遗传算法优化BP网络的汛期降水预测模型  

Optimization of BP neural network for short-term climate prediction of rainfall amount in flood season based on genetic algorithm

作　　者：吴有训[1] 王周青 汪文烈 束长汉 章爱国 汪小逸 陶曙华[1] 

机构地区：[1]安徽省宣城市气象局,宣城242000 [2]安徽省黄山市气象局,黄山245021

出　　处：《安徽农业大学学报》2013年第2期299-303,共5页Journal of Anhui Agricultural University

基　　金：中国气象局气象新技术推广项目(CMATG2005M34)资助

摘　　要：针对BP算法易陷入局部极小、收敛速度慢等缺点,遗传算法是全局优化算法和具有很强的全局搜索能力,遗传算法优化BP神经网络初始连接权值和阈值形成混合算法。以安徽宣城市为例,将汛期降水量作为预测对象,前期74项大气环流特征量、500 hPa、100 hPa月平均高度场、月平均海平面气压场和月平均海温场资料中选取预测因子,建立汛期降水短期气候预测模型。结果表明,该方法计算稳定,预报误差小,具有实用价值。BP algorithm has disadvantage of falling easily into local solution and slow rate of convergence,while genetic algorithm is global optimization algorithm and has strong global search ability.Genetic algorithm can be used to optimize the initial connection weights and thresholds of BP neural network,forming a hybrid algorithm.In this article,we took precipitation in flood season as the forecast objective in Anhui Xuancheng,selected factors from the monthly 74 circumfluent eigen values,the monthly mean height values at 500 hPa and 100 hPa,the monthly mean sea level pressure and SST field,to establish a model for prediction of rainfall amount in flood season.The results show that genetic algorithm has practical value for its stability and small forecast error.

关 键 词：遗传算法 BP(back propagation)神经网络 降水短期气候预测 

分 类 号：P457.6[天文地球—大气科学及气象学]