前馈网络泛化性能改进的目的规划方法研究

THE GOAL PROGRAMMING METHOD FOR IMPROVING GENERALIZATION PERFORMANCE OF THE FEEDFORWARD NEURAL NETWORKS

以目的规划模型为基础，将前馈网络准则函数改进、网络灵敏度降低、先验知识运用有机地结合在一起，提出了前馈网络泛化性能改进的目的规划方法．文中给出了该方法的数学模型。

One of the important characteristic of feedforward neural networks is their ability to generalize the input/output behavior of functions based on a set of training examples.Yet many aspects of the problem of the improving generalization in feedforward neural networks have not been stuided well.The goal programmming method for neural network is proposed in this paper.Three enhancements are made in this method,those are:1)improvement of objective function;2)use of prior knowledge,3)decrease of the sensitivity of the neural network.Theoretical analysis and numerical example show that both generalization performance and the precision of forecasting of the neural networks are improved effectively.