Prediction of breaking waves with neural networks

The height of a wave at the time of its breaking, as well as the depth of water in which it breaks, are the two basic parameters that are required as input in design exercises involving wave breaking. Currently the designers obtain these values with the help of graphical procedures and empirical equations. An alternative to this in the form of a neural network is presented in this paper. The networks were trained by combining the existing deterministic relations with a random component. The trained network was validated with the help of fresh laboratory observations. The validation results confirmed usefulness of the neural network approach for this application. The predicted breaking height and water depth were more accurate than those obtained traditionally through empirical schemes. Introduction of a random component in network training was found to yield better forecasts in some validation cases.     

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.     

Derivation of coastal wind and wave parameters from offshore measurements of TOPEX satellite using ANN

The paper discusses an artificial neural network (ANN) approach to project information on wind speed and waves collected by the TOPEX satellite at deeper locations to a specified coastal site. The observations of significant wave heights, average wave period and wind speed at a number of locations over a satellite track parallel to a coastline are used to estimate corresponding values of these three parameters at the coastal site of interest. A combined network involving an input and output of all the three parameters, viz., wave height, period and wind speed instead of separate networks for each one of these variables was found to be necessary in order to train the network with sufficient flexibility. It was also found that network training based on statistical homogeneity of data sets is essential to obtain accurate results. The problem of modeling wind speeds that are always associated with very high variations in their magnitudes was tackled in this study by imparting training in an innovated manner.     

GA-BP神经网络在罗源湾口波浪模拟研究中的应用

由于BP神经网络存在收敛速度慢和易于陷入极小值等缺点,引入遗传算法(GA)对网络的权值和阈值加以优化,并采用不同组合的输入因子和样本数,对福建省罗源湾口的波浪进行模拟研究.对输入因子的分析结果表明,研究区域的波浪主要受台湾海峡波浪传播影响,与局地气象因子(风速、风向、海气温差)的相关性较弱.训练样本试验表明,30 d以上的波浪历史数据可使GA-BP神经网络充分学习研究区域的波浪特征,从而实现对波浪要素的高精度模拟.模拟结果显示,对春、夏季实测波浪数据的模拟效果均很好,其中相关性分别为0.967和0.938,均方根误差分别为0.112 m和0.107 m,表明GA-BP神经网络在近岸波浪模拟预报中有较广阔的应用前景.     

基于人工神经网络的高频雷达风速反演

风速是重要的海洋状态参数之一,对海面风速的准确提取是实现海洋环境监测和沿海工程应用的重要保证。目前,作为新兴海洋环境监测设备,高频雷达在风速提取方面仍然存在挑战。本文提出了一种基于人工神经网络的风速提取方法,利用历史浮标测量海态数据训练风速提取网络,实现风速与有效波高、波周期、风向及时间因素之间的非线性映射。测试结果表明了这一网络在时间和空间上的稳定性;进而将已训练的网络应用到便携式高频地波雷达OSMAR-S的风速反演中,得到的风速与浮标测量风速间的相关系数达到0.849,均方根误差为2.11 m/s。这一结果明显优于常规由浪高反演风速的SMB方法,验证了该方法在高频雷达风速反演中的可行性。     

Real time wave forecasting using neural networks

Forecasting of ocean wave heights, with warning time of a few hours or days, is necessary in planning many operation-related activities in the ocean. Such information is currently derived by numerically solving the differential equation representing wave energy balance. The solution procedure involved is extremely complex and calls for very large amounts of meteorological and oceanographic data. This paper presents a complementary and simple method to make a point forecast of waves in real time sense based on the current observation of waves at a site. It incorporates the technique of neural networks. The network involved is first trained by different algorithms and then used to forecast waves with lead times varying from 3 to 24 h. The results of different training algorithms are compared with each other. The neural output is further compared with the statistical AR models.     

Real time wave forecasting using neural networks

Forecasting of ocean wave heights, with warning time of a few hours or days, is necessary in planning many operation-related activities in the ocean. Such information is currently derived by numerically solving the differential equation representing wave energy balance. The solution procedure involved is extremely complex and calls for very large amounts of meteorological and oceanographic data. This paper presents a complementary and simple method to make a point forecast of waves in real time sense based on the current observation of waves at a site. It incorporates the technique of neural networks. The network involved is first trained by different algorithms and then used to forecast waves with lead times varying from 3 to 24 h. The results of different training algorithms are compared with each other. The neural output is further compared with the statistical AR models.     

Artificial neural network to translate offshore satellite wave data to coastal locations

Owing to the spatial averaging involved in satellite sensing, use of observations so collected is often restricted to offshore regions. This paper discusses a technique to obtain significant wave heights at a specified coastal site from their values gathered by a satellite at deeper offshore locations. The technique is based on the approach of Artificial Neural Network (ANN) of Radial Basis Function (RBF) and Feed-forward Back-propagation (FFBP) type. The satellite-sensed data of significant wave height; average wave period and the wind speed were given as input to the network in order to obtain significant wave heights at a coastal site situated along the west coast of India. Qualitative as well as quantitative comparison of the network output with target observations showed usefulness of the selected networks in such an application vis-à-vis simpler techniques like statistical regression. The basic FFBP network predicted the higher waves more correctly although such a network was less attractive from the point of overall accuracy. Unlike satellite observations collection of buoy data is costly and hence, it is generally resorted to fewer locations and for a smaller period of time. As shown in this study the network can be trained with samples of buoy data and can be further used for routine wave forecasting at coastal locations based on more permanent flow of satellite observations.     

