Abstract
A new data-driven approach is developed in order to provide a detailed analysis of fuel quality. Our approach is constructed by combining the recent development of applied mathematical statistics to high-resolution mass spectrometry. Precisely, from the modern mathematical statistics, we use new models, recently introduced, to fit a big data sample collected by the Near-Infrared Reflectance (NIR) spectroscopy procedure. Such a method allows to provide exhaustive information about the chemico-physical properties of diesel fuel such as Boiling Point, the Cetane Number, the density, the total aromatics and the viscosity. The big-data models used to conduct this fuel-quality analysis are the classical regression, the local linear regression and the relative regression. We show that the used models improve the accuracy more than the standard models, such as the Principal Component Regression (PCR) or the Partial Least Squares Regression (PLS). Moreover, the main features of the conduct data-driven approach are the possibility to make accurate, non-destructive, fast and interactive tools that allow real-time analysis of the fuel quality. Such fast analysis allows to provide a portable NIR spectrometry that helps to control the diesel fuel quality in both production and transportation which permit us to simplify significantly the cost and the time-testing.