Regression-based models for prediction of oxides of nitrogen in diesel exhaust with electric discharge-based treatment

Abstract

A prior prediction of oxides of nitrogen, i.e., NOX (sum of NO and NO2), in diesel exhaust while treating with electric discharge-based nonthermal plasma (NTP) technique, would assist the researchers in planning the resources required for the treatment. In this present study, the performance of different regression-based models, i.e., linear, support vector regression and Gaussian process regression (GPR), has been analyzed for predicting the NOX concentrations based on the values of five dominating parameters of the NTP treatment. Experiments have been conducted and collected a dataset of 4032 number of input–output pairs to be used for training and testing of the regression models. The performances of these models have been assessed while testing them for the unseen set of data. A comparison of root-mean-square error (RMSE) has been made, where Matern 3/2 type of GPR model has been found to be the best among all the considered models with an RMSE of 1.86 ppm for a test data of 1210 sets. The model is shown to perform consistently well even when the test data are increased to 50% of total data. Regression analysis shows that the NOX can be predicted with very good accuracy using the Matern 3/2 type of GPR model.