摘要
构造了一种带多个核函数的结构自适应径向基函数神经网络(RBF NN)降水预报集成模型.该网络结构同时使用6个核函数,采用基于输入输出全部信息的模糊相似矩阵的平均矩阵元法自动确定隐节点数,自适应地生成RBF神经网络集成个体,最后建立多元回归模型集成.对日本的细网格资料数据建立平均日降水量预报模型,利用MATLAB进行仿真实验,结果表明,该模型预报性能明显优于同期中国气象局的T213(中国气象局的全球中期天气数值预报产品预报值)降水预报,可为气象预报研究提供新思路,为降水预报决策的制定提供重要的参考.
This paper constructs an integrated model of radial basis function neural network （RBF NN）with self-a-daptive structure of multi-kernel function on precipitation forecast. The network structure using six nuclear functions at the same time adapts the method of mean matrix element based on the fuzzy similarity matrix of the input and output information and automatically ensures the number of hidden nodes. The generated RBF neural network ensembles indi-vidual and finally establishes integrated model of multiple regression. It also establishes the mean daily precipitation forecast model on Japanese fine grid data and does simulation experiment by using the method of MATLAB. The result shows that the predictability of this model is obviously superior to the T213 （the forecast value tested by the global me-dium-range numerical forecasting product of China Meteorological Bureau）,which can provide new ideas for the study of weather forecast and important references for the decision making of the precipitation forecast.
出处
《昆明理工大学学报（自然科学版）》
CAS
北大核心
2014年第2期50-57,73,共9页
Journal of Kunming University of Science and Technology(Natural Science)
基金
国家自然科学基金项目(51065002
11161029)
广西自然科学基金青年项目(2012GXNSFBA053161)
广西教育厅高校科研项目(201203YB186)
柳州师范高等专科学校项目(LSZ2012B003)
关键词
多核径向基函数
集成神经网络
多方法集成建模
降水预报
multi-kernel radial basis function
integrated neural network
multi-method integrated modeling
precipitation forecast
