支持向量回归的机器学习方法在海浪预测中的应用

基于支持向量回归(SVR)方法,建立了渤海海域近岸海浪有效波高短期预测模型,并设计了多组风浪信息组合输入方案,开展了有效波高预测敏感性试验。研究发现:综合考虑当前风浪信息作为模型的输入,对3 h和6 h有效波高预测具有较高的预报技巧,但随着预测时效的延长其预测准确性迅速降低;若此时引入未来预测风速信息作为模型输入,则可极大提高对12 h和24 h有效波高的预测能力;此外,若输入信息与预测对象之间不存在显著相关,多个信息的输入对有效波高预测效果提高无显著作用。建立的机器学习模型对小样本数据集具有良好的适应能力,能够有效解决海浪预报中的非线性问题,可为近岸海浪有效波高短期预测提供合理的技术参考。     

基于变分模态分解和注意力机制的浪高预测

针对海洋中的海浪高度数据存在非线性和非平稳性的特点,海浪高度的预测就变得相对复杂。基于变分模态分解(VMD),在引入注意力机制(AM)的基础上,对传统长短期记忆(LSTM)神经网络算法进行了改进,提出了一种基于混合模型的海浪高度预测算法。算法通过预处理、预测和重构3个主要步骤,对海浪高度的时间序列进行预测。为了比较和说明,以太平洋东北海盆海域和马尾藻海域的4个站点浮标数据进行实验。实验结果表明,本文提出的混合模型(VALM)将海浪高度数据分解为更平稳和更规则的子序列;可以更好的区分数据之间的重要程度,并能够携带更多信息的数据;与支持向量回归(SVR)、人工神经网络(ANN)和LSTM等模型进行比较,VALM模型的预测效果最好且具备一定的鲁棒性。     

基于海浪谱分解与重构的资料同化试验

针对有效波高资料提出一种海浪谱分解与重构的资料同化方案:利用历史时段内的有效波高观测资料和模式计算波高场,采用最优插值方法得到分析波高场;在WAVEWATCH-Ⅲ模式的波浪能量密度谱和有效波高分析值之间引入一个变异系数矩阵,描述模式的误差,以此为状态向量构建卡尔曼滤波系统,对分解过的海浪谱进行修正和重构,得到同化后的海浪谱初始场。利用美国阿拉斯加湾北部海域的7个浮标站进行同化和72 h预报试验,对连续1个月的预报结果进行统计表明:采用该同化方案后24 h预报结果的有效波高均方根误差比未同化的结果降低了0.13 m;同化方案对预报效果的影响可持续36 h左右,随着预报时效延长,同化的效果减弱。     

基于Sentinel-1A的全球有效波高的反演研究

本文利用神经网络的技术手段,针对Sentinel-1A二级波模式数据提出一种用于海浪有效波高(Hs)反演的模型——N_N模型。该模型在基于ERS2 SAR波模数据开发的双参数模型的基础上,加入经度、纬度、方位向截断波长(λ_c)、图像偏斜(skewness,skew)、图像峰度(kurtosis,kurt)、卫星平台距目标物的距离与卫星飞行速度之比(β)等其他参数信息,根据不同输入参数的组合,建立了14个模型用于Hs反演,旨在分析各参数对有效波高反演的影响。通过分析表明,14个N_N模型相关系数都在0.8以上。随着λ_c、β参数的加入,N_N模型性能均大幅上升,且λ_c参数对模型性能的改善作用更加明显,相关系数提升0.06左右,均方根误差(Root Mean Squared Error,RMSE)下降0.12m左右。另外,skew与kurt的加入也使N_N模型性能有所改善,RMSE下降0.03m左右,相关系数提升0.01左右。其中,N_N10模型效果最佳且性能最稳定,与欧洲中程天气预测中心(the European Centre for Medium-Range Weather Forecasts,ECMWF)数据对比,相关系数(CORR)达到0.905,散射指数(Scattering Index,SI)与RMSE最低,分别为18.74%、0.502m,与独立测量的浮标数据的相关系数达到了0.894。     

