Research on typhoon wave height prediction method based on BO-LSTM neural network model
-
摘要: 随着海平面上升和风暴增强等气候变化的影响，快速准确地预报台风浪波高对于海岸保护和海洋灾害预防显得格外重要。本文首先基于TCWiSE模型生成大量的虚拟台风，利用SWAN数值模式计算台风期间测站处的有效波高，并构建台风浪样本数据库；然后对BO-LSTM神经网络模型的输入因素和超参数进行评估和选取，结合样本数据库对其进行训练和测试。结果表明：所构建的虚拟台风与历史台风具有很好的相似性，可以为台风浪波高智能预报模型的搭建提供充足的数据基础；所搭建的BO-LSTM模型可以快速实现单一站点处的台风浪波高智能预报，并具有媲美SWAN的预报精度，其在长时间预报场景中的预报精度显著优于RF模型和BPNN模型；将预报的未来台风数据添加到BO-LSTM模型的输入中，进一步提高了模型的预报精度和预报未来时长，其预报未来24 h的Bias、RMSE和R2分别为−0.102 m、0.494 m和0.855。研究成果为极端天气下台风浪的智能预报提供一种可行的实现途径。Abstract: With the impact of climate change such as rising sea levels and intensified storms, it is particularly important to quickly and accurately predict typhoon wave heights for coastal protection and marine disaster prevention. This article first generates a large number of virtual typhoons based on the TCWiSE model, uses the SWAN numerical model to calculate the significant wave height at the observation station during the typhoon, and constructs a sample database of typhoon waves; Then evaluate and select the input factors and hyperparameters of the BO-LSTM neural network model, and train and test it using a sample database. The results show that the constructed virtual typhoon has good similarity with historical typhoons, which can provide sufficient data basis for the construction of intelligent typhoon wave height prediction models; The BO-LSTM model built can quickly achieve intelligent prediction of typhoon wave height at a single station, and has prediction accuracy comparable to SWAN. Its prediction accuracy in long-term forecasting scenarios is significantly better than RF and BPNN models; Adding future typhoon data to the input of the BO-LSTM model further improves the accuracy and duration of the model’s forecast. Its Bias, RMSE, and R2 for predicting the next 24 h are −0.102 m, 0.494 m, and 0.855, respectively. The research results provide a feasible approach for intelligent forecasting of typhoon waves under extreme weather conditions.
-
Key words:
- typhoon waves /
- wave height forecast /
- LSTM model /
- bayesian optimization /
- virtual typhoon
-
图 1 1979–2022年西北太平洋地区历史台风路径图
Fig. 1 Map of historical typhoon tracks in the Northwest Pacific region from 1979 to 2022
图 2 波浪测站位置、网格范围及水深示意图
Fig. 2 Schematic diagram of wave measurement station location, grid range, and water depth
图 4 台风浪波高智能预报模型的构建流程图
虚线框内为贝叶斯优化过程
Fig. 4 The construction process diagram of an intelligent typhoon wave height prediction model
图 5 历史台风与随机抽取
1128 场虚拟台风的起点对比图（a）与终点对比图（b）Fig. 5 Comparison diagram of starting point (a) and ending point (b) of historical typhoons and randomly selected
1128 virtual typhoons
图 6 验证站点位置示意图（a）和影响各站点的历史台风与虚拟台风的参数对比图（b、c）
Fig. 6 Verification site location diagram (a) and parameter comparison diagram of historical and virtual typhoons affecting each site (b, c)
图 7 3场典型历史台风期间不同测站处的有效波高验证对比图
Fig. 7 Comparison chart of significant wave height verification at different stations during three typical historical typhoons
图 8 输入因素不同前时序条件下各组次的均方根误差
Fig. 8 Root mean square error of each group under different pre time series conditions of input factors
图 9 3场典型台风期间不同机器学习模型对舟山站有效波高的预报结果与SWAN模拟结果的对比
