Comparison of extreme learning machine models for gasoline octane number forecasting by near-infrared spectra analysis

Abstract

The octane number is often used in the evaluation of the gasoline quality due to its antiknock property. The traditional ASTM-CFR internal combustion engine method costs an enormous amount of expenditure and time to determine its level of each product upgrade. The extreme learning machine (ELM) and its improved models are a new type of single-hidden layer feed-forward neural network (SLFN). This structure has advantages over the traditional neural network technology in regression tasks. Based on the response relationship between the near-infrared spectra and the octane number, 60 groups of collected gasoline data are divided into training and test sets. The ELM, online sequential extreme learning machine (OS-ELM), and self-adaptive evolutionary extreme learning machine (SaDE-ELM) models are applied to the prediction task of the octane number. The statistical parameters RMSE, CORR, and R2 and the execution time are used as the performance comparison criteria for the three models. The simulation results show that the OS-ELM and the SaDE-ELM improve the prediction accuracy, the generalization ability, the stability of the ELM in different levels.