启发式改进BPNN在模式分类领域内的对比研究  

Comparative study of BPNNs improved by heuristic method in the field of pattern classification

作　　者：丁硕[1] 常晓恒[1] 巫庆辉[1] 杨友林[1] 

机构地区：[1]渤海大学工学院,辽宁锦州121013

出　　处：《电子设计工程》2014年第11期137-140,共4页Electronic Design Engineering

基　　金：国家自然科学基金(61104071)

摘　　要：采用附加动量BP算法、自适应最速下降BP算法、自适应动量BP算法、弹性BP算法4种启发式改进方法分别对标准BP算法进行改进,并构建了相应的BP神经网络分类模型,将构建的4种分类模型应用于二维向量模式的分类,并进行了泛化能力测试,将4种BP网络分类模型的分类结果进行对比。仿真结果表明,对于中小规模的网络而言,弹性BP算法改进的BP网络的分类结果最为精确,收敛速度最快,分类性能最优;附加动量BP算法改进的BP网络的分类结果误差最大,收敛速度最慢,分类性能最差;自适应学习速率BP算法改进的BP网络的分类结果的误差值、收敛速度及分类性能介于上述两种算法之间。Four kinds of heuristic methods including additional momentum BP algorithm, adaptive steepest descent BP algorithm, adaptive momentum BP algorithm and resilient BP algorithm are used to improve standard BP algorithm, and the corresponding BP neural networks are also established. The four kinds of classification methods are applied to classification of two-dimensional vectors. Then their generalization abilities are tested and the classification results of the four BP network are compared with each other. The simulation results show that for small and medium scale networks, BP neural network improved by resilient BP algorithm has the most accurate classification result, the fastest convergence speed and the best classification ability; the one improved by additional momentum BP algorithm has the biggest classification error, the slowest convergence speed and the worst classification ability; while the classification error, convergence speed and classification ability of BP neural network improved by adaptive steepest descent BP algorithm lie between the above two algorithms.

关 键 词：启发式方法 算法改进 BP神经网络 模式分类 泛化能力 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]