基于粒子群与BP混合算法的神经网络学习方法

Hybrid neural network algorithm method based on particle swarm optimization and BP neural network algorithm

针对标准BP算法收敛速度慢及易陷入局部极值等问题,提出一种基于粒子群优化与BP混合算法的神经网络学习方法。该方法在网络的训练过程中,同时利用粒子群算法与BP算法进行最优网络权值的搜索,从而既充分利用了粒子群算法的全局搜索性又较好地保持了BP算法本身的反向传播特点。将该混合学习算法应用于复杂函数的拟合仿真,并与标准BP算法以及传统的粒子群优化BP神经网络学习算法进行比较。实验结果表明所提的混合学习算法具有较高的收敛精度,且收敛速度更快。

For the standard BP （error Back Propagation） algorithm usually has the limitations of slow convergence and local extreme values, a new hybrid BP neural network learning algorithm based on particle swarm optimization （PSO） and BP algorithm was proposed in this paper. The main idea of the model is to find the optima weight for the network by using PSO method and BP algorithm simultaneously. Therefore, it can not only have the global property of PSO but also contain the feature of error back propagation of BP algorithm. The authors evaluate the model by simulation test on some typical complex functions and compare it with other models including standard BP network and traditional PSO based BP network. Experimental results show that the proposed hybrid learning algorithm has higher convergence accuracy, and faster convergence speed.