BP 网络改进算法的性能对比研究

通过实例对几种具有代表性的用以训练BP网络权值的改进算法进行性能对比研究，首先分析了基于标准梯度下降法和基于标准数值优化方法获得的各种改进算法的优缺点，然后对各种改进算法在训练中所需的收敛时间及其所厉害误差进行对比分析，其结果为选择训练网络的算法，开阔人们对算法改进的思路提供了一些借鉴。

Comparative Study on Fast Learning Algorithms of BP Networks

A comparative study on some typical faster learning algorithms of BP networks is proposed. Firstly, the advantages and disadvantages of two main categories, the fast algorithms based on standard steepest descent and standard numerical optimization, are analysed by means of their work principles. Then the comparative study on convergence times and error performance of training of those faster algorithms is done by two numerical examples. The study can be used for the reference in selecting learning algorithms and developing high performance algorithms.