遗传算法在BP网络权值学习中的应用

Application of genetic algorithm to weights learning in BP network

针对BP算法学习神经网络权值收敛速度慢、易陷入局部最优等缺点 ,采用遗传算法学习BP网络的权值 ;并给出了“异或”(XOR)问题的仿真示例 ,通过比较 ,发现遗传算法不会陷入局部最优 。

Aimed at the shortcoming of back propagation algorithm such as slow convergence speed of the weights during learning process, easy to be trapped into local optimum, etc, a genetic algorithm is adopted to learn the weights in BP neural network and, moreover, a simulation illustration of 'XOR' problem is given. It is found by comparison that genetic algorithm can avoid the trap of local optimum and effectively improve the convergence speed.