.644. 智能系统学报 第12卷 1989,1(2):270-280 3 结束语 [11 ELMAN J L.Finding structure in time J].Cognitive 本文采用递归神经网络对风暴潮增水进行预 science,1990,14(2):179-211. [12]BOTVINICK MM,PLAUT D C.Short-term memory for 测。传统的BP神经网络只能用来提取静态数据之 serial order:a recurrent neural network model [J]. 间的关系,而由于增添了反馈连接,递归神经网络 Psychological review,2006,113(2):201-233. 就更加适合用于时序数据的处理。本文在真实的 [13]SAINATH T N,KINGSBURY B,SAON G,et al.Deep 数据集上进行了验证,结果表明,相对于BP神经网 convolutional neural networks for large-scale speech tasks 络,递归神经网络能更好地对风暴潮增水进行预 [J].Neural networks,2015,64:39-48. 测,误差精度更低。 [14]SARIKAYA R,HINTON G E,DEORAS A,et al. Application of deep belief networks for natural language 在实验中发现,随着预测时间的增加,递归神 understanding[J.IEEE transactions on audio,speech, 经网络预测误差会逐步加大,并且预测结果与真实 and language processing,2014,22(4):778-784. 测量值的相位差会越来越大。如何采取有效的措 [15]KRIZHEVSKY A,SUTSKEVER I,HINTON G E,et al. 施对这一现象进行缓解,是作者下一步需要研究的 ImageNet classification with deep convolutional neural 工作。 networks C ]//International Conference on Neural Information Processing Systems.Nevada,America,2012: 参考文献: 1097-1105. [16]SVOZIL D,KVASNICKA V,POSPICHAL J,et al. [1]王喜年.风暴潮预报知识讲座[J].海洋预报,2001,18 Introduction to multi-layer feed-forward neural networks (1):73-78 [J].Chemometrics and intelligent laboratory systems, [2]高清清,曹兵,高鑫鑫,等.南通沿海台风风暴潮分析及 1997,39(1):43-62. 其经验预报初探[J].海洋预报.2014,31(1):29-35. [17]HORNIK K,STINCHCOMBE M B,WHITE H,et al. GAO Qingqing,CAO Bing,GAO Xinxin,et al.Analysis of the Multilayer feedforward networks are universal approximators typhoon storm surge in the Nantong coastal zone and the [J].Neural networks,1989,2(5):359-366. forecasting formulas[J].Marine forecasts,2014,31(1):29-35. [18]RUMELHART D E,HINTON G E,WILLIAMS R J,et al. [3]曾德美.青岛港风暴潮经验统计预报[J].海洋预报 Learning representations by back-propagating errors[J]. 1992,9(3):66-73. Nature,1988,323(6088):533-536. ZENG Demei.A statistical forecasting of storm surge in 19]KOSMATOPOULOS E B.POLYCARPOU MM.CHRISTODOULOU qingdao harbor[J].Marine forecasts,1992,9(3):66-73. M A,et al.High-order neural network structures for [4]谢亚力,黄世昌.钱塘江河口风暴潮经验预报[J].海洋 identification of dynamical systems[J].IEEE transactions 预报,2006,23(1):54-58. on neural networks,1995,6(2):422-431. XIE Yali,HUANG Shichang.Effects on watercourse and [20]WERBOS P J.Backpropagation through time:what it does hydrological factor in downstrean nanxi river due to water and how to do it[J].Proceedings of the IEEE,1990,78 diversion[J].Marine forecasts,2006,23(1):54-58. (10):1550-1560. [5]朱建荣,朱首贤.ECOM模式的改进及在长江河口、杭州 作者简介: 湾及邻近海区的应用[J].海洋与湖沼,2003,34(4): 雷森,男,1992年生,博士研究生 364-373. 主要研究方向为图像处理、机器学习、 ZHU Jianrong,ZHU Shouxian.Improvement of the ECOM 遥感影像质量提升。 with application to the changjiang river estuary,Hangzhou bay and adjacent waters [J.Oceanologia Et limnologia sinica,2003.34(4):364-373. [6]黄世昌,李玉成,赵鑫,等.浙江沿海超强台风作用下 风暴潮增水数值分析[J].海洋工程,2008,26(3): 58-64. 史振威,男,1977年生,教授,博士 HUANG Shichang,LI Yucheng,ZHAO Xin,et al. 生导师,博士,主要研究方向为图像处 Numerical analysis of storm surge due to a super typhoon in 理、模式识别、机器学习、遥感影像处 coastal region of Zhejiang province [J].The ocean 理。发表SCI国际期刊检索论文70 engineering,2008,26(3):58-64. 余篇。 [7]王培涛,于福江,刘秋兴,等.福建沿海精细化台风风暴 潮集合数值预报技术研究及应用[J].海洋预报,2010 27(5):7-15. [8]LEE T L.Neural network prediction of a storm surge[J] 石天阳,男,1994年生,硕士研究 Ocean engineering,2006,33(3):483-494. 生,主要研究方向为机器学习和人工 [9]RAJASEKARAN S,GAYATHRI S,LEE T L,et al.Support 智能。 vector regression methodology for storm surge predictions[]. 0 cean engineering,2008,35(16):1578-1587 [10]WILIIAMS R J,ZIPSER D.A learning algorithm for continually running fully recurrent neural networks[J.Neural computation,3 结束语 本文采用递归神经网络对风暴潮增水进行预 测。 传统的 BP 神经网络只能用来提取静态数据之 间的关系,而由于增添了反馈连接,递归神经网络 就更加适合用于时序数据的处理。 