基于深度学习的潮位预报订正技术研究-海洋预报

基于深度学习的潮位预报订正技术研究

单位：1. 国家海洋局东海预报中心, 上海 200136;
2. 上海市海洋监测预报中心, 上海 200062

分类号：P731.34

出版年·卷·期（页码）：2022·39·第二期（70-79）

摘要：

引入长短期记忆神经网络（LSTM）深度学习先进算法，通过3折滑动时间序列交叉验证结合网格搜索方法确定最佳参数，构建了吴淞口潮位预报订正深度学习模型。结合风、压气象参数，对调和分析的预报潮位进行订正，得到更为准确的订正潮位，并与非线性自回归动态神经网络（NARX）浅层学习预报订正结果对比。结果表明：120 h预报潮位经LSTM模型订正后的均方根误差为0.102 m，平均绝对误差为0.084 m，订正后误差降低了52.8%；前72 h预报潮位经LSTM模型订正后误差降低了57.3%。对比发现，LSTM模型在短期和中长期潮位预报订正中均有较好表现，NARX模型在短期预报订正中表现出色。“海神”台风风暴潮过程期间，120 h、前72 h和“主振”48 h特征时段预报潮位经LSTM模型订正后，均方根误差为0.114～0.119 m，平均绝对误差为0.100～0.102 m。

In this paper, a deep learning model to correct the tide level prediction in Wusongkou area is established based on the advanced long short term memory (LSTM) neural network and the best parameters determined by 3-fold cross validation of sliding time series and grid search method. The forecast tide level of harmonic analysis is corrected with higher accuracy using wind and pressure meteorological parameters, and is compared with the correction result of the nonlinear autoregressive exogenous (NARX) dynamic neural network. The results show that the root mean square error (RMSE) and the mean absolute error (MAE) of the 120 h forecast tide level corrected by LSTM model is 0.102 m and 0.084 m, respectively, and the error is reduced by 52.8%. Moreover, the error of 72 h forecast tide level corrected by LSTM model is reduced by 57.3%. By comparison, it is found that the LSTM model performs well in the correction of short-term and medium-term and long-term tide level forecast, while the NARX model performs well only in the correction of short-term forecast. During the storm surge process caused by typhoon Haishen, the RMSE and MAE of the forecast tide level corrected by LSTM model for the characteristic periods of 120 h, the first 72 h and 48 h are 0.114～0.119 m and 0.100～0.102 m.

