摘要
针对标准BP算法易陷入局部极值及收敛速度慢等问题,提出一种基于粒子群优化与BP算法的协同神经网络学习方法。该方法在网络的学习过程中,同时利用PSO优化算法与BP算法进行最优网络权值的协同搜索,从而充分利用粒子群算法的全局搜索性及BP算法的反向传播特点。将该算法应用于4个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神经网络算法进行比较。实验结果表明所提的协同算法的性能优于传统的BP网络优化算法。
Abstract For the standard BP algorithm usually has the limitations of local extreme values and slow convergence, a cooperative neural network learning method based on particle swarm optimisation (PSO) and BP algorithm is proposed in this paper. During the process of network learning, this method makes use of both IX50 and BP algorithms simuhaneously to carry out the. cooperative search of optimal network weight, so that it takes the full advantages of global search property of PSO and back propagation feature of BP algorithm. We apply this algorithm in fitting simulation with four complex functions and compare it with the BP neural network algorithms based on either standard BP network or traditional PSO. Experimental results show that the cooperative algorithm proposed performs better than the traditional BP network optimisation algorithms.
出处
《计算机应用与软件》
CSCD
北大核心
2013年第9期19-20,99,共3页
Computer Applications and Software
基金
国家自然科学基金项目(61074071
61104022)
关键词
BP算法
粒子群算法
优化
函数拟合
协同算法
BP algorithm Particle swarm optimisation algorithm Optimising Function fitting Cooperative algorithm