本文在真实的 数据集上进行了验证,结果表明,相对于 BP 神经网 络,递归神经网络能更好地对风暴潮增水进行预 测,误差精度更低。 在实验中发现,随着预测时间的增加,递归神 经网络预测误差会逐步加大,并且预测结果与真实 测量值的相位差会越来越大。 如何采取有效的措 施对这一现象进行缓解,是作者下一步需要研究的 工作。 参考文献: [1]王喜年. 风暴潮预报知识讲座[ J]. 海洋预报, 2001, 18 (1): 73-78. [2]高清清,曹兵,高鑫鑫,等. 南通沿海台风风暴潮分析及 其经验预报初探[J]. 海洋预报, 2014, 31(1): 29-35. GAO Qingqing, CAO Bing, GAO Xinxin, et al. Analysis of the typhoon storm surge in the Nantong coastal zone and the forecasting formulas[J]. Marine forecasts, 2014, 31(1): 29-35. [3]曾德美. 青岛港风暴潮经验统计预报[ J]. 海洋预报, 1992, 9(3): 66-73. ZENG Demei. A statistical forecasting of storm surge in qingdao harbor[J]. Marine forecasts, 1992, 9(3): 66-73. [4]谢亚力, 黄世昌. 钱塘江河口风暴潮经验预报[ J]. 海洋 预报, 2006, 23(1): 54-58. XIE Yali, HUANG Shichang. Effects on watercourse and hydrological factor in downstrean nanxi river due to water diversion[J]. Marine forecasts, 2006, 23(1): 54-58. [5]朱建荣, 朱首贤. ECOM 模式的改进及在长江河口、杭州 湾及邻近海区的应用[ J]. 海洋与湖沼, 2003, 34( 4): 364-373. ZHU Jianrong, ZHU Shouxian. Improvement of the ECOM with application to the changjiang river estuary, Hangzhou bay and adjacent waters [ J ]. Oceanologia Et limnologia sinica, 2003, 34(4): 364-373. [6]黄世昌, 李玉成, 赵鑫, 等. 浙江沿海超强台风作用下 风暴潮增水数值分析 [ J]. 海洋工程, 2008, 26 ( 3): 58-64. HUANG Shichang, LI Yucheng, ZHAO Xin, et al. Numerical analysis of storm surge due to a super typhoon in coastal region of Zhejiang province [ J ]. The ocean engineering, 2008, 26(3): 58-64. [7]王培涛,于福江,刘秋兴,等. 福建沿海精细化台风风暴 潮集合数值预报技术研究及应用[ J]. 海洋预报,2010, 27(5): 7-15. [8]LEE T L. Neural network prediction of a storm surge[ J]. Ocean engineering, 2006, 33(3): 483-494. [9]RAJASEKARAN S, GAYATHRI S, LEE T L, et al. Support vector regression methodology for storm surge predictions[J]. Ocean engineering, 2008, 35(16): 1578-1587. [10]WILLIAMS R J, ZIPSER D. A learning algorithm for continually running fully recurrent neural networks[J]. Neural computation, 1989, 1(2): 270-280. [11] ELMAN J L. Finding structure in time [ J]. Cognitive science, 1990, 14(2):179-211. [12] BOTVINICK M M, PLAUT D C. Short-term memory for serial order: a recurrent neural network model [ J ]. Psychological review, 2006, 113(2): 201-233. [13] SAINATH T N, KINGSBURY B, SAON G, et al. Deep convolutional neural networks for large⁃scale speech tasks [J]. Neural networks, 2015, 64: 39-48. [14 ] SARIKAYA R, HINTON G E, DEORAS A, et al. Application of deep belief networks for natural language understanding[ J]. IEEE transactions on audio, speech, and language processing, 2014, 22(4): 778-784. [15]KRIZHEVSKY A, SUTSKEVER I, HINTON G E, et al. ImageNet classification with deep convolutional neural networks [ C ] / / International Conference on Neural Information Processing Systems. Nevada, America, 2012: 1097-1105. [16 ] SVOZIL D, KVASNICKA V, POSPICHAL J, et al. Introduction to multi⁃layer feed⁃forward neural networks [ J ]. Chemometrics and intelligent laboratory systems, 1997, 39(1): 43-62. [17] HORNIK K, STINCHCOMBE M B, WHITE H, et al. Multilayer feedforward networks are universal approximators [J]. Neural networks, 1989, 2(5): 359-366. [18]RUMELHART D E, HINTON G E, WILLIAMS R J, et al. Learning representations by back⁃propagating errors [ J]. Nature, 1988, 323(6088): 533-536. [19]KOSMATOPOULOS E B,POLYCARPOU M M,CHRISTODOULOU M A, et al. High⁃order neural network structures for identification of dynamical systems[ J]. IEEE transactions on neural networks, 1995, 6(2): 422-431. [20]WERBOS P J. Backpropagation through time: what it does and how to do it[ J]. Proceedings of the IEEE, 1990, 78 (10): 1550-1560. 作者简介: 雷森,男,1992 年生,博士研究生, 主要研究方向为图像处理、机器学习、 遥感影像质量提升。 史振威,男,1977 年生,教授,博士 生导师,博士,主要研究方向为图像处 理、模式识别、机器学习、遥感影像处 理。 发表 SCI 国 际 期 刊 检 索 论 文 70 余篇。 石天阳,男,1994 年生,硕士研究 生,主要研究方向为机器学习和人工 智能。 ·644· 智 能 系 统 学 报 第 12 卷