基于BP神经网络方法的近岸数值海温预报释用技术

发布时间：2018-06-14 16:20

  本文选题：数值预报 + 释用　； 参考：《海洋与湖沼》2016年06期

【摘要】：为了提高近岸精细化海温预报精度,利用神经网络方法,分析了海温数值预报及观测数据在释用中的作用,研究了定点近岸海温影响因子的最优配置方案,建立了定点海温精细化数值预报释用模型,评估了释用模型性能。误差分析结果显示,数值海温产品及其观测在建模中起到了降低和稳定模型误差的作用;释用模型将定点数值预报的误差从2.2°C减少至0.7°C;预报误差较调训误差略高,但考虑到预报误差的稳定性,数值释用与人工经验预报水平持平,因此,该方法具有十分广阔的拓展空间和应用前景。
[Abstract]:In order to improve the precision of inshore fine SST prediction, the function of SST numerical forecast and observation data in interpretation is analyzed by using neural network method, and the optimal configuration scheme of the influence factors of fixed NLS is studied. A precise numerical prediction model of SST is established and the performance of the model is evaluated. The results of error analysis show that the numerical SST products and their observations play a role of reducing and stabilizing the model errors in modeling, that the numerical prediction errors of fixed points are reduced from 2.2 掳C to 0.7 掳C, and that the prediction errors are slightly higher than those of training. However, considering the stability of prediction error, the numerical interpretation level is equal to that of artificial experience, so this method has a very broad space and application prospect.
【作者单位】： 国家海洋环境预报中心;
【基金】：国家自然科学基金项目,41222038号 海洋公益性行业科研专项经费项目,201305031号
【分类号】：P731.31
，

本文编号：2018119