Performance Prediction And Optimization Of Gdi Turbocharged Engine Based On Grnn And SVM

GDI(Gasoline Direct Injection)turbocharged engine is a complex non-linear system.The effect of engine performance prediction and optimization based on single factor or traditional methods are not so good.So in this paper,firstly,the influence of multiparameters on engine performance is studied and analyzed by Taguchi method.And then,the generalized regression neural network(GRNN)and support vector machine(SVM)are used to model the engine performance with multi-parameters.Finally,NSGA-Ⅱ(Nondominated Sorting Genetic Algorithms-II)is used to optimize engine performance.The results show,Taguchi method can effectively analyze the influence of multi-parameters on engine performance,and the engine performance can be optimized by combining machine learning and intelligent optimization algorithms,which can quickly optimize the engine performance and greatly reduce the experimental cost.This paper is based on the key technology innovation project of key industries in Chongqing,“Research and Application of Key Technology for Supercharger Design and Control”(Project No.: CSTC2015ZDCY-ZTZX60014).The main research contents are as follows:Firstly,based on a certain engine test bench data and engine physical size,the engine professional simulation software GT-POWER is used to establish the engine full load characteristic simulation model.The simulation results show that the relative error between the simulation model and the experimental value is less than 5%.The established simulation model can be used to study the engine full load characteristics.Then,based on the engine simulation model,the signal-to-noise ratio analysis and variance analysis in Taguchi method are used to analyze the engine IVVT(Intake Variable Valve Timing),EVVT(Exhaust Variable Valve Timing),Start of Injection(SOI),CA50(Crank Angle 50),and the length of engine intake and exhaust manifold and diameter on engine torque,BSFC(Brake Specific Fuel Consumption)and exhaust temperature.The results show that Taguchi method can effectively analyze the influence of multiple inputs on output,and the contributions of IVVT,EVVT,injection time and CA50 to engine performance parameters are greater.Further,on the basis of the above research on engine parameters,taking IVVT,EVVT,injection timing and CA50 as inputs,the generalized regression neural network and support vector machine are used to predict engine performance in small samples.The results show that the generalized regression neural network and support vector machine can predict engine torque,BSFC and exhaust temperature well with small samples,and the prediction accuracy and stability of support vector machine are slightly higher than that of the generalized regression neural network.Finally,combining support vector machine and NSGA-Ⅱ algorithm are used to optimize the engine full load characteristics.The experimental results show that the machine learning algorithm combined with the intelligent optimization algorithm optimizes the engine performance,can quickly optimize the engine performance,and has high accuracy,which provides an advantageous tool for engine design and development.