基于观测和再分析数据的LSTM深度神经网络沿海风速预报应用研究

王国松; 王喜冬; 侯敏; 齐义泉; 宋军; 刘克修; 吴新荣; 白志鹏

doi:10.3969/j.issn.0253-4193.2020.01.008

基于观测和再分析数据的LSTM深度神经网络沿海风速预报应用研究

doi: 10.3969/j.issn.0253-4193.2020.01.008

王国松^{1, 2,},
王喜冬¹,
侯敏^3, ,,
齐义泉¹,
宋军⁴,
刘克修²,
吴新荣²,
白志鹏⁵

1.
河海大学海洋学院，江苏南京 210098
2.
国家海洋信息中心，天津 300171
3.
天津市滨海新区气象局，天津 300457
4.
大连海洋大学海洋科技与环境学院应用海洋研究所，辽宁大连 116023
5.
61741部队，北京 100094

基金项目: 国家重点研发计划（2016YFC1401903）；国家自然科学基金（41776004）；河海大学中央高校基本科研业务费（2016B12514）。

详细信息

作者简介:
王国松（1987—），男，天津市人，助研，主要从事数值模拟与资料处理研究。E-mail：xifengbishu2110@163.com

通讯作者:
侯敏，工程师，主要从事海洋气象预报业务。E-mail：ande0604@163.com

中图分类号: P732.4
计量
- 文章访问数: 2034
- HTML全文浏览量: 80
- PDF下载量: 222
- 被引次数: 0
出版历程
- 收稿日期: 2019-01-15
- 修回日期: 2019-06-22
- 网络出版日期: 2021-04-21
- 刊出日期: 2020-01-25

Research on application of LSTM deep neural network on historical observation data and reanalysis data for sea surface wind speed forecasting

Wang Guosong^{1, 2
,},
Wang Xidong¹,
Hou Min^{3
, ,},
Qi Yiquan¹,
Song Jun⁴,
Liu Kexiu²,
Wu Xinrong²,
Bai Zhipeng⁵

1.
College of Oceanography, Hohai University, Nanjing 210098, China
2.
National Marine Data and Information Service, Tianjin 300171, China
3.
Tianjin Binhai New Area Meteorology Administration, Tianjin 300457, China
4.
Operational Oceanographic Institution, School of Marine Science and Environment Engineering, Dalian Ocean University, Dalian 116023, China
5.
61741 Troops, Beijing 100094, China

摘要

摘要: 基于海洋气象历史观测资料和再分析数据等，利用LSTM深度神经网络方法，开展在有监督学习情况下的海面风场短时预报应用研究。以中国近海5个代表站为研究区域，通过气象台站观测数据和ERA-Interim 6 h再分析数据构建数据集。选取21个变量作为预报因子，分别构建两个LSTM深度神经网络框架(OBS_LSTM和ALL_LSTM)。经与2017年WRF模式6 h预报结果对比分析，得出如下结论：构建的两个LSTM风速预报模型可以大幅降低风速预报误差，RMSE分别降低了41.3%和38.8%，MAE平均降低了43.0%和40.0%；风速误差统计和极端大风分析发现，LSTM模型能够抓住地形、短时大风和台风等敏感信息，对于大风过程预报结果明显优于WRF模式；两种LSTM模型对比发现，ALL_LSTM模型风速预报误差最小，具有很好的稳定性和鲁棒性，OBS_LSTM模型应用范围更广泛。
- 深度学习 /
- LSTM /
- 海面风速 /
- 短时预报 /
- WRF模式
Abstract: Based on historical meteorological observation data and reanalysis data, the application of LSTM (longs short-term memory) deep neural network in short-term forecasting of sea surface wind speed under supervised learning was studied. Five representative meteorological stations in the offshore were taken as research areas. Twenty-one variables, which has been quality control and preprocessing, were selected as the prediction factors, and two LSTM deep neural network frameworks (OBS_LSTM and ALL_LSTM) were constructed. The 6 h wind speed forecast of WRF model over two-nesting domains in 2017 was included to validate the real performance of the proposed model. The result indicate that, the LSTM wind speed forecast models could significantly reduce the forecast error, RMSE was reduced by 41.3% and 38.8%, and MAE was reduced by 43.0% and 40.0%, respectively; wind speed error statistics and strong wind events comparison shows that, LSTM model can grasp sensitive information such as topography and typhoon, and superior to WRF model. The ALL_LSTM model has the smallest prediction error, good stability and robustness, and OBS_LSTM model has a wider range of application.
- deep neural network /
- LSTM /
- sea surface wind speed /
- short-term forecasting /
- WRF model

