基于遗传算法和梯度下降的RBF神经网络组合训练方法

在使用基于梯度下降的径向基函数（RBF）神经网络学习方法时，由于网络目标函数误差曲面极其复杂，因而产生了网络收敛速度慢，且容易陷入局部极小，网络初始值的设置对网络训练结果影响很大等问题。基于遗传算法的训练方法能够摆脱陷入局部最优的困扰，但遗传算法的局部搜索能力不够，从而影响网络的训练效果。为了解决上述问题，在研究两种算法特点的基础上，提出一种组合训练方法，用提出的训练方法对UCI中的部分数据集进行了仿真实验，并将实验结果与传统方法下的结果进行了比较，实验结果表明新方法是有效的。

Combined algorithms for training RBF neural networks based on genetic algorithms and gradient descent

The deficiencies of gradient descent method include the slow speed of convergence, the problem of local minima and the great influence of initial parameters on the performance of the network. Genetic Algorithm (GA) based methods can get rid of the problem of local optima, but they are not very effective to refine an existent good solution. For resolving these problems, in this paper we proposed a new algorithm. The experimental results show the algorithm performs well, and it is better than both Gradient descent algorithm and genetic algorithm.