring of air pollutants. Air pollution monitoring through FTIR is mainly affected by water vapor, which interferes with the measurement of pollutants that share the same spectral region with water vapor, particularly NO_X and SO₂. One of the approaches for increasing the accuracy of a method for measuring pollutants is removing water vapor interference from a sample spectrum by subtractingthe background water vapor spectrum. This method is greatly useful in analyzing spectra with water vapor interference. Water vapor spectra in different concentrations do not ideally follow the Beer-Lambert’s law because of several nonlinear effects, such as water molecule clusters and instrument line shape function. Thus, a numerically calculated water vapor spectrum by Beer-Lambert’s law presents substantial error. Therefore, the background water vapor spectrum is usually directly measured by the same FTIR instrument that measures the sample spectrum at the same water vapor concentration as that of a sample spectrum. The background water vapor spectrum can be measured by two methods. One is adjusting water vapor concentration in a water vapor/nitrogen mixture toresemble a sample spectrum. This method is time consuming and extremely difficult to use in the field. The other method involves premeasuring multiple water vapor reference spectra with different concentrations and fitting. After sample spectrum measurement, two reference spectra are selected from premeasured water vapor spectra, and a background water vapor spectrum is linearly fitted by these reference spectra. The fitting method can obtain a highly approximated water vapor background spectrum when the water vapor concentrations in the reference spectra are extremely close to the sample spectrum’s water vapor concentrations and the water vapor concentration of sample spectrum is in the middle of two reference spectra. Currently, the fitting method cannot be applied in the rapid automatic elimination of water vapor interference due to lack of automatic algorithm. Thus, this study proposes an automatic algorithm, which includes reference spectrum selection and background spectrum fitting, for the fitting method. In the reference spectrum selection, the sample spectrum deducts the premeasured water vapor spectra from low to high concentrations. Two reference spectra are selected by the criteria according to the number of wavenumbers of negative absorbance in the subtracted spectra. The background water vapor spectrum is fitted through the iterative least square method, which gradually deletes the wavenumbers that are interfered by pollutants, and a background water vapor spectrum, which has an absorption feature that is consistently identical to that of water vapor in the sample spectrum, is linearly fitted by the remaining wavenumbers. The water vapor interference in the sample spectrum is eliminated by subtracting its background water vapor spectrum. In this study, we automatically remove water vapor interference in an air spectrum that contains NO₂. Results show that the proposed algorithm accurately remove water vapor interference. After the elimination of water vapor interference, NO₂ can be detected by its absorption peak located at 1 629 cm-1. The detection limit of NO₂ remarkably improves when compared with detecting by its weaker absorption peak located at 2 917 cm-1 that is not interfered by water vapor.