HTML全文

图 1 气象观测台站分布（蓝点）与WRF模式嵌套区域

黑红点线为2017年影响汕头和西沙两站的台风最佳路径

Fig. 1 Locations of meteorological observation stations (blue points) and nested domain of WRF model

Black dot line is the best track of typhoon affecting Shantou and Xisha stations in 2017

下载: 全尺寸图片幻灯片

图 2 预处理后测试集主要变量序列分布

Fig. 2 Time series of main variable of the test set after preprocessing

下载: 全尺寸图片幻灯片

图 3 西沙站与观测风速相关性最高的9个变量热力图

Fig. 3 The heat map of highest correlation between 9 variables and observed wind speed at Xisha Station

下载: 全尺寸图片幻灯片

图 4 ALL_LSTM预报模型结构图

Fig. 4 ALL_LSTM prediction model structure diagram

下载: 全尺寸图片幻灯片

图 5 ALL_LSTM，OBS_LSTM和WRF模式日平均风速预报结果与台站观测资料对比

Fig. 5 Comparison of daily mean wind speed forecast results of ALL_LSTM, OBS_LSTM and WRF models with observational data

下载: 全尺寸图片幻灯片

表 1 中国近海5个气象台站基本信息

Tab. 1 Basic information of the five meteorological stations in China offshore

站号	站名	缩写	站类	纬度	经度	观测场海拔高度/m
54662	大连	DL	基准站	38°54'N	121°38'E	91.5
54857	青岛	QD	基本站	36°04'N	120°20'E	76.0
58265	吕泗	LS	基准站	32°04'N	121°36'E	5.5
59316	汕头	ST	基准站	23°24'N	116°41'E	2.9
59981	西沙	XS	基准站	16°50'N	112°20'E	4.7

下载: 导出CSV

表 2 WRF模式设置

Tab. 2 Specification of the WRF model

区域与选项	具体设置
区域与分辨率	三重嵌套，Lambert conformal投影，中心点（42°N，115°E）
区域与分辨率	D01:水平分辨率 30 km	D02:水平分辨率 10 km	D03:水平分辨率 3.3 km	D04:水平分辨率 3.3 km	D05:水平分辨率 3.3 km
垂直分辨率	44η层
输出时间间隔	6 h
边界层方案	Lin et al. scheme方案
积云方案	Kain-Fritsch方案
微物理方案	WSM 5-class 方案
辐射方案	长短波辐射方案:RRTMG方案
陆面过程	Noah陆面模式
背景误差	CV5
热启动	是

下载: 导出CSV

表 3 验证集风速误差统计

Tab. 3 Compare the performance of 6 h wind speed on validation sets

站位	误差统计	ALL_LSTM 模型	OBS_LSTM 模型	WRF 模型
大连	均方根误差/m·s⁻¹	1.44	1.51	1.87
	绝对误差/m·s⁻¹	1.09	1.17	1.43
	相关系数	0.55	0.48	0.70
青岛	均方根误差/m·s⁻¹	1.47	1.53	3.00
	绝对误差/m·s⁻¹	1.10	1.15	2.33
	相关系数	0.60	0.53	0.11
吕泗	均方根误差/m·s⁻¹	1.15	1.21	1.17
	绝对误差/m·s⁻¹	0.90	0.94	0.93
	相关系数	0.64	0.58	0.73
汕头	均方根误差/m·s⁻¹	0.74	0.77	1.13
	绝对误差/m·s⁻¹	0.58	0.61	0.85
	相关系数	0.55	0.50	0.51
西沙	均方根误差/m·s⁻¹	1.16	1.20	2.99
	绝对误差/m·s⁻¹	0.88	0.92	2.44
	相关系数	0.73	0.71	0.70
各地平均	均方根误差/m·s⁻¹	1.19	1.24	2.03
	绝对误差/m·s⁻¹	0.91	0.96	1.60
	相关系数	0.61	0.56	0.55
注：最小均方根误差、最小绝对误差和最大相关系数分别用粗体表示。

