Research of a data preprocessing method for near infrared spectrum of coal

For noise existed in original near infrared spectral data of coal, a data preprocessing method for near infrared spectrum of coal was proposed based on De-SNV and wavelet threshold denoising. Spectrum data processed by Savitzky-Golay smoothing and De-SNV was further processed by default soft threshold denoising method. Then PLS calibration models of moisture, ash and volatile were established. Effect of the method was evaluated by analyzing predicting performance of the models. The experiments show that performance of PLS model based on spectrum data processed by the method is much better than the one based on original spectrum data. The root-mean-square errors of prediction of the three PLS calibration models are decreased to 0.007 07, 0.040 8, 0.008 66 respectively, and the determination coefficients are increased to 0.858 7, 0.743 8, 0.778 5.