Research On Modeling And Multi-objective Optimization Of Urea-SCR System For A Diesel Engine

Urea-SCR technology has many advantages,including high NO_X conversion efficiency,good reliability and strong durability and so on.It is an effective and key technique for reducing NO_X emission of diesel engines,one of the major problems of which is how to precisely control the urea injection amount to balance the trade-off relationship between NO_X conversion efficiency and NH₃ slip.Therefore,this thesis presented an ensemble method on modeling and multi-objective optimization of NO_X conversion efficiency and NH₃ slip for Urea-SCR system.Firstly,the engine bench experiment was designed to obtain the original data for modeling,and the pre-processing of the original data was then carried out by means of normalization and principal component analysis to reduce the dimension of the original data matrix and extract the effective information from the original data,to reduce the modeling computational complexity and improve model efficiency.The different prediction models based on the partial least squares method?PLS?,artificial neural network?NN?,support vector machine?SVM?were then established,the optimization methods of different models were also put forward according to the characteristics of different algorithms.Secondly,the model error was analyzed,and cross-validated dataset partitioning method was presented to deal with the problem that the traditional dataset partitioning method is easy to lead to over-fitting model.The K-fold cross-validation with k=7was chosen as the method of data sampling for modeling after comparing the accuracy in training and testing datasets between K-fold cross validation and leave group out cross validation.Genetic algorithm?GA?combined with the exponential sequence grid search method was then used to optimize the parameters of SVM model for improving the model accuracy and generalization ability,with cross-validation root mean square error taken as the fitness function.Statistical indicators,including root mean square error,average absolute percentage error and correlation coefficient,were used to evaluate the accuracy of PLS model,NN model,SVM model and GA-SVM model both on training and testing datasets.The optimal prediction models of upstream NO_X emission,downstream NO_X emission and NH₃slip were built,with GA-SVM taken as the best modeling method.The two objective functions of NO_X conversion efficiency and NH₃ slip were obtained based on these prediction models,and the non-dominated sorting genetic algorithm?NSGA-II?was then implemented to optimize the urea injection amount to maximize the NO_X conversion efficiency while minimizing NH₃ slip under different operating points,and the optimal Pareto set between NO_X conversion efficiency and NH₃ slip and their corresponding urea injection amount were also obtained.Finally,the optimal urea injection amount was obtained under different operating points by adding specific constraints to the optimal Pareto set.This thesis proposed the Urea-SCR system modeling and multi-objective optimization method based on the machine learning and genetic algorithm,and validated its validity and feasibility,which exhibited certain theoretical significance and practical value.