摘要
针对BP神经网络收敛速度慢、易陷入局部极小的缺点,提出将改进的人工鱼群算法与BP算法相结合的混合算法训练人工神经网络,建立了相应的优化训练模型及训练过程。通过基于生物免疫机制改进的人工鱼群算法优化训练多层前向神经网络,使神经网络对训练初值和参数要求不高,扩大了权值的搜索空间,提高了收敛速度和学习精度,有效地协调全局和局部搜索能力。仿真结果表明,该算法性能优于其它算法,具有均方误差值小、收敛速度快和计算精度高等特点,是一种更有效的神经网络训练算法。
According to weak points of slow convergence and being apt to local minimum about BP neural network, hybrid algorithm combining improved artificial fish-swarm algorithm with BP （error back propagation） algorithm is suggested to train the artificial neural network. Besides, the corresponding optimized training model and training process are set up. The biological immunity mechanism is introduced into the AFSA to optimize the weight and threshold of neural network, to enlarge the search space of the weight of it, better the convergence speed and learning accuracy, and effectively coordinate the search ability. Simulation result shows that because of the advantages of low request for initial values and parameters, little mean square errors, fast convergence and accurate calculation, this algorithm is better than other algorithms, so it is a more efficient neural network training algorithm.
出处
《计算机工程与设计》
CSCD
北大核心
2009年第20期4719-4721,4765,共4页
Computer Engineering and Design
基金
国家星火计划基金项目(2007EA780068)
广东省自然科学基金项目(7010116)
广东省粤港关键领域重点突破基金项目(2006A25007002)
湛江市科技计划基金项目(2008C08017)
关键词
改进人工鱼群算法
BP神经网络
免疫算子
组合优化
随机搜索
improved artificial fish-swarm algorithm
BP artificial neural network
immune operator
combinatorial optimizing
random search
