Multi-objective Optimization Design Of High-speed Train Head Based On Machine Learning

With the increasing speed of high-speed trains,more and more researchers pay attention to the aerodynamic characteristics of high-speed trains in order to save energy and reduce emissions.Among many factors,the head shape of high-speed train has an obvious influence on aerodynamic characteristics such as aerodynamic drag and aerodynamic lift.Therefore,it is very necessary to carry out research on the head shape design of high-speed train.Under the background of the rise of artificial intelligence,machine learning algorithm based on big data has achieved vigorous development.In this paper,two machine learning algorithms,support vector machine(SVM)and BP neural network(BPNN),are used to construct the proxy model of high-speed train aerodynamic characteristics.The multi-objective optimization of aerodynamic drag and aerodynamic lift is realized by using genetic algorithm.The main research work is as follows:Firstly,the aerodynamic characteristics of high-speed train are simulated and analyzed.A simplified three-section model and fluid calculation domain are established.Sculptor uses sculptor to select eight shape design variables in the locomotive part.Then the Fluent software is used to calculate the aerodynamic characteristics of the medium-high speed train in the sample space,and the aerodynamic drag and aerodynamic lift are obtained.Secondly,two machine learning algorithms,SVM and BPNN,are briefly described,and the two algorithms are used to construct the aerodynamic drag and aerodynamic lift proxy models.The influences of kernel function and penalty factor on the accuracy of SVM model and the influences of the number of BPNN hidden layer neurons on the accuracy of the model are studied respectively.For the SVM model,the radial basis function is used as the kernel function,and the accuracy of the model is the highest,and different kernel parameters and penalty factors will have great differences on the accuracy of the model.For BPNN model,different number of neurons in the hidden layer will also lead to different accuracy of the model,but the model accuracy can be maintained at a high level.Thirdly,the parameters in the SVM model are optimized by particle swarm optimization algorithm(PSO),and then the genetic algorithm(GA)is used to optimize the BPNN model parameters.By comparing different indicators of the four models before and after optimization,The accuracy of the BP neural network(GA-BP)model optimized by genetic algorithm and the support vector machine(PSO-SVM)model optimized by particle swarm optimization has been improved in different degrees,and the error has been reduced.Among the four models,GA-BP model has the highest accuracy.Therefore,GA-BP proxy model is used to carry out the following multi-objective research on aerodynamic characteristics of high-speed trains.Finally,a multi-objective optimization design model of aerodynamic drag and aerodynamic lift of high-speed train head was constructed,which was solved by NSGA-Ⅱ,and a reasonable optimal solution was selected from the multi-objective Pareto solution set.The aerodynamic drag was reduced by 4.3%,the lift was reduced by 8.0%,and the aerodynamic performance was effectively improved.