Optimization Of An Atkinson Cycle Gasoline Engine’s Performance Through NSGA Ⅱ Algorithm Using Machine Learning

In response to the global energy crisis and increasingly severe environmental problems,the exhaust emissions of vehicles have obtained more stringent restrictions under the requirements of a series of policies and regulations.This allows engine manufacturers to withstand huge research and development pressure,and the engine’s energy saving and emission reduction is facing greater challenges.During the engine’s research and development,optimization of the control parameters thus enabling an improvement in engine performance is one of the targets of internal combustion engine researchers and engine vendors.This paper constructs a GT-Simulink simulation platform via the coupling of GTPower and Simulink,and NSGA II algorithm is used to drive the simulation platform in MATLAB to achieve automatic optimization of engine control parameters.In addition,in order to further verify the effectiveness of the optimized parameters,this paper establishes the machine learning model based on experimental data,and the optimized control parameters are put into a machine learning model that has been fully trained,and the predicted value obtained by the machine learning model are compared with the simulation value obtained by GT-Power under the corresponding control parameters.Finally,the optimization of the engine based on machine learning model driven by NSGA II is explored.The main research work of this article is as follows:(1)The acquisition of the sample geometric parameters,structural dimensions,etc.,and the test is carried out for the prototype.Based on the acquired data,a high precision one dimensional simulation model is established in GT-power.(2)The ignition advance angle(SA),exhaust gas recirculation rate(EGR),intake variable valve timing(VVT-I)and exhaust gas variable valve timing(VVT-E)are selected as the control parameters to be optimized and be set in the algorithm.At the same time,the optimization range of these four control parameters are given.Thus,the GT-Simulink coupling simulation platform can be driven by NSGA II algorithm to optimize the control parameters of the engine.(3)The machine learning models based on the Support Vector Machine(SVM)and the BP Neural Network are established.And the machine learning models are trained and tested based on experiment scanning data.In addition,the performance of the two machine learning models is evaluated,and the machine learning model that is finally applied to the research is determined.After the machine learning model is fully trained,the optimized control parameters obtained in(2)are input to the machine learning model,thereby obtaining the predicted value under the corresponding control parameters.Under the same control parameters,the simulation value of the GTSimulink coupling platform and the predicted value of the machine learning model are compared and analyzed.The results show that the engine performance obtained by machine learning model are optimized compared to the original engine according to the optimized control parameters.(4)The machine learning model is driven by the NSGA II algorithm to optimize the engine.And 30 sets of optimized control parameters are selected and input to GTPower model for simulation verification.