小波系数压缩用于柴油十六烷值近红外光谱分析建模

Compression of Wavelet Coefficients Applied in Calibration Model for Diesel Cetane Number by Near-Infrared Spectroscopy(NIRS)

近红外光谱分析建模中存在多变量高维数据处理问题,导致计算量大,不利于过程控制中应用。为此提出利用小波变换压缩近红外光谱数据的算法与准则,并结合柴油十六烷值定量分析研究压缩数据的建模效果。研究表明,经小波方法处理后,变量维数压缩30倍左右,光谱主要信息基本保留,而模型的预测精度和常规预处理方法分析相比有明显提高。光谱数据压缩的同时包含了噪声滤除和基线校正,简化数据处理步骤,有利于NIRS实际应用时提高建模效率。

Regression is a commonly used method for NIRS analysis model dealing with a large numbers of variables resulting in huge calculations and lower efficiency of operations. In this paper, wavelet transform is employed to preprocess individual signals of NIRS,which reduces the number of variables drastically by extracting a few of larger wavelet coefficients according to Some proposed criterion without loss of important information. Partial I.east Squares （PLS） is used to calculate a regression model for cetane number prediction in diesel fuel on the compressed data set with few variables and the denoising and correcting baseline. The prediction results with the new model are improved compared to that of conventional method. The NIRS analysis model can be obtained more straightforwardly and efficiently.