Feasibility study on prediction of gasoline octane number using NIR spectroscopy combined with manifold learning and neural network

Octane number is an anti-knock index of fuel gasoline, which has an important impact on the service life of engine components and the safety of vehicles. Therefore, it is a basic work involving safety to predict the gasoline octane number accurately. This work was aimed to predict the octane number of near infrared (NIR) spectroscopy by combining dimension reduction algorithm with neural network. Covariance matrix estimation (CME), known as a mathematical statistic tool, was applied to estimating the intrinsic dimensions of octane spectrum dataset. Landmark-Isometric feature mapping (L-Isomap), as a novel manifold learning algorithm, was used for dimensionality reduction of spectral data. A new method, beetle antennae search optimization BP neural network (BAS-BP), was proposed to realize the prediction of octane number. In order to verify the performance of CME-L-Isomap-BAS-BP model presented in this paper, it is compared with other models. The results showed that when CME-L-Isomap was combined with BAS-BP, the average recovery rate (AR), mean square error (MSE), mean absolute percentage error (MAPE), correlation coefficient (R) and running time were superior than other models. The satisfying results demonstrated that the CME-L-Isomap-BAS-BP model is more suitable for prediction of gasoline octane number. (c) 2019 Elsevier B.V. All rights reserved.