基于改进遗传算法的藻类神经网络识别  

Neural Network Recognization for Algae Based on Improved Genetic Algorithms

作　　者：姚志红[1] 费敏锐[2] 孔海南[3] 谢雳[1] 孙林峰[1] 

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240 [2]上海大学机电工程与自动化学院,上海200072 [3]上海交通大学环境科学与工程学院,上海200240

出　　处：《上海交通大学学报》2007年第11期1801-1805,共5页Journal of Shanghai Jiaotong University

基　　金：国家高技术研究发展计划(863)项目(2003AA601040-02);上海市科委国际合作处资助项目(052307055);高等学校博士学科点专项科研基金(20040280017)资助项目

摘　　要：为了借助智能方法对日益严重的水污染进行识别预测,提出了一种改进遗传神经网络识别算法.该算法通过各个分算子的多元冲突、融合、协作和互补等方式,有机地结合形成一种整体优化算子,它包含了降维差异选择、暂态自适应交叉和冲突自适应变异等3个新的彼此相关联的分算子,能有效地生发多样性,提高解空间处处可达性.数值优化及湖泊蓝绿藻神经网络识别实验表明,该算法在克服早熟、提高全局收敛速度和增强神经网络的泛化能力等方面,均取得满意的藻类识别效果.In order to recognize algae of lakes and reservoirs, an improved ethics genetic algorithm （EGA） and application of improved genetic neural network were introduced. The EGA works in the measure that much operators conflict, syncretize, cooperate and complement each other to come into being an integer optimized algorithm. It includes difference choice on reduce dimension, temporary state adaptive crossover operator and collision adaptive mutation operator and can effectively procreate diversity. The results of simulation of numerical value optimization and recognition of blue-green algae indicate that this approach can get over prematurity, accelerate globe convergence speed, improve generalization capability of neural networks （NN） and acquire a satisfactory effect of optimizing and recognition.

关 键 词：藻类神经网络识别 改进遗传算法 逆向选择 暂态交叉 冲突变异 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]