遗传算法在前向神经网络参数估计中的应用

An Application of Genetic Algorithms to Parameter Estimation for Feed-forward Neural Networks

人工神经网络在很多领域有着成功的应用。神经网络参数估计有许多训练算法 ,BP算法是前向多层神经网络的典型算法 ,但BP算法有时会陷入局部最小解。遗传算法是一种随机优化技术 ,它可以发现全局最优解。本文介绍了遗传算法在前向多层神经网络参数估计中的应用 ,并对标准遗传算法进行了适当的改进。结合具体例子给出了算法实现的操作步骤和实验结果。实验数据表明采用遗传算法得到的神经网络参数是最优的 ,神经网络的性能优于基于BP算法的神经网络性能。

Artificial neural networks were successfully applied to solve actual problems in many areas. There are a few training algorithms for parameter estimation of neural networks, in which back propagation (BP) algorithm is the typical algorithm for feed-forward multi-layer neural networks. However, BP algorithm sometimes traps into the local minimum. Genetic algorithm (GA) is a stochastic optimization technique.It can lead to global optimal solutions for complex problems. In this paper an application of GA to parameter estimation for feed-forward multi-layer neural networks is introduced. Suitable improvements are made for the standard genetic algorithms. An example of curve fitting is presented, and the steps of GA realization and experimental results are given. Experimental results have shown that parameters of the network obtained by GA can be ensured to be optimal and the performance of the network is better than that of the network based on BP algorithm.