PLS Regression on Coal Infrared Spectrum with Wavelet Pre-Processing

Study on multivariate calibration for infrared spectrum of coal was presented. The discrete wavelet transformation as pre-processing tool was carried out to decompose the infrared spectrum and compress the data set. The compressed data regression model was applied to simultaneous multi-component determination for coal contents. Compression performance with several wavelet functions at different resolution scales was studied, and prediction ability of the compressed regression model was investigated. Numerical experiment results show that the wavelet transform performs an effective compression preprocessing technique in multivariate calibration and enhances the ability in characteristic extraction of coal infrared spectrum. Using the compressed data regression model, the reconstructing results are almost identical compared to the original spectrum, and the original size of the data set has been reduced to about 5% while the computational time needed decreases significantly.