LAUSR

Abstract Prediction of the cetane numbers (CN) of hydrocarbons and oxygenates was considered using the Adaptive-network-based fuzzy inference system (ANFIS) based on the quantitative structure-property relationship (QSPR) approach. In order to construct a general model, descriptors of pure compounds were selected regarding their applicability to all kinds of compounds and ability to provide an accurate knowledge about combustion and evaporation process of fuels inside compression-ignition engines. The predictive capability of optimized ANFIS models based on the Back Propagation (BP) algorithm, Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Differential Evolution (DE), and Ant Colony Optimization (ACO) have been evaluated for estimating CN. In this regard, an extensive databank containing almost all available single-compound CN data of 204 hydrocarbons and 292 oxygenate was gathered from the literature. Obtained results indicated that the PSO-ANFIS approach has the most satisfactory prediction of all considered approaches. Furthermore, an outlier analysis was applied to enhance the model accuracy and detect suspected data points. The statistical coefficients of R-squared (R 2 ), Mean Squared Error (MSE) and Mean Relative Error (MRE%) were obtained for testing data set as 11.49 & 2.95, 12.53 & 3.28, 25.7 & 7.77, 34.35 & 9.92 & 47.43 & 14.1 for the PSO-ANFIS, GA-ANFIS, ACO-ANFIS, BP-ANFIS, and DE-ANFIS models, respectively. Accordingly, the PSO-ANFIS strategy appeared to be a great tool for estimating CN.