基于结构的神经网络在参数优化中的应用

在对传统人工神经网络优化方法的认识基础上，针对复杂非线性系统的优化问题，提出了一种基于结构的神经网络优化方法。它将一个复杂系统转化为若干个较简单的子系统，分别建立各子系统的函数链神经元模型，然后根据原系统的结构特点将它们连接起来构成一个基于结构的神经网络。网络权值与系统的结构参数相对应，具有明确的物理意义，通过调整权值即可实现系统结构参数的优化。对Y2-Hc10型先导式溢流阀的优化研究表明，该方法为大型、严重非线性系统的结构参数优化提供了一条新的途径。

Application of Architecture-Based Neural Networks in Parameter Optimization

With the cognition of the optimization method of generic artificial neural networks, a novel method using architecture based neural networks was presented for the parameter optimization of complicated nonlinear systems. This method translates a complicated system into some simple sub systems. Each sub system is modeled with a functional link neuron. Then these neurons are connected into an integrated network according to the relationship between the sub systems. Corresponding to the structural parameters of the system, the weight of this networks has explicit physical meanings. The parameter optimization can be done by adjusting the weight. The validity of the method was illustrated by an optimizing on structure parameters of Y 2 Hc10 pilot operated relief valve. And a new approach was provided for the parameter optimization of severe nonlinear system.