基于混沌PSO-BP混合算法的神经网络  

Neural Network Trained by Hybrid Algorithms Based on Chaos Particle Swarm Optimization and Back-Propagation Algorithm

作　　者：孟非[1] 潘朋朋[2] 

机构地区：[1]江苏科技大学经济管理学院,江苏镇江212003 [2]江苏科技大学电子信息学院,江苏镇江212003

出　　处：《计算机仿真》2011年第2期196-199,共4页Computer Simulation

基　　金：江苏省高校自然科学基础研究项目(07KJB510032)

摘　　要：研究神经网络的优化问题,将粒子群优化(PSO)算法同误差反向传播(BP)算法采用两种算法相结合,形成两种混合算法,可用于训练神经网络的优化。提出两种方法,第一种混合算法是在PSO算法优化神经网络权值的同时注入BP算法,第二种混合算法是在PSO算法训练神经网络之后继之以BP算法。同时根据混沌映射的随机性和遍历性,将其引入到混合算法中,进一步提高算法的寻优能力。将这两种混合算法同基于PSO算法和基于BP算法的神经网络训练方法相比较,通过数值仿真实验表明,混合算法的性能优于所比较的两种算法的性能,且第一种混合算法要好于第二种混合算法。Two kinds of hybrid algorithms based on particle swarm optimization（PSO） and error back propagation（BP） are developed to train neural network（NN）.One of the hybrid algorithms is to train NN by PSO;simultaneously,inject BP algorithm into the PSO-based train process.The other is to train NN by PSO;after that,BP algorithm is used to train the network continuously.In order to enhance the global exploitation,the hybrid algorithms integrate chaotic mechanism for its stochastic property,ergodicity,and regularity.Compared the two hybrid algorithms with the NN training methods based on PSO algorithm and BP algorithm respectively,the numerical experiment shows that performances of the hybrid algorithms are superior to the other two methods.Moreover,the performances of the first hybrid algorithm are better than that of the second hybrid algorithm.

关 键 词：神经网络 粒子群优化 反向传播 混沌 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]