基于自适应花授粉算法的BP神经网络结构优化

传统的BP神经网络通过梯度下降法来调整网络的权值和阈值，使网络存在易陷入局部最优且收敛速度慢等缺陷，在很大程度上限制了BP神经网络的应用。针对BP网络存在的不足，提出利用自适应花授粉算法来优化BP网络的权值和阈值。对花授粉算法（FPA）中的转换概率做自适应调整，并引入自适应的变异因子，提出了自适应的花授粉算法（SFPA）；通过两种不同的结合方式将SFPA与BP神经网络进行融合，从而提出了SFPA1-BP神经网络和SFPA2-BP神经网络；通过函数逼近实验和数据集分类实验对新网络的性能进行验证。结果表明，SFPA1-BP和SFPA2-BP在函数逼近和分类方面都优于标准BP网络，且SFPA1-BP具有更高的泛化能力及学习能力。

New BP neural network based on adaptive flower pollination algorithm

The standard BP neural networks adjusts the weights and threshold value by the gradient descent method, which is easy to fall into the local optimum and slow convergence rate. Thus, the application of BP neural network is limited. In this paper, a Self-adaptive Flower Pollination Algorithm（SFPA） is proposed and used to optimize the weights and threshold value of BP neural networks. Firstly, the self-adaptive SFPA is put forward by adjusting the switch probability of FPA during iterations and as in the mutation operator. Then, the SFPA is integrated with the BP in two different forms：SFPA1-BP and SFPA2-BP. Finally, the performance of the new BP network is tested by using function approximation experiments and Iris classification data set. The?study shows that SFPA1-BP and SFPA2-BP are superior to other networks, especially in the function approximations and classification.