排序方式： 共有2条查询结果，搜索用时 15 毫秒
1
1.
动态学习率神经网络预测气温的尝试 总被引：3，自引：1，他引：3
采用单站气象资料建立动态学习率的神经网络预测模型，进行逐日气温模拟预测，并与逐步回归预测模型和固定学习率的神经网络模型比较，结果表明，神经网络模型预测能力较好，但训练时间长。采用动态学习率的网络模型在不损失预测精度的前提下大大减少了训练时间，为神经网络在气象中的应用提供了一种方法。 相似文献
2.
The Global and Regional Assimilation and Prediction System (GRAPES), a limited-area
regional model, was used to simulate the onset of South China Sea summer monsoon. In view of the
relatively insufficient information about the initial field in simulation predictions, the Advanced
Microwave Sounding Unit-B (AMSU-B) data from a NOAA satellite were introduced to improve the initial
values. By directly using the 3-dimensional variational data assimilation system of GRAPES, two schemes
for assimilation tests were designed. In the design, Test 1 (T1) assimilates both sounding and AMSU-B
data, and Test 2 (T2) assimilates only the conventional sounding data, before applying the model in
simulation forecasts. Comparative experiments showed that the model was very sensitive to initial fields
and successful in reproducing the monsoon onset, allocation of high- and low-level wind fields during the
pentad of onset, and the northward advancement of the monsoon and monsoon rain bands. The scheme,
however, simulated rainfall and the location of the subtropical high with deviations from observations. The
simulated location of the subtropical high was more westward and northward and the simulated rainfall for
the South China Sea was larger and covered a broader area. 相似文献
1