下载: 导出CSV

参考文献(61)

[1]	陈德花, 刘铭, 苏卫东, 等. BP人工神经网络在MM5预报福建沿海大风中的释用[J]. 暴雨灾害, 2010, 29(3): 263−267. doi: 10.3969/j.issn.1004-9045.2010.03.010 Chen Dehua, Liu Ming, Su Weidong, et al. Interpretation and application of BP artificial neural network in MM5 model forecasting gale for coastal regions of Fujian province[J]. Torrential Rain and Disasters, 2010, 29(3): 263−267. doi: 10.3969/j.issn.1004-9045.2010.03.010
[2]	刘伟东, 扈海波, 程丛兰, 等. 灰色关联度方法在大风和暴雨灾害损失评估中的应用[J]. 气象科技, 2007, 35(4): 563−566. doi: 10.3969/j.issn.1671-6345.2007.04.023 Liu Weidong, Hu Haibo, Cheng Conglan, et al. Application of grey correlation degree to disaster loss evaluation of strong wind and heavy rainfall[J]. Meteorological Science and Technology, 2007, 35(4): 563−566. doi: 10.3969/j.issn.1671-6345.2007.04.023
[3]	曲海涛, 刘学萍. 黄渤海大风统计分析和预报方法[J]. 辽宁气象, 2002, 18(4): 11−12. Qu Haitao, Liu Xueping. Statistical analysis and forecasting method of gale in Huang-Bohai Sea[J]. Liaoning Meteorological Quarterly, 2002, 18(4): 11−12.
[4]	吴庆丽, 尹福杰, 陈敏, 等. “一一·二四”海难渤海风场的数值模拟[J]. 自然灾害学报, 2002, 11(1): 85−90. doi: 10.3969/j.issn.1004-4574.2002.01.014 Wu Qingli, Yin Fujie, Chen Min, et al. Wind field numerical simulation for a great shipwreck in Bohai Sea on November 24, 1999[J]. Journal of Natural Disasters, 2002, 11(1): 85−90. doi: 10.3969/j.issn.1004-4574.2002.01.014
[5]	尹尽勇, 刘涛, 张增海, 等. 冬季黄渤海大风天气与渔船风损统计分析[J]. 气象, 2009, 35(6): 90−95. doi: 10.7519/j.issn.1000-0526.2009.06.012 Yin Jinyong, Liu Tao, Zhang Zenghai, et al. Statistical analysis of the weather system types causing strong winds and fishery boat windage loss accidents in Bohai Sea and Yellow Sea in winter[J]. Meteorological Monthly, 2009, 35(6): 90−95. doi: 10.7519/j.issn.1000-0526.2009.06.012
[6]	Sutskever I, Vinyals O, Le Q V. Sequence to sequence learning with neural networks[C]//Advances in Neural Information Processing Systems 27 (NIPS). 2014: 1-5.
[7]	Zeiler M D, Fergus R. Visualizing and understanding convolutional networks[C]//European Conference on Computer. Cham: Springer, 2014.
[8]	Glorot X, Bordes A, Bengio Y. Deep sparse rectifier neural networks[J]. Journal of Mechine Learing Research, 2010(1): 315-323.
[9]	Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems. Lake Tahoe, Nevada: Curran Associates Inc, 2012.
[10]	Hinton G, Deng L, Yu D, et al. Deep neural networks for acoustic modeling in speech recognition: the shared views of four research groups[J]. IEEE Signal Processing Magazine, 2012, 29(6): 82−97. doi: 10.1109/MSP.2012.2205597
[11]	Russakovsky O, Deng J, Su H, et al. ImageNet large scale visual recognition challenge[J]. International Journal of Computer Vision, 2015, 115(3): 211−252. doi: 10.1007/s11263-015-0816-y
[12]	Erhan D, Bengio Y, Courville A, et al. Why does unsupervised pre-training help deep learning?[J]. Journal of Machine Learning Research, 2010, 11(3): 625−660.
[13]	Schmidhuber J. Deep learning in neural networks: an overview[J]. Neural Networks, 2015, 61: 85−117. doi: 10.1016/j.neunet.2014.09.003
[14]	Lecun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278−2324. doi: 10.1109/5.726791
[15]	Hinton G E, Salakhutdinov R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504−507. doi: 10.1126/science.1127647