参考文献：

[1] Doodson A T. The analysis and predictions of tides in shallow water [J]. International Hydrographic Review, 1957, 33(1): 85-126.
[2] 叶琳. 潮汐预报与服务[J]. 海洋预报服务, 1985(1): 74-76. Ye L. Tide forecast and service[J]. Marine Forecast Service, 1985(1): 74-76.
[3] 方国洪, 于克俊. 浅水港口潮汐预报的一个方法[J]. 海洋与湖沼, 1981, 12(5): 383-390. Fang G H, Yu K J. A method for predicting the tides of shallowwater ports[J]. Oceanologia et Limnologia Sinica, 1981, 12(5): 383-390.
[4] 黄辰虎, 刘雁春, 暴景阳, 等. 相关法在提高浅水港口潮汐预报质量方面的应用[J]. 海洋测绘, 2004, 24(3): 9-11. Huang C H, Liu Y C, Bao J Y, et al. Increasing the quality of tide prediction in Netric port with correlation[J]. Hydrographic Surveying and Charting, 2004, 24(3): 9-11.
[5] Box G E P, Jenkins G M, Reinsel G C, et al. Time series analysis: forecasting and control, 5th edition[J]. Journal of the Operational Research Society, 2015, 22(2): 199-201.
[6] Yen P H, Jan C D, Lee Y P, et al. Application of Kalman filter to short-term tide level prediction[J]. Journal of Waterway, Port, Coastal, and Ocean Engineering, 1996, 122(5): 226-231.
[7] 柳成, 尹建川. 一种高精度的短期潮汐预报模型[J]. 上海海事大学学报, 2016, 37(3): 74-80. Liu C, Yin J C. A high-accuracy short-term tide prediction model [J]. Journal of Shanghai Maritime University, 2016, 37(3): 74-80.
[8] Meena B L, Agrawal J D. Tidal level forecasting using ANN[J]. Procedia Engineering, 2015, 116: 607-614.
[9] Lee T L, Jeng D S. Application of artificial neural networks in tideforecasting[J]. Ocean Engineering, 2002, 29(9): 1003-1022.
[10] LeCun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521(7553): 436-444.
[11] Schmidhuber J. Deep learning in neural networks: an overview[J]. Neural Networks, 2015, 61: 85-117.
[12] Salim A M, Dwarakish G S, Liju K V, et al. Weekly prediction of tides using neural networks[J]. Procedia Engineering, 2015, 116: 678-682.
[13] 杨兴果. 钱塘江河口咸潮入侵预警研究[D]. 杭州: 浙江大学, 2014. Yang X G. Saltwater intrusion warning research in Qiantang estuary[D]. Hangzhou: Zhejiang University, 2014.
[14] Le X H, Ho H V, Lee G, et al. Application of long short-term memory (LSTM) neural network for flood forecasting[J]. Water, 2019, 11(7): 1387.
[15] Ding Y K, Zhu Y L, Feng J, et al. Interpretable spatio-temporal attention LSTM model for flood forecasting[J]. Neurocomputing, 2020, 403: 348-359.
[16] 刘亚新, 樊启祥, 尚毅梓, 等. 基于LSTM神经网络的水电站短期水位预测方法[J]. 水利水电科技进展, 2019, 39(2): 56-60, 78. Liu Y X, Fan Q X, Shang Y Z, et al. Short-term water level prediction method for hydropower station based on LSTM neural network[J]. Advances in Science and Technology of Water Resources, 2019, 39(2): 56-60, 78.
[17] Leontaritis I J, Billings S A. Input-output parametric models for non-linear systems Part I: deterministic non-linear systems[J]. International Journal of Control, 1985, 41(2): 303-328.
[18] Billings S A. Models for linear and nonlinear systems[M]//Billings S A. Nonlinear System Identification: NARMAX Methods in the Time, Frequency, and Spatio-Temporal Domains. Hoboken: John Wiley & Sons, Ltd., 2013.
[19] Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780.
[20] Greff K, Srivastava R K, Koutník J, et al. LSTM: a search space odyssey[J]. IEEE Transactions on Neural Networks and Learning Systems, 2017, 28(10): 2222-2232.
[21] 方正飞. 智能网格化推动海洋预报更精确[N]. 中国海洋报, 2017-06-16(001). Fang Z F. Intelligent grid promotes ocean forecast more accurate [N]. China Ocean News, 2017-06-16(001).
[22] Diaconescu E. The use of NARX neural networks to predict chaotic time series[J]. WSEAS Transactions on Computer Research, 2008, 3(3): 182-191.
[23] 陈宗镛. 潮汐学[M]. 北京: 科学出版社, 1980. Chen Z Y. Tidology[M]. Beijing: Science Press, 1980.
[24] 徐文龙, 姜海燕, 张翼翔. 吴淞口警戒区通航风险分析[J]. 水运管理, 2020, 42(1): 34-36. Xu W L, Jiang H Y, Zhang Y X. Analysis on navigation risk in Wusongkou precautionary area[J]. Shipping Management, 2020, 42(1): 34-36.
[25] 陈吉余, 沈焕庭, 恽才兴, 等. 长江河口动力过程和地貌演变[M]. 上海: 上海科学技术出版社, 1990. Chen J Y, Shen H T, Yun C X, et al. Dynamic process and morphological evolution in the Changjiang Estuary[M]. Shanghai: Shanghai Scientific and Technical Publishers, 1990.
[26] 陈杰. 短期潮位预报分析探讨[C]//中国航海学会航标专业委员会测绘学组2008年学术研讨会论文集. 北京: 中国航海学会, 2008: 478-487. Chen J. Analysis and discussion of short-term tidal level forecast [C]//Proceedings of Academic Exchange at the Symposium of Surveying and Mapping Group, Navigation Mark Professional Committee. Beijing: Chinese Nautical Association, 2008: 478- 487.
[27] 吴喜之, 刘苗. 应用时间序列分析R软件陪同[M]. 2版. 北京: 机械工业出版社, 2018. Wu X Z, Liu M. Applied time series analysis with R[M]. 2nd ed. Beijing: China Machine Press, 2018.
[28] Ribeiro G H T, De M Neto P S G, Cavalcanti G D C, et al. Lag selection for time series forecasting using Particle Swarm Optimization[C]//The 2011 International Joint Conference on Neural Networks. San Jose: IEEE, 2011: 2437-3444.
[29] Hermans M, Schrauwen B. Training and analyzing deep recurrent neural networks[C]//Proceedings of the 26th International Conference on Neural Information Processing Systems. Lake Tahoe: ACM, 2013: 190-198.
[30] Reimers N, Gurevych I. Optimal hyperparameters for deep LSTMnetworks for sequence labeling tasks[J]. arXiv: 1707.06799, 2017.
[31] 崔巍, 顾冉浩, 陈奔月, 等. BP与LSTM神经网络在福建小流域水文预报中的应用对比[J]. 人民珠江, 2020, 41(2): 74-84. Cui W, Gu R H, Chen B Y, et al. Comparison of BP and LSTM neural network for hydrologic forecasting of a small watershed in Fujian[J]. Pearl River, 2020, 41(2): 74-84.
[32] Abbasimehr H, Shabani M, Yousefi M. An optimized model using LSTM network for demand forecasting[J]. Computers & Industrial Engineering, 2020, 143: 106435.
[33] 杜维仲, 王硕. 基于LSTM的辅助动力装置系统辨识与仿真[J]. 计算机测量与控制, 2020, 28(2): 157-161. Du W Z, Wang S. System identification and simulation of auxiliary power plant based on LSTM[J]. Computer Measurement & Control, 2020, 28(2): 157-161.
[34] 盛敏, 李兰. 基于动作主视图和LSTM网络模型的人体行为识别[J]. 安庆师范大学学报(自然科学版), 2020, 26(1): 73-76. Sheng M, Li L. Human behavior recognition based on LSTM network model and front action view[J]. Journal of Anqing Normal University (Natural Science Edition), 2020, 26(1): 73- 76.

服务与反馈：

【文章下载】【发表评论】【查看评论】【加入收藏】