BP神经网络优化算法研究

Study of the Optimized Algorithms of BP Neural Network

为解决BP神经网络收敛速度慢和易陷入局部极小值的缺点,利用遗传算法(GA)和基因表达式编程(GEP)的各自特点,基于BP算法提出了两种改进算法:其一是GA-BP算法,即利用GA优化BP神经网络的权值和阈值;其二是GEP-BP算法,即利用GEP对BP网络进行调整,包括网络结构、权值和阈值。用样本数据进行了测试并与基本BP算法进行了比较,结果表明两种改进算法具有很强的可行性和高效性。

To solve the BP neural network's disadvantages of trapping to a local optimum and being prone to converge to minimum,the authors propose two new improved algorithms based on BP neural network based on the characteristic of Genetic Algorithm and Gene Expression Programming respectively. One is GA-BP algorithm in which the weights and thresholds of BP neural network were optimized with GA; the other is GEP-BP which uses GEP to modify BP neural network ,including the architecture ,the weights and thresholds. Finally,the two new algorithms were implemented,and standard data was used to test them. Compared to BP neural network, the results show that these two new algorithms are effective and feasible method in real application.