基于PCA和蜂群算法优化的BP神经网络

BP NEURAL NETWORK BASED ON PCA AND ARTIFICIAL BEE COLONY ALGORITHM OPTIMISATION

在研究BP神经网络的基础上,针对其收敛速度慢、容易陷入局部极小值等问题进行分析,设计实现一种新的混合神经网络模型。通过引入主成分分析的思想对样本进行降维,简化BP网络的结构,之后采用蜂群算法来优化BP网络的权值,把得到的最优权值赋予该神经网络,从而使优化后的神经网络具有结构简单、泛化性好和不易陷入局部极小值等优点。仿真实验结果表明,该网络模型能够达到比较高的分类精度。

After studying the disadvantages of BP neural network in low convergent speed and easy to be trapped into local minimum, we design and implement a new hybrid neural network model. First, by introducing the idea of principal component analysis （PCA） we reduce the dimensions of the samples, simplify the BP network structure, then by using artificial bee colony algorithm （ABC） we optimise the weight value of BP neuralnetwork. Finally, we assign the derived weight value to BP neural network, thereby enable the optimised BP neural network to possess the advantages of simple in structure, good in generalisation and not to be prone to falling into local minimum. Simulation experimental results show that this network model can reach higher classification precision.