一种新的加权目标函数下的神经网络训练方法

采用罚函数算法的思想构造一个新的加权目标函数,可以用一个无约束优化过程实现约束条件下的参数寻优·基于此种新的加权目标函数,采用遗传算法训练了神经网络控制器参数·仿真表明,该方法比采用Clarke目标函数及其改进方案使系统具有更好的输出响应性能,更具有工程实用性·

Training Algorithm for Neural Network Based on a New Weighted Object Function

A new weighted object function based on the concept of penalty function algorithm was fabricated. Using this function the parameter optimization with constraints can be performed through a non constraint process. According to this new object function, the weight training of neurocontroller was performed using GA. The new object function provides better response than the method based on Clarke object function does, and it is more effective in practical application.