BP自适应网络在数值预报产品的释用

提出了动态指标库的概念，即不同时次采取不同的指标库，同时次在网络自适应时采取不断更新指标库的阈值，采取先建立消空指标库对样本进行消空，提高了降水样本在整体样本中的概率，然后利用BP(Back Propagation)网络作了大连短期的12h和24h降水有、无的预报，设计了网络的自适应算法，并在降水有、无预报中加以应用。结果表明，此方法比没有做消空样本和网络自适应的降水TS评分有所提高，在数值产品解释应用中更具有实际意义。

Self -adapt Network in Numerical Forecast Products

The conception of dynamic indexes is presented, namely different time taking different indexes. When network takes self-adapt, same time update value of the indexes. In this paper, using the indexes to get rid of the inconformity samples improves the sample of precipitation in the general sample, then the use of BP to forecast the precipitation of 12 hours and 24 hours is given. The self-adapt is also designed and used in the forecast. The result indicates that this method is better and has more practical value in explaining the numerical products.