基于格点场数据的沙尘暴双预报模型

为提高沙尘暴预报的准确率，以描述大气环流形式的物理场格点数据作为建模样本，采用自组织神经网络对物理格点场数据样本进行聚类，构建出由大规模阵列式数据格式表示的建模样本的低维特征，再用模糊神经网络综合建模样本的一般性规律，用非典型样本进行二次建模以反映建模样本的特殊性，并设计隶属度调整方案对一般性和特殊性进行协调，由此形成兼顾建模样本一般性和特殊性的双预报模型．测试结果表明，基于特征提取方案的双预报模型体系使沙尘暴预报准确率达到80．4％．

Dual Forecasting Model of Sand-Dust Storm Based on the Lattice Position Field Data

To improve the forecasting precise ratio of the sand-dust storm, the physical lattice position field data,which is used to describe atmosphere circumfluence,constitutes modeling sample. The physical lattice position field data are clustered by self-organizing neural network. Based on these processed data, the low-dimensional features for modeling samples, which is depicted by large-scale point lattice data, are extracted successfully. A fuzzy neural network is trained to embody the generic rules hidden in modeling samples. Sequentially based on non-typical samples, further model is constructed to reflect their particularity. Then a membership degree adjusting formula is proposed to blend the universality with the particularity. Thus the dual forecasting model comes into being, which includes the generic rules and the particularity of modeling samples. The experimental results show that the forecasting precise ratio of the sand-dust storm predicted by the proposed dual forecasting model is 80.4%.