Prediction of Methanol Content in Biodiesel by Near Infrared Spectroscopy Combined with Partial Least Square Method

The quantitative analysis of methanol in biodiesel has important scientific significance and practical value for its quality monitoring, and the title study was conducted. The 44 biodiesel samples containing different volume fractions of methanol were prepared and their corresponding near infrared spectra were collected. The original spectra obtained were preprocessed using the combination of Savitzky Golay smoothing filtering (SG) and standard normal transformation (SNV), and feature variables were extracted using synergy interval partial least square method (SIPLS). Wavelength ranges of 6 270-6 640 cm⁻¹ and 10 900-11 240 cm⁻¹ were determined for modeling with partial least square method (PLS). It was shown that the determination coefficient R²_cv was 0.999 4, and the root mean square error RMSECV of the SG-SNV-SIPLS-PLS model from leave-one-out cross validation was 0.048 8. The determination coefficient R²_p was 0.999 6, the root mean square error RMSEP of prediction set was 0.056 3, and the number of variables was 207, which were superior to those given by the PLS model and univariate linear regression model.