实验设计和神经网络法对柴油机性能的优化研究  

Research on Performance Optimization of Diesel Engine Based on DOE and ANN Method

作　　者：李静[1] 姜峰[2] 牛彩云[2] LI Jing;JIANG Feng;NIU Cai-yun(School of Vehicle Engineering,Jinzhong Vocational and Technical College,Shanxi Jinzhong030600,China;School of Mechanical and Transportation,Guangxi University of Science and Technology,Guangxi Liuzhou545006,China)

机构地区：[1]晋中职业技术学院车辆工程学院,山西晋中030600 [2]广西科技大学机械与交通工程学院,广西柳州545006

出　　处：《机械设计与制造》2020年第6期100-104,共5页Machinery Design & Manufacture

基　　金：广西自然科学基金项目(2013GXNSFAA019317);广西高校中青年教师基础能力提升项目(2017KY0357)。

摘　　要：利用GT-Power软件搭建一款3.1L电控增压柴油机仿真模型,采用拉丁超立方采样算法进行实验设计与计算,确立了6个实验因子和4000个实验数目.通过神经网络径向基算法对不同响应变量因子进行建模,最终确定了转速与EGR率两个实验因子对多目标优化影响的贡献度最大.通过建立实验因子和响应变量模型关联,完成了基于模型的多目标遗传优化.优化结果表明:通过优化柴油机扭矩和燃油消耗率,可使柴油机扭矩值最大提升12.3%,且燃油消耗率最大能下降2.6%.GT-Power software is used to build up a simulation model of 3.1L electronically controlled and turbocharged diesel engine.We adopt Latin hyper-cube sampling algorithm to do the design of experiment(DOE)and the calculation,and to set up 6 experimental factors and 4000 experimental items.Model of different responsive variables is built up by the radial basis function(RBF)algorithm of artificial neural network(ANN)to finally determine that two experimental factors(that is,rotational speed and EGR rate)contribute the most to the multi-objective optimization.A connective model between experimental factors and responsive variables is built up to complete the model-based multi-objective optimization.The optimization results show that by optimizing diesel engine torque and fuel consumption rate,the maximum torque increases by 12.3%as the maximum fuel consumption decreases by 2.6%.

关 键 词：实验因子 拉丁超立方采样 实验设计 神经网络 

分 类 号：TH16[机械工程—机械制造及自动化] TK421[动力工程及工程热物理—动力机械及工程]