Moving window correlation coefficient differences partial least squares (MWCC-DPLS) quantitative calibration method based on spectral differences between calibration samples: Application to the fast determination of gasoline octane number with near-infrared spectroscopy

A new quantitative calibration method based on spectral differences between calibration samples, named “moving window correlation coefficient differences partial least squares (MWCC-DPLS)”, was proposed for the fast determination of gasoline research octane number (RON) with near infrared (NIR) spectroscopy. Such method takes full advantage of the exact spectral searching performance of MWCC and the proper compensation performance of differences PLS (DPLS), which can possibly provide enhanced prediction accuracy. The prediction performance of MWCC-DPLS was evaluated by using a set of exclusive gasoline dataset from a refinery affiliated to SINOPEC, China. The results indicated that the proposed MWCC-DPLS method significantly improved the prediction results of traditional MWCC method for research octane number (RON) of unknown gasoline samples that are not included in the calibration set. More importantly, it attained excellent improvement of about 34%, 29%, 33%, 26%, and 30% on the average value of MAE and RMSE in testing set as compared with PLS, KPLS, SVR, GPR, and ANN calibration methods, respectively. As can be seen, the proposed MWCC-DPLS possesses the outstanding advantage of prediction accuracy in employing it to address the calibration problem of «NIR spectrum-gasoline RON» non-linear analysis system and is a promising method of multivariate quantitative calibration.