基于深度学习和CatBoost的海浪波高预测方法研究  

Research on wave height prediction method based on deep learning and CatBoost

作　　者：卢鹏[1] 年圣全 邹国良[1] 王振华[1] 郑宗生[1] LU Peng;NIAN Shengquan;ZOU Guoliang;WANG Zhenhua;ZHENG Zongsheng(College of Information Technology,Shanghai Ocean University,Shanghai 201316,China)

机构地区：[1]上海海洋大学信息学院,上海201316

出　　处：《海洋湖沼通报（中英文）》2024年第5期28-34,共7页Transactions of Oceanology and Limnology

基　　金：上海市地方能力建设项目(19050502100);上海海洋大学科技发展专项(A2-2006-20-200211)。

摘　　要：基于深度学习和CatBoost技术,本文提出了一种混合模型(LACM)对墨西哥湾、芬迪湾和阿拉斯加湾附近的海浪波高进行预测,进而可以应用于其他海域的浪高预测。首先,对从美国国家数据浮标中心(NDBC)获取的海浪数据信息进行预处理;其次构建基于加入了注意力机制(AM)的长短期记忆(LSTM)神经网络模型和CatBoost集成学习模型;对预测后的结果进行重构。经过与LSTM、支持向量回归(SVR)、CatBoost等方法的比较,实验结果表明,LACM模型的平均绝对误差(MAE)、均方根误差(RMSE)和平均绝对百分比误差(MAPE)都是最优,曲线拟合效果最接近,且具备一定的鲁棒性。Based on deep learning and CatBoost technology,a hybrid model(LACM)was proposed to predict wave heights in the Gulf of Mexico,Bay of Fundy and Gulf of Alaska,which could then be applied to other sea areas.Firstly,the wave data obtained from the National Data Buoy center(NDBC)were preprocessed.Secondly,the LSTM neural network model and CatBoost integrated learning model were constructed.The predicted results were reconstructed.Compared with the LSTM,support vector regression(SVR),CatBoost and other methods,the experimental results showed that the mean absolute error(MAE),root mean square error(RMSE)and mean absolute percentage error(MAPE)of the LACM model were the lowest,and the prediction result was the best and the fitting effect was the closest,and it had a certain robustness.

关 键 词：AM LSTM CatBoost 海浪波高预测 重构 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]