集合最优插值方法在北印度洋海浪同化中的应用

基于第三代海浪模式WaveWatch III,采用集合最优插值(EnOI)方法对北印度洋海浪进行同化数值实验研究。在集合样本选取方案上,针对不同的实验分别选取有效波高(SWH)的历史后报场(样本A)、24h变化(样本B)以及以同一时刻72h预报时效和24h预报时效的差异(样本C)用于估计背景误差协方差。样本A和样本B是为海浪模拟而设计,样本C是为海浪预报而设计;通过与由高度计数据确定的模式背景误差进行比较,认为样本B优于样本A。采用样本B对2011年北印度洋海浪场进行同化模拟,结果表明2011-03-11相对误差改进都在5%及以上,其中7月份改进效果最佳。采用样本C对2013-07的有效波高进行0~72h预报,发现同化使0~24h预报改进最明显:均方根误差改进0.12m,相对误差改进5%。浮标检验结果支持上述结论。     

降雨条件下的导航X波段雷达海浪参数反演算法研究

X波段的电磁波受降雨影响容易产生衰减,这导致导航X波段雷达在降雨时无法用于海浪观测。本文提出了一种新的降低降雨影响的算法来反演海浪参数。首先,对X波段雷达图像做主成分分析,获得波浪变化的主成分,利用一维傅里叶变换得到波数谱,对其滤波减小降雨对雷达图像的影响;然后,选取JONSWAP(Joint North Sea Wave Project)谱作为理论谱,建立以观测谱与理论谱的最小化差异为目标函数的模型,求解该模型估算海浪的有效波高。与浮标测量的有效波高相比,该方法反演的有效波高的均方根误差是0.23 m,证明了该方法的可行性。     

JASON-2卫星高度计波高数据同化对中国渤黄海海浪数值预报影响研究

利用原国家海洋局北海分局浮标所测有效波高数据对Jason-2卫星高度计所测有效波高数据进行验证,采用50km空间窗和0.5h时间窗,得到219个时空配准点。对配准结果进行统计分析表明,Jason-2卫星高度计测得有效波高与浮标测量结果存在-0.277m的偏差,均方根误差为0.372m。利用最小二乘回归(OLR)对Jason-2有效波高数据进行校正可使其均方根误差下降至0.247m,减少34.5%。基于第三代海浪模式WAVEWATCH III对Jason-2有效波高数据进行最优插值同化试验,其中背景误差相关函数取为指数形式,相关距离尺度选为500km。与浮标观测数据比较表明,同化后模式有效波高均方根误差比未同化时减少11.56%,,能够有效地改善模式精度。以此为初始场进行为期3d的数值预报实验。与未同化实验相比,卫星高度计有效波高数据同化对模式0～72 h预报有不同程度的改善,改善程度随预报时间的增加而降低。     

基于两种嵌套方案数值模拟烟台北部海域海浪的对比研究

采用第三代海浪模式WAVEWATCH Ⅲ-SWAN三层嵌套以及SWAN三层自嵌套两种方式建立两套烟台市北部近岸海域的海浪数值模拟系统,利用ERA-Interim再分析风场对几次大浪过程进行后报,对比分析两种嵌套方式的模拟效果,以选取更适合烟台近岸海域海浪数值模拟的嵌套方案。结果表明:两种嵌套方案模拟的有效波高在空间分布上,除边界附近存在0.1 m左右的差异外,其他地区计算结果基本无区别,模拟的波向空间分布基本也无差别;两种嵌套方案模拟的有效波高和波向的单点时间序列也基本无区别。数值模拟结果与观测结果的对比表明,模式计算结果在整体上能反映出波浪的变化趋势,有效波高的均方根误差为0.2～0.6 m,波向的均方根误差在40°以内。     

X-波段雷达近海海浪频谱反演的神经网络模型

X-波段雷达作为国内海浪观测的一种新工具,在海浪频谱获取和有效波高反演方面仍存在较多问题.本文利用非线性回归方法,将现场实测浮标数据频谱和雷达一维图像谱分别与标准频谱模型进行拟合,发现浮标频谱和一维图像谱具有标准频谱的特征,能够较准确地获取相应的谱参数.提出了建立由雷达一维图像谱参数反演海浪频谱参数的神经网络模型,同时在模型中加入影像序列信噪比,进而反演有效波高,并将反演结果与现场实测数据和传统算法(建立影像序列信噪比与有效波高之间的线性回归方程)进行了对比,结果表明,获取谱参数的误差和反演有效波高的平均误差在20％以内,而传统算法计算有效波高平均误差在20％以上.     

