一种神经网络的云图短时预测方法

依据6hT213数值预报产品的资料，采用EOF展开和人工神经网络等方法，对卫星云图短时预报方法进行研究。首先对卫星云图灰度值样本序列进行EOF展开，将提取出来的时间系数作为建模的预报量，以数值预报产品的物理量场作为预报因子，建立人工神经网络预测模型。将预报得到的时间系数与空间特征向量进行时空反演，实现对未来6h云图的预测。预报方法的独立样本试验证明，预测结果与实际云图的主要特征基本吻合，尤其在预测云图的大体分布和发展趋势上得到了较好效果。

A Short-term Cloud Forecast Model by Neural Networks

A short-term cloud forecast model, based on the numerical forecast products data, is studied in this paper by means of the empirical orthogonal function （EOF） and artificial neural network （ANN） method. Firstly, the time coefficient of the EOF of the sample sequences of gray scale images of clouds was taken as the predictand and physical factors of numerical forecast products as the predictors, and an ANN forecast model was established. The future 6 h cloud forecast was made by the space-time inversion from the predicted time coefficient and the corresponding eigenmodes. The ANN cloud forecast model was verified by independent samples and the results show that the forecast results are better accorded with observed cloud pictures in the principle characteristics, especially in the general distribution and developing trend.