人工神经网络方法在短期天气预报中的应用

Application of Aritifical Netural Network to Forecast of Short-Term Precipitation in Nanjing

将人工神经网络方法试用于南京夏季短期降水分级预报,根据南京夏季梅雨期的天气特点,用统计和动力学方法从HLAFS(高分辨率有限区域预报系统)预报产品中寻找预报因子,然后用2种方法选取输入因子分别对人工神经网络进行训练,并利用抽取的5天雨量实况作降水分级预报检验.通过对人工神经网络方法预报降水的结果与HLAFS降水预报以及逐步回归预报的结果对比发现:与HLAFS数值模式的降水预报相比,人工神经网络降水预报方法的准确率提高了20%以上,而且漏报、错报明显减少;特别是与逐步回归预报相比,大到暴雨的预报准确率得到了明显提高,这一研究表明人工神经网络方法在短期天气预报中也会有较大的应用价值.

Artificial neural network （ANN） is used to make categorical forecasting of short-term precipitation in Meiyu period of Nanjing in summer. At first, based on the weather characteristics of Meiyu period, in terms of the statistical and dynamic method , the predictors are searched in data of HLAFS（High resolution Limited Area Forecast System） . Secondly, utilizing import predictors that are selectedby two methods trains the ANN, and Finally, five days precipitation data are taken out to test the level forecast of precipitation. The results show that the simulating effect is very good and the ability of short-term precipitation forecast is obviously improved.