[16]	Silver D, Huang A, Maddison C J, et al. Mastering the game of Go with deep neural networks and tree search[J]. Nature, 2016, 529(7587): 484−489. doi: 10.1038/nature16961
[17]	Hochreiter S, Schmidhuber J. LSTM can solve hard long time lag problems[C]//Proceedings of the 9th International Conference on Neural Information Processing Systems. Cambridge, MA, USA: MIT Press, 1997.
[18]	Gers F A, Schmidhuber J, Cummins F. Learning to Forget: Continual Prediction with LSTM[J]. Neural Computation, 2014, 12(10): 2451−2471.
[19]	Graves A. Long short-term memory[M]//Supervised Sequence Labelling with Recurrent Neural Networks. Berlin Heidelberg: Springer, 2012.
[20]	Graves A. Supervised Sequence Labelling with Recurrent Neural Networks[M]. Berlin Heidelberg: Springer, 2012.
[21]	Hochreiter S, Schmidhuber J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735−1780. doi: 10.1162/neco.1997.9.8.1735
[22]	Cheng Lilin, Zang Haixiang, Ding Tao, et al. Ensemble recurrent neural network based probabilistic wind speed forecasting approach[J]. Energies, 2018, 11(8): 1958. doi: 10.3390/en11081958
[23]	Kumar N K, Savitha R, Al Mamun A. Regional ocean wave height prediction using sequential learning neural networks[J]. Ocean Engineering, 2017, 129: 605−612. doi: 10.1016/j.oceaneng.2016.10.033
[24]	Wei Jun, Jiang Guoqing, Liu Xin. Parameterization of typhoon-induced ocean cooling using temperature equation and machine learning algorithms: an example of typhoon Soulik (2013)[J]. Ocean Dynamics, 2017, 67(9): 1179−1193. doi: 10.1007/s10236-017-1082-z
[25]	Haque A U, Mandal P, Meng Julian, et al. Wind speed forecast model for wind farm based on a hybrid machine learning algorithm[J]. International Journal of Sustainable Energy, 2015, 34(1): 38−51. doi: 10.1080/14786451.2013.826224
[26]	Klein B, Wolf L, Afek Y. A dynamic convolutional layer for short rangeweather prediction[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston, MA, USA: IEEE, 2015.
[27]	Kambekar A R, Deo M C. Real time wave forecasting using wind time history and genetic programming[J]. Ocean Engineering, 2014, 5(4): 249−259.
[28]	Najeebullah, Zameer A, Khan A, et al. Machine learning based short term wind power prediction using a hybrid learning model[J]. Computers & Electrical Engineering, 2014, 45: 122−133.
[29]	Shi Xingjian, Chen Zhourong, Wang Hao, et al. Convolutional LSTM network: A machine learning approach for precipitation nowcasting[C]//Advances in Neural Information Processing Systems 28 (NIPS). 2015: 802-810.
[30]	Jiang Guoqing, Xu Jing, Wei Jun. A deep learning algorithm of neural network for the parameterization of typhoon-ocean feedback in typhoon forecast models[J]. Geophysical Research Letters, 2018, 45(8): 3706−3716. doi: 10.1002/2018GL077004
[31]	王辉, 刘娜, 逄仁波, 等. 全球海洋预报与科学大数据[J]. 科学通报, 2015, 60(5/6): 479−484. Wang Hui, Liu Na, Pang Renbo, et al. Global ocean forecasting and scientific big data[J]. Chinese Science Bulletin, 2015, 60(5/6): 479−484.
[32]	Alexandre E, Cuadra L, Nieto-Borge J C, et al. A hybrid genetic algorithm-extreme learning machine approach for accurate significant wave height reconstruction[J]. Ocean Modelling, 2015, 92: 115−123. doi: 10.1016/j.ocemod.2015.06.010
