前馈神经网络的混沌学习方法研究

Chaos Optimization Design for Feed-forward Neural Networks

采用混沌优化策略 ,提出一种前馈神经网络权参数的最优学习方案 .由于 BP算法优化神经网络权参数时存在收敛速度慢、自身参数选取困难、易陷入局部极小等缺陷 .采用混沌变量优化神经网络权参数 ,具有全局性、快速性、并行性的特点 .仿真实验表明采用该方案对强非线性问题的逼近具有精度较高、学习较快的优点 .

In this paper, the optimization design for feed-forward neural network is proposed based on chaos optimization. The BP algorithm is applied for optimizing feed-forward neural network, it has local minimum, slow convergence speed and difficulty in selecting self-parameters. The chaotic variables are applied to optimize neural network weight parameters, which have many advantages of global minimum, fast convergence speed and in parallel. Simulation results show control results to the strong nonlinear problem process have high precision, fast convergence speed.