变尺度混沌算法的BP网络优化

BP network optimization based on mutative scale chaos optimization algorithm

采用变尺度混沌优化方法代替梯度下降法融入BP神经网络,在优化搜索过程中不断缩小搜索空间,克服了标准BP算法易陷入局部极小的缺点,能有效地寻找到BP神经网络权值的全局最优值;此外,进一步提出变尺度混沌优化与梯度下降法有机结合的算法,能有效缩短单一的变尺度混沌优化BP算法的训练时间。仿真结果表明,改进的BP神经网络具有实现简单、寻优性强和优化效率高等特点。

This paper presented a kind of BP network training method that based on mutative scale chaos optimization algorithm.By continually reducing the searching space of variable optimized,the method overcame the problem that conventional BP algorithm was liable to trap in local minimum value and could effectively search out the global optimization weight values of neural network.Based on the improved BP algorithm and gradient decline algorithm,further proposed a new hybrid optimization algorithm,which could effectively shorten the training time of the single BP algorithm based on mutative scale chaos optimization algorithm.The simulation results show that the improved BP network has characteristic of simple structure,strong optimization and high optimization efficiency.