[33]	Casas-Prat M, Wang Xiaolan, Sierra J P. A physical-based statistical method for modeling ocean wave heights[J]. Ocean Modelling, 2014, 73: 59−75. doi: 10.1016/j.ocemod.2013.10.008
[34]	Jain P, Deo M C, Latha G, et al. Real time wave forecasting using wind time history and numerical model[J]. Ocean Modelling, 2011, 36(1/2): 26−39.
[35]	Wang X L, Feng Yang, Swail V R. Changes in global ocean wave heights as projected using multimodel CMIP5 simulations[J]. Geophysical Research Letters, 2014, 41(3): 1026−1034. doi: 10.1002/2013GL058650
[36]	李泽椿, 毕宝贵, 金荣花, 等. 近10年中国现代天气预报的发展与应用[J]. 气象学报, 2014, 72(6): 1069−1078. Li Zechun, Bi Baogui, Jin Ronghua, et al. The development and application of the modern weather fore-cast in China for the recent 10 years[J]. Acta Meteorologica Sinica, 2014, 72(6): 1069−1078.
[37]	Warner T T. Numerical Weather and Climate Prediction[M]. Cambridge: Cambridge University Press, 2011.
[38]	Bauer P, Thorpe A, Brunet G. The quiet revolution of numerical weather prediction[J]. Nature, 2015, 525(7567): 47−55. doi: 10.1038/nature14956
[39]	Li Dele, Storch H, Geyer B. High-resolution wind hindcast over the Bohai Sea and the Yellow Sea in East Asia: Evaluation and wind climatology analysis[J]. Journal of Geophysical Research: Atmospheres, 2016, 121(1): 111−129. doi: 10.1002/2015JD024177
[40]	Zhang Dalin, Fritsch J M. A case study of the sensitivity of numerical simulation of mesoscale convective systems to varying initial conditions[J]. Monthly Weather Review, 1986, 114(12): 2418−2431. doi: 10.1175/1520-0493(1986)114<2418:ACSOTS>2.0.CO;2
[41]	杨悦, 高山红. 黄海海雾 WRF 数值模拟中垂直分辨率的敏感性研究[J]. 气象学报, 2016, 74(6): 974−988. Yang Yue, Gao Shanhong. Sensitivity study of vertical resolution in WRF numerical simulation for sea fog over the Yellow Sea[J]. Acta Meteorologica Sinica, 2016, 74(6): 974−988.
[42]	Cohen A E, Cavallo S M, Coniglio M C, et al. A review of planetary boundary layer parameterization schemes and their sensitivity in simulating southeastern U. S. cold season severe weather environments[J]. Weather and Forecasting, 2015, 30(3): 591−612. doi: 10.1175/WAF-D-14-00105.1
[43]	Draxl C, Hahmann A N, Peña A, et al. Evaluating winds and vertical wind shear from Weather Research and Forecasting model forecasts using seven planetary boundary layer schemes[J]. Wind Energy, 2014, 17(1): 39−55. doi: 10.1002/we.1555
[44]	Peña A, Hahmann A N, Hasager C B, et al. South Baltic wind atlas: south Baltic offshore wind energy regions project[J]. Risø Dtu National Laboratory for Sustainable Energy, 2011.
[45]	常蕊, 朱蓉, 周荣卫, 等. 高分辨率合成孔径雷达卫星反演风场资料在中国近海风能资源评估中的应用研究[J]. 气象学报, 2014, 72(3): 606−613. Chang Rui, Zhu Rong, Zhou Rongwei, et al. An application of high resolution SAR wind retrievals to off-shore wind resources assessment in China[J]. Acta Meteorologica Sinica, 2014, 72(3): 606−613.
[46]	Durán-Rosal A M, Hervás-Martínez C, Tallón-Ballesteros A J, et al. Massive missing data reconstruction in ocean buoys with evolutionary product unit neural networks[J]. Ocean Engineering, 2016, 117: 292−301. doi: 10.1016/j.oceaneng.2016.03.053
[47]	Nitsure S P, Londhe S N, Khare K C. Wave forecasts using wind information and genetic programming[J]. Ocean Engineering, 2012, 54(4): 61−69.
