基于人工神经网络的台风浪高快速计算方法

doi:10.11978/2019089

摘要/Abstract

摘要：

采用2010—2017年南海5个浮标波高观测资料和中国气象局热带气旋最佳路径集中的热带气旋参数, 基于前馈型误差反向传播(Forward Feedback Back Propagation, FFBP)神经网络(Artificial Neural Network, ANN)方法, 分别建立了各浮标站的台风浪高快速计算模型。研究显示, 基于热带气旋中心坐标、中心最低气压、近中心最大风速、热带气旋中心与浮标之间的距离和方位4个参数建立的神经网络模型经反复训练后, 模型输出结果可以很好地拟合观测数据, 各浮标有效波高计算值与观测值的均方根误差小于0.3m, 平均相对误差为5.78%~7.23%, 相关系数大于0.9, 属高度相关。独立测试结果显示, “山竹”( 国际编号: 1822)影响期间有效波高最大值的神经网络模型预报结果与观测值基本吻合, 相对误差为-31.06%~0.98%, 但计算的最大值出现时间和观测情况不完全一致。该计算方法可应用于热带气旋影响期间的有效波高最大值计算, 因而在海洋工程领域和海洋预报领域具有应用前景。

关键词: 台风浪高, 人工神经网络模型, 权值, 热带气旋参数, 快速计算

Abstract:

Based on the wave height observation data of five buoys and parameters from the CMA Tropical Cyclone Database for the South China Sea from 2010 to 2017, fast calculation models of tropical cyclone-generated wave height are established by using Forward Feedback Back Propagation (FFBP) Artificial Neural Network (ANN). The results show that significant wave height calculated by the ANN model based on the tropical cyclone central location, minimum sea level pressure (MSLP), maximum sustained wind (MSW), the distance and azimuth between tropical cyclone center and buoy can well fit the observations. The root mean square errors between the calculation and observation of significant wave height are less than 0.3 m, the average relative errors are between 5.78% and 7.23%, and the correlation coefficients are greater than 0.9. Test results show that the model calculation of the maxima of significant wave height basically coincides with the observation, the relative errors being between -31.06% and 0.98%; however, when the calculated maxima of significant wave height appear is not in agreement with observation during tropical cyclone Mangkhut ( International number ID: 1822). The method introduced in this paper can be applied to calculating the maximum of significant wave height during tropical cyclone for ocean engineering and marine forecast.

Key words: tropical cyclone wave height, Artificial Neural Network Model, weight, tropical cyclone parameters, fast calculation

中图分类号:

P731.33

周水华, 洪晓, 梁昌霞, 江丽芳. 基于人工神经网络的台风浪高快速计算方法[J]. 热带海洋学报, 2020, 39(4): 25-33.

Shuihua ZHOU, Xiao Hong, Changxia LIANG, Lifang JIANG. A method of tropical cyclone wave height calculation based on Artificial Neural Network[J]. Journal of Tropical Oceanography, 2020, 39(4): 25-33.