基于同步风场的机载SAR海浪参数反演方法

针对机载合成孔径雷达(SAR)对海探测特点,采用多入射角法从SAR数据本身得到与海浪参数反演区域时空匹配的同步海面风速和风向,并结合线性变换关系,计算得到海浪初猜谱对应的仿真SAR图像谱,将仿真SAR图像谱和观测SAR图像谱输入代价函数中进行迭代运算,通过非线性方程的解算得到最适海浪谱;采用交叉谱法去除海浪传播180°方向模糊,最终得到海浪参数。论文提出的基于同步风场的机载SAR海浪参数反演方法,充分利用了机载SAR海洋环境探测的优势,解决了传统SAR海浪参数反演中初猜谱构造依赖外部风场的问题,机载同步飞行试验的海浪参数反演结果与浮标观测值的有效波高、波向的均方根误差分别为0.23 m和13.23°,验证了该方法的有效性,可为机载SAR海浪参数反演业务化提供支持。     

A forecasting model for wave heights based on a long short-term memory neural network

To explore new operational forecasting methods of waves, a forecasting model for wave heights at three stations in the Bohai Sea has been developed. This model is based on long short-term memory(LSTM) neural network with sea surface wind and wave heights as training samples. The prediction performance of the model is evaluated,and the error analysis shows that when using the same set of numerically predicted sea surface wind as input, the prediction error produced by the proposed LSTM model at Sta. N01 is 20%, 18% and 23% lower than the conventional numerical wave models in terms of the total root mean square error(RMSE), scatter index(SI) and mean absolute error(MAE), respectively. Particularly, for significant wave height in the range of 3–5 m, the prediction accuracy of the LSTM model is improved the most remarkably, with RMSE, SI and MAE all decreasing by 24%. It is also evident that the numbers of hidden neurons, the numbers of buoys used and the time length of training samples all have impact on the prediction accuracy. However, the prediction does not necessary improve with the increase of number of hidden neurons or number of buoys used. The experiment trained by data with the longest time length is found to perform the best overall compared to other experiments with a shorter time length for training. Overall, long short-term memory neural network was proved to be a very promising method for future development and applications in wave forecasting.     

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

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。     

Probability distribution of wave heights attenuated by salt marsh vegetation during tropical cyclone

Zero-crossing wave heights, obtained from the field measurement of random waves propagating through salt marsh vegetation (Spartina alterniflora) during a tropical storm, were analyzed to examine their probability distribution. Wave data (significant wave heights up to 0.4 m in 0.8 m depth) were collected over a two-day period along a 28 m transect using three pressure transducers sampling at 10 Hz. Wave height distribution was observed to deviate from the Rayleigh distribution. The observed probability densities of the larger wave heights were reduced significantly by vegetation, producing wave heights lower than those predicted by the Rayleigh distribution. Assuming Rayleigh distributed wave heights for the incident waves to the vegetation patch, existing vegetation-induced wave attenuation formulations are used to derive a special form of two-parameter Weibull distribution for wave heights in the inundated wetland. The scale parameter of the distribution is theoretically shown to be a function of the shape parameter, which agrees with the measurements, effectively reducing the proposed distribution to a one-parameter type. The derived distribution depends on the local parameters only and fits well to the observed distribution of wave heights attenuated by vegetation. Empirical relationships are developed to estimate the shape parameter from the local wave parameters.     

基于长短时记忆神经网络的台风路径临近预报模型

It is of vital importance to reduce injuries and economic losses by accurate forecasts of typhoon tracks. A huge amount of typhoon observations have been accumulated by the meteorological department, however, they are yet to be adequately utilized. It is an effective method to employ machine learning to perform forecasts. A long short term memory(LSTM) neural network is trained based on the typhoon observations during 1949–2011 in China's Mainland, combined with big data and data mining technologies, and a forecast model based on machine learning for the prediction of typhoon tracks is developed. The results show that the employed algorithm produces desirable 6–24 h nowcasting of typhoon tracks with an improved precision.     