Fig. 9 Comparison between the prediction results of significant wave height at Zhoushan Station using different machine learning models and SWAN simulation results during three typical typhoons
图 10 BO-LSTM预报有效波高与SWAN模拟有效波高之间的散点密度图
Fig. 10 Scatter density plot between BO-LSTM predicted significant wave height and SWAN simulated significant wave height
表 1 输入因素不同种类条件下各组次的均方根误差
Tab. 1 Root mean square errors of each group under different types of input factors
输入因素种类 验证集RMSE/m 台风因素 0.878 测站气象因素 0.744 前时序波高 0.078 台风因素+测站气象因素 0.687 台风因素+前时序波高 0.066 测站气象因素+前时序波高 0.036 所有因素 0.025 
下载: 导出CSV
表 2 不同预报时长条件下LSTM最优超参数的选取方案
Tab. 2 Selection schemes for optimal hyperparameters of LSTM under different forecast duration conditions
超参数名称 寻优范围 选取方案 1 h 3 h 6 h 12 h 神经元个数 （2，512） 214 174 196 272 丢弃率 （0.01，0.2） 0.05 0.03 0.11 0.08 批尺寸 （8，32） 12 15 16 8 迭代次数 （10，100） 34 68 52 76
下载: 导出CSV
表 3 不同预报方案条件下BO-LSTM模型在测试集上的Bias、RMSE和R2
Tab. 3 Bias, RMSE, and R2 of the BO-LSTM model on the test set under different forecasting schemes
预见期 不添加预报风场 添加预报风场 Bias/m RMSE/m R2 Bias/m RMSE/m R2 1 h 0.001 0.017 0.999 0.001 0.012 0.999 3 h −0.013 0.092 0.997 0.009 0.064 0.998 6 h 0.018 0.194 0.978 0.014 0.122 0.986 12 h −0.088 0.425 0.892 −0.063 0.316 0.931 24 h / / / −0.102 0.494 0.855
下载: 导出CSV
-
[1] 陶爱峰, 沈至淳, 李硕, 等. 中国灾害性海浪研究进展[J]. 科技导报, 2018, 36(14): 26−34.Tao Aifeng, Shen Zhichun, Li Shuo, et al. Research progrecs for disastrous waves in China[J]. Science & Technology Review, 2018, 36(14): 26−34. [2] 屈远, 高志一, 蔡靖泽, 等. 数值模型和智能模型的海浪预报能力比较[J]. 海洋预报, 2022, 39(5): 17−26. doi: 10.11737/j.issn.1003-0239.2022.05.003Qu Yuan, Gao Zhiyi, Cai Jingze, et al. Comparison of wave prediction ability between numerical model and AI model[J]. Marine Forecasts, 2022, 39(5): 17−26. doi: 10.11737/j.issn.1003-0239.2022.05.003 [3] Wilson B W. Numerical prediction of ocean waves in the North Atlantic for December, 1959[J]. Deutsche Hydrografische Zeitschrift, 1965, 18(3): 114−130. doi: 10.1007/BF02333333 [4] 许富祥, 许林之. 海浪预报方法综述(二)[J]. 海洋预报, 1989, 6(4): 50−58.Xu Fuxiang, Xu Linzhi. Overview of wave forecasting methods (Ⅱ)[J]. Marine Forecasts, 1989, 6(4): 50−58. [5] 刘凡, 陆小敏, 徐丹, 等. 海浪预报方法研究进展[J]. 河海大学学报(自然科学版), 2021, 49(5): 387−393.Liu Fan, Lu Xiaomin, Xu Dan, et al. Research progress of ocean waves forecasting method[J]. Journal of Hohai University (Natural Sciences), 2021, 49(5): 387−393. [6] Fan Shuntao, Xiao Nianhao, Dong Sheng. A novel model to predict significant wave height based on long short-term memory network[J]. Ocean Engineering, 2020, 205: 107298. doi: 10.1016/j.oceaneng.2020.107298 [7] Zhou Shuyi, Xie Wenhong, Lu Yuxiang, et al. ConvLSTM-based wave forecasts in the South and East China Seas[J]. Frontiers in Marine Science, 2021, 8: 680079. doi: 10.3389/fmars.2021.680079 [8] Gao Zhiyi, Liu Xing, Yv Fujiang, et al. Learning wave fields evolution in North West Pacific with deep neural networks[J]. Applied Ocean Research, 2023, 130: 103393. doi: 10.1016/j.apor.2022.103393 [9] Pan Yi, Chen Yongping, Li Jiangxia, et al. Improvement of wind field hindcasts for tropical cyclones[J]. Water Science and Engineering, 2016, 9(1): 58−66. doi: 10.1016/j.wse.2016.02.002 [10] Ying Ming, Zhang Wei, Yu Hui, et