摘要
基于遗传算法对某型涡扇发动机全状态实时NARX神经网络模型结构参数进行了优化,通过采用二进制编码形式,将输入输出延迟信息转换成两个输入状态,节省了遗传算法求解空间。而后,对优选后的模型输出结果进行了讨论。采用优选算法获得的模型结构在包含更多数据样本信息的同时,也能够保持神经网络结构的简便性,在高压转速、低压转速的模型优选上效果较为显著。而在涡轮后温度的模型结构优选上,由于数据样本量最大延迟的限制,未能得到可靠的针对涡轮后温度的优选结构,后续工作可对此进行完善。本文采用的方法可推广至神经网络模型其他经验确定、未定参数的优选中,同时还可基于遗传算法对建模样本点进行优选,以获得最小数目的试验样本。
In this paper,the structural parameters of real-time NARX neural network model of a turbofan engine are optimized based on genetic algorithm.The binary coding is used to convert the input and output delay information into two input states,which saves the solving space of genetic algorithm.The model structure obtained by using the optimization algorithm can contain more information of data samples,and can also maintain the simplicity of neural network structure,which has significant effect on high speed and low speed model optimization.Because of the limitation of data sample maximum delay,it is failed to obtain the optimal structure of temperature model of turbine outlet.The problem should be improved in the following works.The method used in the paper can be extended to other empirical determination and undetermined parameter optimization of neural network model.At the same time,based on the genetic algorithm,the model sample points can be optimized to obtain the minimum number of test samples.
作者
王霖萱
Wang Linxuan(Guizhou University, Guiyang 550025, Guizhou, China)
出处
《工程与试验》
2018年第4期64-66,94,共4页
Engineering and Test
关键词
遗传算法
全状态实时模型
模型辨识
神经网络
genetic algorithm
full-state real-time model
model identification
neural network