Design estimates of surface wave interaction with compliant deepwater platforms

The extreme behavior of surface waves as they encounter and pass compliant deepwater platforms is an important class of problems for offshore engineers attempting to specify the platform deck elevation. In this study analytical expressions for the probability density and cumulative distribution functions that utilize empirical coefficients in an attempt to accurately model surface wave runup and airgap problems are presented. The analysis focuses upon interpreting the tails of the measured data histograms using two parameter Weibull distribution models. The appropriate empirical constants, assumed to be solely dependent upon the significant wave height, were evaluated and compared for all the test data. Based upon a small select set of data, for a mini-TLP and two Spar platforms, the airgap problem was found to be adequately modeled using a Rayleigh distribution. Further, for the seven seastates analyzed, the Weibull shape parameter was nearly constant and the data confirmed that the exclusive fit of the scale parameter assuming dependence only on the significant wave height was a reasonable approach for modeling the wave runup. Finally, by combining these models with a Poisson return model for each storm the associated reliability estimates for various deck heights were estimated.     

基于支持向量机预测安达曼海南部内孤立波传播特性

内孤立波是发生在密度稳定层化海水中的一种特殊的海洋内波。预测内孤立波传播难度较大。本文提出了一种方法,利用美国麻省理工学院大气环流模型（MITgcm）的内孤立波模型计算了大量模拟数据,建立数据库。采用机器学习的方法,建立一个基于支持向量机（support vector machine,SVM）的安达曼海南部内孤立波传播预测模型。最后运用安达曼海南部的Sentinel-1A合成孔径雷达（SAR）图像对内孤立波传播预测模型结果进行检验。结果表明：基于SVM的内孤立波传播时间预测模型预测的时间平均绝对百分比误差为8.43%,平均绝对误差为1.00 h。基于SVM的内孤立波到达位置预测模型预测的位置平均绝对百分比误差为0.071%,平均绝对误差为0.069°。基于SVM的内孤立波振幅预测模型预测的振幅范围为23.80~84.98 m。     

Prediction of significant wave height using regressive support vector machines

Wave parameters prediction is an important issue in coastal and offshore engineering. In this literature, several models and methods are introduced. In the recent years, the well-known soft computing approaches, such as artificial neural networks, fuzzy and adaptive neuro-fuzzy inference systems and etc., have been known as novel methods to form intelligent systems, these approaches has also been used to predict wave parameters, as well. It is not a long time that support vector machine (SVM) is introduced as a strong machine learning and data mining tool. In this paper, it is used to predict significant wave height (H_s). The data set used in this study comprises wave wind data gathered from deep water locations in Lake Michigan. Current wind speed (u) and those belonging up to six previous hours are given as input variables, while the significant wave height is the output parameter. The SVM results are compared with those of artificial neural networks, multi-layer perceptron (MLP) and radial basis function (RBF) models. The results show that SVM can be successfully used for prediction of H_s. Furthermore, comparisons indicate that the error statistics of SVM model marginally outperforms ANN even with much less computational time required.     

Computation of Long-Term Statistical Properties of Wave Agitation in Harbors

In this paper,the long-term statistical properties of wave height in an idealized square harborwith a partial opening are studied.The incident waves are propagated into the harbor numerically by the fi-nite/infinite element method using three different wave models:(1)monochromatic wave train,(2)long-crested random wave train,and(3)short-crested random wave train.This study shows that for a giv-en incident wave,the wave height in the harbor is affected by the wave model used.For long-term estima-tion of wave height exceedance probability,it is recommended that the waves be propagated into the har-bor using the random wave model,and that wave heights be computed by use of the Rayleigh probabilitydistribution.     

基于SVR和LSTM算法反演散射计下有效波高研究

有效波高反演对于海洋工程及海洋环境安全具有重要意义。我国海洋二号(HY-2A)卫星载有散射计和高度计等获取海洋要素的仪器。散射计可获取海洋风场数据但无法直接获取有效波高数据,高度计可获取海洋有效波高数据但覆盖区域狭小。本文将散射计与高度计各自优势结合,利用支持向量回归(SVR)和长短期记忆(LSTM)智能算法反演散射计下有效波高,提升高度计有效波高利用率。实验结果表明,长短期记忆智能算法更能有效反演散射计下有效波高。