基于人工神经网络的海流预报研究

Study of tidal current prediction based on artificial neural network

以调和分析预报结果、海面风速、风向作为输入因素,采用神经网络拟合实测海流与各输入因素之间的关系。该方法不但反映了海面风对潮流的影响,对调和分析预报本身的误差也有一定的改善。研究还表明多步变量输入的预报模式可以反映输入变量的变化趋势,提高预报的准确性。在宁波-舟山港某航道海流预报计算实例中,该方法流速预报的均方误差比直接调和分析预报下降了22%,流向预报的均方误差下降了18%。

Neural network is used to simulate the relationship between the measured tidal current and the input factors such as Harmonic predict results and sea surface wind. This method can reflect the influence of the sea surface wind and reduce the error of the harmonic prediction. The study also shows that the prediction model of multi-step input can reflect the changing trend of input variables and improve the accuracy. The method is used to predict tidal current of a navigation channel in Ningbo-Zhoushan harbor. In the case, mean square error of velocity prediction is 22% lower than Traditional harmonic method, and the mean square error of direction prediction also decreases by 18%.