摘要
针对传统方法难以对性能参数进行有效预测的问题,提出一种基于过程神经网络的性能参数预测方法。为解决反向传播学习算法收敛速度慢、易陷于局部极小点等问题,开发了一种基于正交基函数展开的Leven-berg-Marquardt学习算法。为提高过程神经网络的泛化能力,从提高训练样本的质量和规模入手,研究了实际测量数据的预处理方法,并提出一种基于样条函数拟合和相空间重构理论的训练样本集构造方法。最后,将该方法用于某型航空发动机性能参数的预测,获得了满意的结果。
It was difficult for the traditional methods to predict performance parameters effectively,aiming at this problem,a performance parameter prediction method based on the process neural network was proposed.Back Propagation（BP） learning algorithm was with low convergence speed and it was easy to a local minimum point.To solve these problems,a Levenberg-Marquardt learning algorithm based on the expansion of the orthogonal basis functions was developed.To improve the generalization capability of process neural network,from the quality and scale of the training samples,data pretreatment for the actual measurement data was studied,and a method for the construction of the training samples based on the spline functions approximation and the phase space reconstruction theory was proposed.Finally,the proposed prediction method was applied to predict the performance parameters of aeroengine,and the test results were satisfactory.
出处
《计算机集成制造系统》
EI
CSCD
北大核心
2011年第1期198-207,共10页
Computer Integrated Manufacturing Systems
基金
国家863计划重点资助项目(2009AA043403-2)
国家自然科学基金资助项目(50805032)~~
