Artificial neural network prediction of NOx emissions from EGR and non‐EGR engines running on soybean biodiesel fuel (B5) during cold idle mode

Abstract

Artificial neural network (ANN) prediction scheme was developed for nitrogen oxides (NOx) emissions in cold idle mode based on the analysis of NOx emissions from on‐road transit buses operating on a blend of biodiesel. The input data necessary for training and testing the proposed ANN scheme was obtained from two different urban transit buses fueled with 5 vol % soybean methyl ester (SME) and 95% of ultra‐low sulfur diesel (ULSD). One bus was equipped with exhaust gas recirculation (EGR) and the other one was not. The reduction of NOx emissions was observed when EGR was implemented. A standard feed forward back‐propagation algorithm was used in this analysis and in building the network structure, whereas Levenberge–Marquardt (LM) learning algorithm was used to predict the NOx emissions. The study was carried out with 70% of total experimental data selected for training the neural network, 15% for the network's validation, and the remaining 15% data were used for testing the performance of the trained network. ANN results showed that the developed ANN model was capable of predicting the NOx emissions of the tested engines with excellent agreement (correlation coefficients: 0.99 < R < 1), while root mean square errors (RMSEs) were 22.1 and 1.7 ppm for non‐EGR and EGR‐equipped engine series, respectively. ANN provided an accurate and simple approach to the analysis of this complex, multivariate problem, where idle NOx emissions from both non‐EGR and EGR engines should be predicted. © 2016 American Institute of Chemical Engineers Environ Prog, 35: 1537–1544, 2016