中文摘要：

对于一个具体的多层前向神经网络设计问题,网络的输入输出以及标准样本数为已知,网络的隐层结构,即隐层层数和每个隐层神经元个数如何选择是神经网络设计中的关键。根据代数方程理论,通过权值和阈值与隐层结构的关系,建立了以权值和阈值为设计变量的目标函数表达式,通过分析,提出了多层前向神经网络合理的隐层层数和每个隐层神经元个数的一般确定方法,给出了确定多层前向神经网络合理结构的优化目标函数及其约束条件。仿真研究结果表明所提出方法确定的多层前向神经网络结构是合理的。

英文摘要：

For the structure design of a multi-layer feedforward neural network （MFNN）, the number of input, output and training datum of MFNN is known, and how to establish the structure of hidden layers, such as numbers of hidden layers and unit number per hidden layer, is the most important process. Based on algebra equation theory, the expression composed by power and threshold was established. Furthermore, a general confirming method of rational construct of MFNN was proposed. And then the target function and restricting condition to optimize the structure of MFNN were given. The simulation results show the structure of MFNN constructed by the proposed method is reasonable.