Research On Intelligent Calibration Method Of Diesel Engine Based On Bayesian Regularization Combustion Model

Virtual calibration technology has been widely valued for its potential to reduce resource consumption and shorten development cycle in the process of diesel engine calibration.Aiming at the complicated problem of diesel engine combustion model calibration,an intelligent calibration method of diesel engine based on Bayesian regularization combustion model is proposed in this paper to solve the problem caused by complicated combustion model calibration,control MAP calibration and optimization of multi-objective,multi-parameter,multi-working condition and strong coupling.Firstly,aiming at the study of intelligent calibration method of diesel engine,a GT-SUITE model of four-cylinder high-pressure common rail diesel engine was established to provide experimental data for the pre-calibration of combustion model and the establishment of combustion prediction model,and provide a comparative verification platform for the subsequent control MAP calibration optimization algorithm.In view of the changeable fuel injection strategy and real-time requirements of the model,a multi-wiebe combustion exothermic rate model is built to develop a method for establishing the combustion prediction model.Secondly,in view of the diesel engine more weber combustion model calibration of weber mutual coupling,many parameters,point by point calibration workload big,lack of predictive models,this paper adopts based on the Bayesian regularization Bayesian regularization algorithm of feedforward neural networks combustion prediction model building method,studies the influence of the relationship between the working condition of boundary parameters and model parameters,solves the model overfitting problem.Compared with neural network model trained by LevenbergMarquardt algorithm and scaled conjugate gradient algorithm and model fitted by multiple linear fitting method.The experimental analysis shows that the prediction model has good adaptability and predictability.The average calibration accuracy of the working condition points has reached 93.19%,and the prediction accuracy of partial working condition points have reached more than 97%.Finally,in consideration of the control MAP calibration optimization of multiobjective,multi-parameter and multi-condition,the strong coupling characteristic,low efficiency,poor effect and strong experiences dependence of manual traversal adjustment and optimization,this paper designs the multi-objective optimization control MAP calibration platform based on the non-dominated sorting genetic algorithm(NSGA Ⅱ),realize the optimization of multi-condition,multi-parameter calibration work.Compared with the manual traversal optimization method,the steadystate condition calibration of electronically controlled diesel engine realized by this platform can improve the economy by 5.58% on average,reduce the original NOx emission by 59.47% on average,save about 40% of the time,reduce the participation of engineers in the calibration process,and reduce the labor cost.