Evaluation of machine learning algorithms on hydrogen boosted homogeneous charge compression ignition engine operation for performance and emission prediction

• Published in: Process safety and environmental protection Vol. 195; p. 106756
• Main Authors: Sathishkumar, S., Ibrahim, M. Mohamed
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.03.2025
• Subjects: Gaussian process regression; Hydrogen homogeneous charge compression ignition operation; K-fold cross-validation; Machine learning; Neural Network; Gaussian process regression; Hydrogen homogeneous charge compression ignition operation; Neural Network; K-fold cross-validation; Machine learning
• ISSN: 0957-5820
• DOI: 10.1016/j.psep.2025.01.010

The present work focuses on utilizing machine learning regression algorithms to estimate the performance and emission characteristics in the hydrogen diesel homogeneous charge compression ignition (HDHCCI) engine. The dataset includes three input parameters namely hydrogen energy share (HES), equivalence ratio and injection timing, five output parameters like brake thermal efficiency (BTE), nitrogen oxides (NOx), smoke, hydrocarbon (HC), carbon monoxide (CO). Using the comprehensive dataset, a total of 26 machine learning algorithms were trained and tested. The efficient machine learning model was identified by synthesized evaluation of mean square error (MSE), root mean square error (RMSE), R-squared (R2) and mean absolute error (MAE) values. Among the algorithms considered, Matern 5/2 GPR, Wide Neural Network and Fine Tree algorithm were excellent for predicting the BTE with the R2 value of 0.9999, 0.9985, and 0.9961. Matern 5/2 GPR and Tri layered Neural Network were more accurate for predicting the CO emissions with the R2 values of 0.9999 and 0.9985. Matern 5/2 GPR, Bilayered Neural Network and Fine Tree were exhibited well for forecasting smoke emissions with the R2 value of 0.9999, 0.9993 and 0.9961. Due to the complexity of the estimation of HC emissions, most of the models were underperformed excluding GPR. [Display omitted]