论文部分内容阅读
针对前向网络反向传播算法（ＢＰ）训练缓慢和易于陷入局部极小的缺点以及反向运算的复杂性，利用ＢＰ算法监督学习特点、模拟退火（ＳＡ）算法在局部极小处的概率突跳特性和遗传算法（ＧＡ）的并行化群体搜索的特点，有效结合ＢＰ和ＳＡ算法以及ＧＡ和ＳＡ算法，提出了前向网络的两种混合学习策略即ＢＰ＆ＳＡ混合策略和ＧＡ＆ＳＡ混合策略。以异或问题为例，通过计算机仿真对混合策略与ＢＰ、改进ＢＰ算法的比较表明混合学习策略较大程度改进了前向网络学习的收敛性能和收敛速度，并一定程度上避免了反向运算的复杂性，是前向神经网络学习的有效算法。
Aiming at the shortcomings of slow training and easy to fall into local minima and the complexity of reverse operation of forward-looking network backpropagation algorithm (BP), the probability of simulated annealing (SA) algorithm at local minima is supervised by BP algorithm. Hopping characteristics and genetic algorithm (GA). By combining BP and SA algorithms and GA and SA algorithms effectively, two mixed learning strategies of forward network, namely BP & SA hybrid strategy and GA & SA hybrid strategy, are proposed. Taking the XOR problem as an example, the comparison between the hybrid strategy and BP and the improved BP algorithm by computer simulation shows that the hybrid learning strategy improves the convergence performance and convergence rate of the forward learning to some extent, and to some extent avoids the reverse operation The complexity of the forward neural network learning algorithm.