摘要
针对一般模块化模糊神经网络(MFNN)的门网络普遍采用模糊C均值聚类算法(FCM),没有对样本特征进行优化的问题,提出了在门网络中采用模糊核聚类算法(FKCA)替代模糊C均值聚类算法,构建了一种新的模糊核聚类模块化模糊神经网络预报模型。进一步采用动力消空算法、切比雪夫多项式展开方法和自然正交展开方法对预报量和预报因子进行计算处理后,分别建立了普通模块化模糊神经网络和模糊核聚类模块化模糊神经网络暴雨预报模型。利用这两种预报模型进行的暴雨预报试验表明,在相同的条件下,改进模型具有更高的暴雨预报TS评分。
Due to the fact that the characteristics of the sample are not optimized in Fuzzy C-mean （FCM） cluster algorithm,commonly used in the gate network of the general Modular Fuzzy Neural Network （MFNN）,Fuzzy Kernel-clustering Algorithm （FKCA） instead of FCM was used to propose a fkca-MFNN model.Further,computing predictand and predictors by using dynamic decreasing FAR algorithm,expression of chebyshev polynomial and Empirical Orthogonal Function,the general fcm-MFNN and fkca-MFNN rainstorm prediction models were set up.Experiments conducted with rainstorm as the predictive target by the two prediction models show that,under the same conditions,the improved model has higher TS valuation of rainstorm forecast compared to the fcm-MFNN model.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2009年第24期7902-7905,7909,共5页
Journal of System Simulation
基金
广西科学研究与技术开发项目(桂攻关052005-2A)
国家自然科学基金(40675023)
关键词
模块化模糊神经网络
暴雨预报
数值预报产品
非线性
modular fuzzy neural network
rainstorm forecast
numerical forecast product
non-linear
