Construction of multi-stage heat release prediction model for HPDI dual-fuel engine based on efficient machine learning algorithm

Abstract

The high-pressure direct injection (HPDI) diesel/natural gas dual-fuel engine holds significant research importance due to its potential to increase thermal efficiency while reducing pollutant emissions. However, its intricate and variable combustion process presents challenges for modeling and simulation efforts. In the current study, a scientific framework for modeling multi-stage heat release for HPDI engine based on machine learning algorithms is proposed to address such challenge. Firstly, the active subspace sensitivity analysis method and the triple Wiebe function fitting method were employed to ascertain the principal engine operating condition parameters as independent variables and the Wiebe function parameters as dependent variables of the multi-stage heat release prediction model, respectively. Subsequently, the support vector regression (SVR) algorithm, which calculated the heat release rate on the target testing set that is closest to the experimental value, was selected among multiple machine learning algorithms to construct the multi-stage heat release prediction model. Meanwhile, this model was highlighted with integrating GridSearchCV method to optimize the hyperparameters, thus achieving a better prediction accuracy of R² greater than 0.95 on the testing set. Finally, in order to enhance the model’s generalization ability, the conditional tabular generative adversarial network (CTGAN) was employed to expand the original limited dataset in training set. As a consequence, the new multi-stage heat release prediction model was calibrated against the combustion parameters CA10 and CA50 for available experimental conditions. It is demonstrated that the proposed new modeling framework exhibited advantages in terms of high prediction accuracy, strong generalization ability, and low amount of experimental data required for model calibration.