Abstract: A method to calibrate multiple Wiebe's heat release rate model was proposed by building predictive neural network model of combustion based on Bayesian regularization to solve the problem that Wiebe's combustion model needed to go through a lot of point-by-point parameter tuning and its universality and predictability was poor. ModeFRONTIER was used to pre-calibrate part points of the multi-Wiebe combustion model, which provided data for the establishment of combustion prediction model. The sensitivity analysis between operating condition boundary parameters and model calibration parameters was carried out, and the relationship between them was established by using the neural network based on Bayesian regularization, which endowed the multiple Wiebe combustion models with prediction and reduced the calibration workload of the combustion model. The results show that the average accuracy of the combustion prediction model is 93.2%, and the prediction accuracy of some operating points is more than 97%, manifesting that the neural network combustion prediction model has high model accuracy and model generalization ability.

Key words:  diesel combustion, multiple Wiebe combustion model, Bayesian regularization algorithm, neural network, predictive mode