基于遗传算法的优化BP神经网络算法和应用研究.pdfVIP

Proceedings of the 29th Chinese Control Conference July 29-31, 2010, Beijing, China 基于遗传算法的优化BP神经网络算法与应用研究* 丁世飞, 苏春阳 中国矿业大学计算机科学与技术学院 徐州 221116 E-mail: dingsf@cumt.edu.cn 摘 要：反向传播(Back-Propagation, BP)神经网络是目前研究最为成熟的神经网络模型之一, 由于具有较好的自学习、 自适应、鲁棒性和泛化能力, 被广泛地运用模式识别、函数逼近和图像处理等各个领域. 然而BP 网络存在着收敛速率 慢、易陷入局部极小、隐层结构难以确定等不足, 针对这些不足所提出的诸多改进虽一定程度上进行了改善, 但仍在 存储空间、算法简洁等方面难以均衡. 本文对基于生物进化理论的遗传算法(Genetic Algorithms, GA)优化BP神经网络 进行了研究, 首先用GA优化BP 网络的权值和阈值, 以提高网络的收敛速率, 克服易陷入局部极小的的不足; 其次通过 对网络结构的优化, 以解决隐 神经元个数确定的难题, 建立了基于GA同时优化BP 网络的连接权值、阈值和隐层结构 的优化神经网络模型. 最后通过实例分析, 证明了算法的有效性. 关键词：遗传算法, BP神经网络, 权值, 拓扑结构 Application of Optimizing BP Neural Networks Algorithm Based on Genetic Algorithm DING Shifei, SU Chunyang School of Computer Science and Technology, China University of Mining and Technology, Xuzhou 221008, P.R.China E-mail: dingsf@cumt.edu.cn Abstract: Back-Propagation (BP) neural networks is one of most mature neural networks models. It has better self-learning, self-adapted, robustness and generalization ability and has been widely applied to pattern recognition, function approximation and image processing, etc. But for BP neural networks algorithm, the convergence rate is slow, it is easy to get stuck in a local minimum, and its structure is hard to designed. A lot of improved algorithms have been proposed to overcome these disadvantages, but these algorithms need more storage space and are not very brief. This paper does a research on the optimized BP neural networks based on Genetic Algorithm (GA), firstly, the optimized BP neural networks algorithm which optimizes the connection weights and structure at the same time with GA is establish