[48]	Malekmohamadi I, Ghiassi R, Yazdanpanah M J. Wave hindcasting by coupling numerical model and artificial neural networks[J]. Ocean Engineering, 2008, 35(3/4): 417−425.
[49]	Gao Shanhong, Wu Wei, Zhu Leilei, et al. Detection of nighttime sea fog/stratus over the Huang-hai Sea using MTSAT-1R IR data[J]. Acta Oceanologica Sinica, 2009, 28(2): 23−35.
[50]	王国松, 高山红, 吴彬贵, 等. 我国近海风能资源分布特征分析[J]. 海洋科学进展, 2014, 32(1): 21−29. doi: 10.3969/j.issn.1671-6647.2014.01.003 Wang Guosong, Gao Shanhong, Wu Bingui, et al. Distribution features of wind energy resources in the offshore areas of China[J]. Advances in Marine Science, 2014, 32(1): 21−29. doi: 10.3969/j.issn.1671-6647.2014.01.003
[51]	Berrisford P, Kållberg P, Kobayashi S, et al. Atmospheric conservation properties in ERA-Interim[J]. Quarterly Journal of the Royal Meteorological Society, 2011, 137(659): 1381−1399. doi: 10.1002/qj.864
[52]	Dee D P, Uppala S M, Simmons A J, et al. The ERA-Interim reanalysis: configuration and performance of the data assimilation system[J]. Quarterly Journal of the Royal Meteorological Society, 2011, 137(656): 553−597. doi: 10.1002/qj.828
[53]	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
[54]	Wang G S, Hou M, Li Y, et al. Homogenization methods for the sea surface temperature data over the South China Seas[J]. IOP Conference Series: Earth and Environmental Science, 2017, 52(1): 012046.
[55]	李琰, 王国松, 范文静, 等. 中国沿海海表温度均一性检验和订正[J]. 海洋学报, 2018, 40(1): 17−28. doi: 10.3969/j.issn.0253-4193.2018.01.003 Li Yan, Wang Guosong, Fan Wenjing, et al. The homogeneity study of the sea surface temperature data along the coast of the China Seas[J]. Haiyang Xuebao, 2018, 40(1): 17−28. doi: 10.3969/j.issn.0253-4193.2018.01.003
[56]	Lee M S, Kuo Y H, Barker D M, et al. Incremental analysis updates initialization technique applied to 10-km MM5 and MM5 3DVAR[J]. Monthly Weather Review, 2006, 134(5): 1389−1404. doi: 10.1175/MWR3129.1
[57]	李欢, 王国松, 李文善, 等. 中国近海海上溢油一体化预测预警系统研究——系统介绍[J]. 海洋信息, 2018, 33(4): 44−49. Li Huan, Wang Guosong, Li Wenshan, et al. Introduction to the China integrated offshore oil spill forecasting and early warning system[J]. Marine Information, 2018, 33(4): 44−49.
[58]	高山红, 齐伊玲, 张守宝, 等. 利用循环3DVAR改进黄海海雾数值模拟初始场Ⅰ: WRF数值试验[J]. 中国海洋大学学报, 2010, 40(10): 1−9. Gao Shanhong, Qi Yiling, Zhang Shoubao, et al. Initial conditions improvement of sea fog numerical modeling over the yellow sea by using cycling 3DVAR Part Ⅰ: WRF numerical experiments[J]. Periodical of Ocean University of China, 2010, 40(10): 1−9.
[59]	高佳, 牟林, 王国松, 等. 马航MH370残骸漂移轨迹分析和预测[J]. 科学通报, 2016, 61(21): 2409−2418. Gao Jia, Mu Lin, Wang Guosong, et al. Drift analysis and prediction of debris from Malaysia Airlines flight MH370[J]. Chinese Science Bulletin, 2016, 61(21): 2409−2418.
[60]	李磊, 张立杰, 陈柏纬. 基于CFD技术的陡峭山体风场模拟方法研究[J]. 气象学报, 2016, 74(4): 613−622. Li Lei, Zhang Lijie, Chen Baiwei. The application of CFD techniques on the wind field simulation over steep mountains: A method study[J]. Acta Meteorologica Sinica, 2016, 74(4): 613−622.
[61]	Lecun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521(7553): 436−444. doi: 10.1038/nature14539