Determinação do poder calorífico de amostras de gasolina utilizando espectroscopia no infravermelho próximo e regressão multivariada

Abstract

The aim this study was quantify the calorific power of 111 gasoline samples available at filling stations using near infrared spectroscopy in conjunction with the multivariate regression. The calorific power value of the fuels was determined using an adiabatic bomb calorimeter (norm ASTM D 4.809). For the construction of multivariate regression models were used 2/3 of the samples for calibration and the remainder to prediction, using the interval partial least squares (iPLS) and synergy interval partial least square (siPLS) algorithms. In the best iPLS model was selected the spectral range from 5561 to 6650 cm-1, obtaining RMSEP of 102 g cal-1 and showing a correlation coefficient (r) of 0.8218 and 0.71% to calibration errors and 0.47% for prediction errors. The siPLS model divided into 32 intervals and grouped into three intervals was the highlighted model, which selected the region below 6000 cm-1 and above 6500 cm-1 with, presenting values of RMSECV of 89.8 cal g-1 and RMSEP of 96.7 cal g-1, and correlation coefficients for the cross-validation and prediction of 0.7834 and 0.7293, respectively. The methodology proposed in this work is efficient, with prediction errors lower than 1%, being a clean alternative, fast, safe and practical.