Autoignition Temperature Prediction Using an Artificial Neural Network with Particle Swarm Optimization

Abstract

The autoignition temperatures of organic compounds were estimated using a hybrid method that includes a simple group contribution method (GCM) implemented in an artificial neural network (ANN) replacing a standard back-propagation algorithm with particle swarm optimization (PSO). A data set of 250 compounds was used for training the network. The optimal condition of the network was obtained by adjusting various parameters by trial-and-error. The capabilities of the designed network were tested in the prediction of the autoignition temperature of 93 compounds not considered during the training step. The proposed model is shown to be more accurate than those of other published works. The results show that the proposed GCM + ANN + PSO method represent an excellent alternative for the estimation of this property with acceptable accuracy (AARD = 1.7 %; AAE = 10 K).

Optimum Fuel for Spark Ignition Engines from Lignin Pyrolysis Oil

“…As a GCM to predict the RON was not available, another approach was employed. Using experimental data, − the following correlation for the RON based on the autoignition temperature ( T auto ), − the number of hydrogen atoms ( y ), and the normal boiling point ( T boil ) , was developed: with a residual standard error of 11.2, an R 2 of 0.8251, and an F-statistic of 141.5.…”

“…As it can be seen, these isomers differ by the position of the substituents in the ring.As a GCM to predict the RON was not available, another approach was employed. Using experimental data[31][32][33][34][35][36][37][38][39] , the following correlation for the RON based on the autoignition temperature (T auto )[40][41][42] , the number of hydrogen atoms (y) and the normal boiling point (T boil )40,43 was developed..36 − 5.94 × 10 14 y 4.36 T 7.45 boil with a residual standard error of 11.2, an R 2 of 0.8251 and an F-statistic of 141.5.…”

Autoignition temperature: comprehensive data analysis and predictive models

“…Since statistical parameters characterizing predictive ability of models largely depends on the test set composition, the real predictive performance of the reported in the literature models is still questionnable. Particularly serious doubts concern the models for which published estimates of prediction errors are considerably lower than experimental measurement errors (19). These shortcomings unfortunately concern many of models published in the last decade.…”

“…In 2011, Lazzus built a model for AIT using a combination of SGC with ANN trained using the PSO (Particle Swarm Optimization) algorithm on a dataset of 250 compounds, and the average absolute error estimated on a test set containing 93 compounds was reported to be 10.5 o C (with R 2 =0.99), while the same measure for ANN trained using the standard backpropagation algorithm appeared to be 45.3 o C (19). Such a huge difference in the quality estimates of the models built using two neural network training algorithms, as well as a surprisingly low prediction error (three times lower than the average measurement error), suggests the presence of strong overfitting (overtraining for ANNs).…”

Artificial Neural Network-Group Contribution Method for Predicting Standard Enthalpy of Formation in the Solid State: C–H, C–H–O, C–H–N, and C–H–N–O Compounds

“…For group contribution methods which are popular and widely used to predict thermodynamic properties of organic compounds [21][22][23][24], there are few models that have been proposed for the prediction of f H • solid in the solid state. The best known and most widely used of these are the methods developed by Domalski and Hearing [19], Cohen [10], and Salmon and Dalmazzone [7,8].…”