Comparison of machine learning algorithms for predicting diesel/biodiesel/iso-pentanol blend engine performance and emissions

In this study, machine learning techniques, namely artificial neural network (ANN), support vector machine (SVM), and extreme gradient boosting (XGBoost), were used to comprehensively evaluate engine performance and exhaust emissions for different fuel blends. To obtain valuable insights on optimizing engine performance and emissions for alternative fuel blends and thus contribute to the advancement of knowledge in this field, we focused on iso-pentanol ratios while maintaining the biodiesel ratios constant. The maximum brake thermal efficiency (BTE) values for the diesel (30.13 %), D(85)B(10)P(5) (29.92 %), D(80)B(10)P(10) (29.89 %), and D(70)B(10)P(20) (29.79 %) blends were achieved at 1600 rpm. At 1600 rpm, the brake-specific fuel consumption (BSFC) values for the diesel, D8(5)B(10)P(5), D(80)B(10)P(10), and D(70)B(10)P(20) blends were 189.93, 200.93, 202.93, and 203.95 g kWh(−1), respectively. In engine performance prediction, the ANN model exhibited superior performance, yielding regression coefficient (R(2)), root mean square error, and mean absolute error values of 0.984, 0.411 %, and 0.112 %, respectively, in BTE prediction, and 0.958 %, 6.9 %, and 2.95 %, respectively, in BSFC prediction. In exhaust gas temperature prediction, the SVM model exhibited the best performance, yielding an R(2) value of 0.981. Although all models successfully predicted NOx emissions, the ANN model exhibited the best performance, achieving an R(2) value of 0.959. In CO(2) and hydrocarbon estimation, the XGBoost model exhibited the best performance, yielding R(2) values of 0.956 and 0.973, respectively. Therefore, the ANN model can be used to accurately predict engine performance, and the XGBoost model can be used to accurately predict emission parameters.