基于BP神经网络方法的近岸数值海温预报释用技术

AN INTERPRETATION SCHEME OF NUMERICAL NEAR-SHORE SEA-WATER TEMPERATURE FORECAST BASED ON BPNN

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。

To enhance the forecast accuracy in near-shore regions, we analyzed the effect of numerical forecast and observation on temperature, discussed the best configuration, established an numerical forecast interpretation model, and assessed the performance of the scheme. Error analysis shows that, the numerical products and observations are essential in reducing and stabilizing the error of the interpretation model. The interpretation model reduced the forecast error （2.2℃） to 0.7℃. Althouth slightly higher than the error of training model, the forecast errors are stable. Thus, the numerical interpretation could perform as empirical forecast, and may have great potential of application for near-shore sea forecast.