图/表 10

图1

表1

图2

表2

表3

表4

图3

图4

表5

图5

参考文献 21

[1]	陈鹏超, 李瑞杰, 李玉婷, 等, 2017. 不同风场模式下“灿鸿”台风浪数值模拟及应用[J]. 水道港口, 38(6):567-573.
	CHEN PENGCHAO, LI RUIJIE, LI YUTING, et al, 2017. Numerical simulation and application of typhoon CHAN-HOM under different wind field models[J]. Journal of Waterway and Harbor, 38(6):567-573 (in Chinese with English abstract).
[2]	陈希, 李妍, 沙文钰, 等, 2003. 人工神经网络技术在台风浪预报中的应用[J]. 海洋科学, 27(2):63-67.
	CHEN XI, LI YAN, SHA WENYU, et al, 2003. Application of the neural network in ocean wave forecast of typhoon[J]. Marine Science, 27(2):63-67 (in Chinese with English abstract).
[3]	丰鉴章, 李元智, 孙书敏, 等, 1987. 海岸工程中的海浪推算方法[M]. 北京: 海洋出版社: 89-153(in Chinese).
[4]	胡克林, 丁平兴, 朱首贤, 等, 2004. 江口附近海域台风浪的数值模拟——以鹿沙台风和森拉克台风为例[J]. 海洋学报, 26(5):23-33.
	HU KELIN, DING PINGXING, ZHU SHOUXIAN, et al, 2004. Numerical simulation of typhoon waves around the waters of the Changjiang estuary-a case study of Typhoon Rusa and typhoon Sinlaku[J]. Acta Oceanologica Sinica, 26(5):23-33 (in Chinese with English abstract).
[5]	梁连松, 李瑞杰, 丰青, 等, 2014. 舟山海域台风浪数值模拟[J]. 水道港口, 35(6):582-588.
	LIANG LIANSONG, LI RUIJIE, FENG QING, et al, 2014. Numerical simulation of typhoon wave in Zhoushan[J]. Journal of Waterway and Harbor, 35(6):582-588 (in Chinese with English abstract).
[6]	唐建, 史剑, 李训强, 等, 2013. 基于台风风场模型的台风浪数值模拟[J]. 海洋湖沼通报, (2):24-30.
	TANG JIAN, SHI JIAN, LI XUNQIANG, et al, 2013. Numerical simulation of typhoon waves with typhoon wind model[J]. Transactions of Oceanology and Limnology, (2):24-30 (in Chinese with English abstract).
[7]	王文娟, 梁昌霞, 李广敏, 等, 2011. 0915号台风“巨爵”不同风场资料对比研究[J]. 海洋预报, 28(2):9-17.
	WANG WENJUAN, LIANG CHANGXIA, LI GUANGMIN, et al, 2011. Studies on sea surface wind fields of TY0915(Koppu) derived from various methods[J]. Marine Forecasts, 28(2):9-17 (in Chinese with English abstract).
[8]	王锡侯, 宋有强, 1982. 台风波浪推算方法——计算推算点深水最大风浪要素的建议方法[J]. 海洋通报, ( 3):9-16 (in Chinese).
[9]	闻斌, 汪鹏, 万雷, 等, 2008. 中国近海海域台风浪模拟试验[J]. 海洋通报, 27(3):1-6.
	WEN BIN, WANG PENG, WAN LEI, et al, 2008. Experiments on the simulation of typhoon waves in the inshore area of China Sea[J]. Marine Science Bulletin, 27(3):1-6 (in Chinese with English abstract).
[10]	闻斌, 薛彦广, 赵建宇, 2014. 基于模型风场的台风浪数值模拟[J]. 海洋预报, 31(2):14-19.
	WEN BIN, XUE YANGUANG, ZHAO JIANYU, 2014. Numerical simulation of typhoon wave field based on typhoon wind field model[J]. Marine Forecasts, 31(2):14-19 (in Chinese with English abstract).
[11]	文圣常, 余宙文, 1985. 海浪理论与计算原理[M]. 北京: 科学出版社: 430-473(in Chinese).
[12]	许林之, 许富祥, 1989. 几种台风浪计算方法的比较[J]. 海洋预报, 6(3):66-71 (in Chinese).
[13]	杨万康, 尹宝树, 伊小飞, 等, 2017. 基于HOLLAND风场的台风浪数值计算[J]. 水利水运工程学报, (4):28-34.
	YANG WANKANG, YIN BAOSHU, YI XIAOFEI, et al, 2017. Numerical calculation and research of typhoon waves based on Holland wind field[J]. Hydro-Science and Engineering, (4):28-34 (in Chinese with English abstract).
[14]	张志旭, 齐义泉, 施平, 等, 2003. 最优化插值同化方法在预报南海台风浪中的应用[J]. 热带海洋学报, 22(4):34-41.
	ZHANG ZHIXU, QI YIQUAN, SHI PING, et al, 2003. Application of an optimal interpolation wave assimilation method in South China Sea[J]. Journal of Tropical Oceanography, 22(4):34-41 (in Chinese with English abstract).
[15]	中华人民共和国交通部, 1998. JTJ213-1998 海港水文规范[S]. 北京: 人民交通出版社: 16-25(in Chinese).
[16]	周水华, 李远芳, 冯伟忠, 等, 2010. “0601”号台风控制下的广东近岸浪特征[J]. 海洋通报, 29(2):130-134.
	ZHOU SHUIHUA, LI YUANFANG, FENG WEIZHONG, et al, 2010. Wave characteristics dominated by typhoon named “Peal”[J]. Marine Science Bulletin, 29(2):130-134 (in Chinese with English abstract).
[17]	DESHMUKH A N, DEO M C, BHASKARAN P K, et al, 2016. Neural-network-based data assimilation to improve numerical ocean wave forecast[J]. IEEE Journal of Oceanic Engineering, 41(4):944-953, doi: 10.1109/JOE.2016.2521222. doi: 10.1109/JOE.2016.2521222
[18]	GOPINATH D I, DWARAKISH G S, 2015. Wave prediction using neural networks at New Mangalore Port along west coast of India[J]. Aquatic Procedia, 4:143-150. doi: 10.1016/j.aqpro.2015.02.020
[19]	LONDHE S N, SHAH S, DIXIT P R, et al, 2016. A coupled numerical and artificial neural network model for improving location specific wave forecast[J]. Applied Ocean Research, 59:483-491. doi: 10.1016/j.apor.2016.07.004
[20]	PUSCASU R M, 2014. Integration of artificial neural networks into operational ocean wave prediction models for fast and accurate emulation of exact nonlinear interactions[J]. Procedia Computer Science, 29:1156-1170. doi: 10.1016/j.procs.2014.05.104
[21]	RIZIANIZA I, AISJAH A S, 2015. Prediction of significant wave height in the Java Sea using artificial neural network[C]// 2015 International Seminar on Intelligent Technology and Its Applications. Surabaya, Indonesia: IEEE, doi: 10.1109/ISITIA.2015.7219944.

