摘要
以南京１９６５～１９９４ 年４ 月平均气温作为预报量， 选取前期５００ ｈ Ｐａ 月平均高度场相关因子， 分别建立了事件概率回归预报模型和神经网络预报模型。通过对比分析发现， 在同等条件下，由于神经网络方法能更好地反映预报量与预报因子间的非线性关系， 并能有效避免采用事件概率回归方法预报建模时， 对预报因子分级造成信息损失的缺点。因此，
In this paper, two forecast models of regression of event probability (REEP)and neural network are established based on mean air temperature of Nanjing in April from 1965 to 1994 and previous monthly mean 500hpa height Comparative analysis shows that, in the same condition, the neural network technique can more effectively reflect the nonlinear relation between predicted facts and predictors and avoid information loss due to grading of predictors when the forecast model is developed by REEP Therefore, effects of fitting and prediction are much better than that of traditional forecast models of REEP
出处
《灾害学》
CSCD
1999年第3期12-16,共5页
Journal of Catastrophology
基金
中国气象局 "九五"项目资助
关键词
事件概率回归
神经网络
对比分析
天气预报
Regression of event probability, Neural network, Comparative analysis, Weather forecast
