研究目的
Investigating the prediction of research octane number (RON) and motor octane number (MON) of hydrocarbon mixtures and gasoline-ethanol blends based on infrared spectroscopy data of pure components.
研究成果
The study demonstrated that infrared spectroscopy data can be used to predict the octane numbers of hydrocarbon blends and gasoline-ethanol blends with high accuracy. Dimensional reduction techniques were effective in extracting important features from the spectra, and nonlinear regression methods, particularly ANN, outperformed linear methods in capturing the nonlinear blending behavior of octane numbers. The mean absolute error from ANN was within the experimental uncertainty of octane testing, making it a promising tool for chemometric studies.
研究不足
The study assumes that the composite spectra of a mixture can be calculated as the molar sum of the spectra of the pure components, which is a reasonable assumption for gas-phase spectra but may not hold for liquid spectra where excess absorbance has been reported. The model's performance is limited by the availability of spectra for pure components in the PNNL database.
1：Experimental Design and Method Selection:
The study utilized infrared spectroscopy data of pure components to generate spectra for hydrocarbon blends and FACE gasoline blends. Dimensional reduction techniques (PCA and SVD) were applied to extract important features from the spectra. Nonlinear regression methods, including ANN, SVM, and PLSR, were used to predict octane numbers.
2：Sample Selection and Data Sources:
Infrared spectra for 61 pure hydrocarbon species were collected from the Pacific Northwest National Laboratory (PNNL) IR spectroscopy database. The spectra of 148 hydrocarbon blends and 38 FACE gasoline blends were calculated based on their compositions.
3：List of Experimental Equipment and Materials:
The spectra were measured by vaporizing the liquid hydrocarbons in a Bruker-66V FTIR at temperatures of 5 C, 25 C, and 50 C.
4：Experimental Procedures and Operational Workflow:
The spectra of blends were calculated by averaging the spectra of their pure components on a molar basis. Dimensional reduction was applied to the spectra to extract scores, which were then used as input for nonlinear regression models.
5：Data Analysis Methods:
The performance of the models was quantified using error metrics such as RMSE, MAE, MPE, MAX, and R2. A 10-fold cross-validation was performed on the training set.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