al. An overview of the China meteorological administration tropical cyclone database[J]. Journal of Atmospheric and Oceanic Technology, 2014, 31(2): 287−301. doi: 10.1175/JTECH-D-12-00119.1 [11] Lu Xiaoqin, Yu Hui, Ying Ming, et al. Western North Pacific tropical cyclone database created by the China meteorological administration[J]. Advances in Atmospheric Sciences, 2021, 38(4): 690−699. doi: 10.1007/s00376-020-0211-7 [12] Nederhoff K, Hoek J, Leijnse T, et al. Simulating synthetic tropical cyclone tracks for statistically reliable wind and pressure estimations[J]. Natural Hazards and Earth System Sciences, 2021, 21(3): 861−878. doi: 10.5194/nhess-21-861-2021 [13] Booij N, Ris C R, Holthuijsen H L. A third-generation wave model for coastal regions: 1. Model description and validation[J]. Journal of Geophysical Research: Oceans, 1999, 104(C4): 7649−7666. doi: 10.1029/98JC02622 [14] Holland G J. An analytic model of the wind and pressure profiles in hurricanes[J]. Monthly Weather Review, 1980, 108(8): 1212−1218. doi: 10.1175/1520-0493(1980)108<1212:AAMOTW>2.0.CO;2 [15] 杨万康, 尹宝树, 伊小飞, 等. 基于Holland风场的台风浪数值计算[J]. 水利水运工程学报, 2017(4): 28−34.Yang Wankang, Yin Baoshu, Yi Xiaofei, et al. Numerical calculation and research of typhoon waves based on Holland wind field[J]. Hydro-Science and Engineering, 2017(4): 28−34. [16] 马秀玲, 魏来. 基于Holland台风模型及三重嵌套海浪模式的台风浪数值模拟研究[J]. 海洋与湖沼, 2024, 55(1): 51−64.Ma Xiuling, Wei Lai. Numerical simulation of typhoon waves based on the Holland typhoon model and triple nested wave pattern[J]. Oceanologia et Limnologia Sinica, 2024, 55(1): 51−64. [17] Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735−1780. doi: 10.1162/neco.1997.9.8.1735 [18] 李亚茹, 张宇来, 王佳晨. 面向超参数估计的贝叶斯优化方法综述[J]. 计算机科学, 2022, 49(S1): 86−92. doi: 10.11896/jsjkx.210300208Li Yaru, Zhang Yulai, Wang Jiachen. Survey on Bayesian optimization methods for hyper-parameter tuning[J]. Computer Science, 2022, 49(S1): 86−92. doi: 10.11896/jsjkx.210300208 [19] Shahriari B, Swersky K, Wang Ziyu, et al. Taking the human out of the loop: a review of Bayesian optimization[J]. Proceedings of the IEEE, 2016, 104(1): 148−175. doi: 10.1109/JPROC.2015.2494218 [20] Georgiou P N. Design wind speeds in tropical cyclone-prone regions[D]. London, Canada: Western University, 1986. [21] Vickery P J, Wadhera D, Twisdale L A, et al. U. S. Hurricane wind speed risk and uncertainty[J]. Journal of Structural Engineering, 2009, 135(3): 301−320. doi: 10.1061/(ASCE)0733-9445(2009)135:3(301) [22] 郑桥. 浙江近海典型台风浪和寒潮浪的精细化数值模拟[D]. 杭州: 浙江大学, 2019.Zheng Qiao. Numerical simulation of typical typhoon waves and cold waves in Zhejiang adjacent seas with refined grids[D]. Hangzhou: Zhejiang University, 2019. [23] 季余, 朱业, 李莉, 等. 浙江沿海台风浪模式的参数适应性研究[J]. 海洋预报, 2023, 40(2): 22−31. doi: 10.11737/j.issn.1003-0239.2023.02.003Ji Yu, Zhu Ye, Li Li, et al. Study on the parameters adaptability of typhoon wave model in Zhejiang coastal area[J]. Marine Forecasts, 2023, 40(2): 22−31. doi: 10.11737/j.issn.1003-0239.2023.02.003 [24] 邱锡鹏. 神经网络与深度学习[M]. 北京: 机械工业出版社, 2020.Qiu Xipeng. Neural Networks and Deep Learning[M]. Beijing: China Machine Press, 2020. -
计量
- 文章访问数: 88
- HTML全文浏览量: 36
- PDF下载量: 30
- 被引次数: 0