		QF305	QF306	QF307	QF308	SF304
观测时长(年份)		2010—2017	2010—2017	2015—2017	2010—2017	2010—2017
热带气旋数/个	超强台风	11	10	8	13	11
	强台风	11	8	3	14	9
	台风	4	3	2	5	4
	强热带风暴	7	9	4	11	10
	热带风暴	7	6	5	13	7
	热带低压	4	3	1	2	5
	合计	44	39	23	58	46
波高阈值/m		2.5	2.5	2.5	2.5	2.5
样本数/个		1573	1395	701	1623	1700

训练方案	输入层输入因子							输出层期望值
训练方案	C₀	P₀	W₀	R₁	R₀	(1010-P₀)	V₀	Hs
CASE1			√	√				√
CASE2		√	√	√				√
CASE3	√	√	√	√				√
CASE4	√	√	√					√
CASE5	√	√	√		√			√
CASE6	√		√	√		√		√
CASE7	√		√			√		√
CASE8	√		√	√		√	√	√

浮标名称	指标	CASE1	CASE2	CASE3	CASE4	CASE5	CASE6	CASE7	CASE8
QF305	num	421	321	9	31	12	9	11	9
	err	0.202	0.155	0.109	0.130	0.149	0.107	0.166	0.145
	r	0.884	0.912	0.939	0.927	0.916	0.940	0.905	0.915
QF306	num	381	406	10	8	13	14	8	10
	err	0.211	0.226	0.113	0.110	0.120	0.124	0.107	0.056
	r	0.893	0.886	0.944	0.945	0.940	0.940	0.947	0.963
QF307	num	13	13	9	8	11	9	8	10
	err	0.141	0.146	0.116	0.083	0.095	0.075	0.063	0.078
	r	0.888	0.884	0.924	0.936	0.926	0.942	0.952	0.954
QF308	num	12	61	12	8	12	10	26	11
	err	0.153	0.136	0.163	0.111	0.131	0.100	0.124	0.127
	r	0.871	0.886	0.862	0.909	0.891	0.900	0.897	0.909
SF304	num	23	10	10	8	14	10	17	10
	err	0.158	0.156	0.110	0.138	0.159	0.108	0.157	0.134
	r	0.879	0.880	0.918	0.895	0.880	0.920	0.880	0.926
平均	num	170.0	162.2	10.0	12.6	12.4	10.4	14.0	10.0
	err	0.173	0.164	0.122	0.114	0.131	0.103	0.123	0.108
	r	0.883	0.890	0.917	0.922	0.911	0.928	0.916	0.933

浮标名	样本数/个	偏差的平方/m²	相关系数R	均方根误差/m	平均相对偏差/%	热带气旋数/个
QF305	1573	0.109	0.94	0.33	7.23	44
QF306	1395	0.079	0.96	0.28	6.00	39
QF307	701	0.072	0.95	0.27	5.89	23
QF308	1623	0.097	0.92	0.31	6.58	58
SF304	1700	0.072	0.95	0.27	5.78	46

	QF305	QF306	QF307	QF308	SF304
观测值/m	9.4	6.4	8.0	10.3	10.2
模型计算值/m	9.8	5.4	7.0	7.1	10.3
误差/m	0.4	-1.0	-1.0	-3.1	0.1
相对误差/%	4.26	-15.63	-12.50	-31.06	0.98