摘要
建立了有限体积法数值仿真、改进的误差反向传递神经网络和遗传算法相结合的铝型材挤压工作带长度优化模型.将型材截面划分单元,由正交试验法得到单元工作带长度值作为网络训练样本的输入值,模型目标值为变形后质点速度均方差.基于有限体积法的数值仿真技术获得样本目标值,模型的全局优化解由遗传算法求得.最后将优化结果应用于工程实际,取得了满意的效果.
An optimization model for designing the die land of aluminum extrusion is presented,which inte- grates numerical simulation with finite volume method,back propagation neural network,genetic algorithm.The area of extrusion section is divided into several elements and the land length values are given as the inputs of network training specimens by using the orthogonal method.The target value of the model is velocity mean- squared error after extrusion.Finite volume method is used in the numerical simulation to get the target value of specimen and the general optimized solution is attained through genetic algorithm.Finally,the optimized results are applied to a practical aluminum extrusion process and satisfactory effect is obtained.
出处
《同济大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2005年第7期937-941,共5页
Journal of Tongji University:Natural Science
基金
上海市启明星跟踪计划资助项目(01QMH1411)
关键词
有限体积法
神经网络
遗传算法
工作带
优化
<Keyword>finite volume method (FVM)
neural network
genetic algorithm
die land
optimization
