基于LightGBM算法的潮位预报修正研究

Application of the LightGBM algorithm to correct the tide level prediction

潮汐预报在数学上属于回归预测,是人工智能算法的经典应用领域之一。文章以上海米市渡站点为例,提出了采用LightGBM算法修正调和分析T_TIDE模型预报潮位的方法。以T_TIDE模型的预报误差序列作为LightGBM算法的输入层参数,训练得到的LightGBM模型可有效预测T_TIDE模型后续的短期(48 h内)预报误差,从而对T_TIDE模型的潮位预报结果进行短期修正。米市渡站测试结果表明,构建的LightGBM模型能将T_TIDE模型的24 h和48 h预报均方根误差分别降低至0.10 m和0.12 m,相应的±0.30 m合格率都提升至95%以上。但是,LightGBM算法在台风期间对T_TIDE模型的预报结果存在误修正,台风期间的潮位预报修正有待进一步研究。

Tide prediction mathematically belongs to regression which is one of the classical application areas of artificial intelligence algorithms.Taking the Mishidu hydrometric station of Shanghai as an example,this study introduced the LightGBM algorithm to correct the prediction results of the harmonic analysis method(T_TIDE model).The historical prediction errors of the T_TIDE model were specified as the input parameters.The trained LightGBM model can effectively predict the subsequent prediction errors of the T_TIDE model.The output of the LightGBM algorithm was used to correct the short-term(within 48 hours)predictions of the T_TIDE model.The testing results of Mishidu station show that the LightGBM algorithm can reduce the root-mean-square-error values of the 24 h and 48 h prediction of the T_TIDE model to 0.10 m and 0.12 m,respectively,while the corresponding pass percentage of the errors within 0.30 m can be increased to more than 95%.However,the LightGBM algorithm presented a wrong correction to the T_TIDE model results during the typhoon period.The correction of the tide level prediction during the typhoon period needs further research.