摘要
利用遗传算法进化设计神经网络的结构和连接权,并针对遗传算法局部调节能力比较弱的问题,采用从进化后的神经网络中用训练样本再次寻优的方法,建立神经网络气象预报模型,该方法克服了神经网络极易陷入局部解和遗传算法局部调节能力比较弱的问题,以广西的月降水量进行实例分析,计算结果表明该方法预报精度高、而且稳定。
Evolving neural network architectures and connection weights by using genetic algorithms, aiming at the problem of weak local regulation ability with genetic algorithms, and adopting secondary optimum from evolved neural network results by trained samples, this method has established a neural network meteorological forecast model. It can overcome the defects of neural network, namely, falling into local solution network and the problem of weak local regulation ability with genetic algorithms. An applied example is built with monthly mean rainfall across the whole area of Guangxi and calculation results indicate that the method has high and stable forecast accuracy.
出处
《热带气象学报》
CSCD
北大核心
2006年第4期411-416,共6页
Journal of Tropical Meteorology
基金
广西科技厅基金项目(0339025)
关键词
神经网络
遗传算法
预报建模
neural network
genetic algorithms
forecast model
