基于模拟退火遗传算法的模糊分类器参数优化及其应用  

THE PARAMETER OPTIMIZATION OF MMNN BASED ON GENETIC ALGORITHM COMBINED WITH SIMULATED ANNEALING AND ITS APPLICATION

作　　者：周越[1] 相敬林[2] 杨杰[1] 

机构地区：[1]上海交通大学图像处理与模式识别研究所,上海200030 [2]西北工业大学航海工程学院,西安710072

出　　处：《电子与信息学报》2001年第10期975-983,共9页Journal of Electronics & Information Technology

摘　　要：该文从结构和算法上研究了Max-Min模糊神经网络(MMNN),找出了其固有的局限性,相应提出了一系列的改进措施形成改进MMNN算法。为了更好地提高网络的性能,同时考虑到优化算法的收敛速度,本文提出了基于模拟退火遗传算法的网络参数优化方法,通过计算机仿真,证明了该方法是可行的。最后,运用它作为分类器对实际的船舶辐射噪声进行了分类实验,与BP等算法进行了比较,显示出其独特的优越性。In this paper, the structure and algorithm of Max-Min fuzzy neural network (MMNN) are studied in detail. In order to get rid of some intrinsic localization of the method and boost up the capability of the MMNN, a series of steps are presented and the improved project (IMMNN) is gained. With a view to making the capability even much better and compressing the time of the convergence, the op-IMMNN is put forward in which the parameters of IMMNN are optimized by genetic algorithm combined with simulated annealing. In the simulation, the result of op-IMMNN is superior over the conventional MMNN's. Finally, a satisfactory result is also obtained when op-IMMNN is regarded as a classifier to distinguish the types of the ships according to their actual radiated noise. Comparing with the neural network based on the back propagation algorithm, the advantages of the op-IMMNN are fully put up.

关 键 词：模糊隶属度函数 神经网络 模拟退火算法 遗传算法 模糊分类器 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]