作　　者： ; ; ;

机构地区： 东莞职业技术学院

出　　处： 《微型机与应用》 2012年第14期61-63,共3页

摘　　要： 为了提高神经网络进行函数拟合的精度,首先在三层径向基神经网络基础上通过增加网络层次和改变激励函数提出了一种四层径向基小波神经网络,并采用遗传算法来确定初始网络参数;其次针对遗传算法中容易早熟的缺点,在遗传算法中引入动态平衡策略,根据适应度的变化来动态改变遗传算法中交叉和变异概率,从而增加算法全局探索和局部开发的平衡能力;最后通过对函数拟合试验并与其他方法相比较表明了算法的有效性。 In order to improve the function fitting accuracy of the neural network,firstly, a four-layer RBF wavelet neural net- work is put forward on the basis of three-layer RBF neural network by increasing the level and setting the activation function to the wavelet function, and take genetic algorithm to determine contraction-expansion factor, size factor and weights of the initial net- work;secondly,aiming to premature convergence in genetic algorithm, dynamic equilibrium strategy is introduced to genetic algorithm, it dynamically change crossover and mutation probability in accordance with the variation of fitness to increase the equilibrium be- tween global exploratory and partial development. Finally,the function fitting experiment compared with other algorithms prove this method is better than others.

关 键 词： 径向基神经网络 小波神经网络 遗传算法 动态平衡 函数拟合

领　　域： [自动化与计算机技术] [自动化与计算机技术]