Wave measurements by pressure transducers using artificial neural networks

Underwater ultrasonic acoustic transducers are frequently used in ocean wave measurements as they measure surface level using acoustic waves. However, their effectiveness can be severely affected in rough sea conditions, when bubbles generated by breaking waves interfere with their acoustic signals. When the seas are rough, one therefore often has to rely on a pressure transducer, which is generally used as a back-up for the acoustic wave gauge. A pressure transfer function is then used to obtain the surface wave information. Alternatively, the present study employed an artificial neural network to convert the pressure signal into significant wave height, significant wave period, maximum wave height, and spectral peakedness parameter using data obtained from various water depths. The results showed that, for water depths greater than 20 m, the wave parameters obtained from the artificial neural network were significantly closer to those obtained by the acoustic measurements than those obtained by using a linear pressure transfer function. Moreover, for a given water depth, the wave heights estimated by the network model from pressure data were not as good as those estimated by linear wave theory for large wave heights (above a 4 m significant wave height in this study). This can be improved if the training data set has more records with large wave heights.     

基于遗传小波神经网络的海底声学底质识别分类

分割海底声纳探测图像,提取单元特征向量进行主成份分析,选取均值、标准差、对比度、相关系数、能量及同质性作为训练特征向量,构建小波神经网络。利用遗传算法优化小波神经网络的初始权值及小波参数,对砂、礁石、泥3种底质类型分别进行训练,并得到3种底质的测试精度都在90%以上,优于单独利用小波神经网络进行训练时的测试精度,克服了小波神经网络训练时易陷入局部极小的固有缺陷,表明基于遗传算法的小波神经网络可有效用于海底底质声纳图像的识别和分类。     

Real-time wave forecasts off the western Indian coast

The wave observations at three locations off the west coast of India have been analyzed using artificial neural network (ANN) to obtain forecasts of significant wave heights at intervals of 3, 6, 12 and 24 h. The most appropriate training method requiring an input of four observations spread over previous 24 h has been selected after considerable trials. Further, the networks are trained after filling in the missing information. Larger gaps in data are filled in using spatial mapping involving observations at nearby locations, while relatively smaller gaps are accounted for by the statistical technique of multiple regressions in temporal mode. It is found that by doing so the long-interval forecasting is tremendously improved, with corresponding accuracy levels becoming close to those of the short-interval forecasts. If the amount of gaps is restricted to around 2% per year or so it is possible to obtain 12 h ahead forecasts with 0.08 m accuracy on an average and 24 h ahead forecast with a mean accuracy of 0.13 m. However, in harsher environments the prediction accuracy can change.     

Ocean wave forecasting using recurrent neural networks

The tremendous increase in offshore operational activities demands improved wave forecasting techniques. With the knowledge of accurate wave conditions, it is possible to carry out the marine activities such as offshore drilling, naval operations, merchant vessel routing, nearshore construction, etc. more efficiently and safely. This paper describes an artificial neural network, namely recurrent neural network with rprop update algorithm and is applied for wave forecasting. Measured ocean waves off Marmugao, west coast of India are used for this study. Here, the recurrent neural network of 3, 6 and 12 hourly wave forecasting yields the correlation coefficients of 0.95, 0.90 and 0.87, respectively. This shows that the wave forecasting using recurrent neural network yields better results than the previous neural network application.     

Neuro-fuzzy based approach for wave transmission prediction of horizontally interlaced multilayer moored floating pipe breakwater

The ocean wave system in nature is very complicated and physical model studies on floating breakwaters are expensive and time consuming. Till now, there has not been available a simple mathematical model to predict the wave transmission through floating breakwaters by considering all the boundary conditions. This is due to complexity and vagueness associated with many of the governing variables and their effects on the performance of breakwater. In the present paper, Adaptive Neuro-Fuzzy Inference System (ANFIS), an implementation of a representative fuzzy inference system using a back-propagation neural network-like structure, with limited mathematical representation of the system, is developed. An ANFIS is trained on the data set obtained from experimental wave transmission of horizontally interlaced multilayer moored floating pipe breakwater using regular wave flume at Marine Structure Laboratory, National Institute of Technology Karnataka, Surathkal, India. Computer simulations conducted on this data shows the effectiveness of the approach in terms of statistical measures, such as correlation coefficient, root-mean-square error and scatter index. Influence of input parameters is assessed using the principal component analysis. Also results of ANFIS models are compared with that of artificial neural network models.     

Neural networks for wave forecasting

The physical process of generation of waves by wind is extremely complex, uncertain and not yet fully understood. Despite a variety of deterministic models presented to predict the heights and periods of waves from the characteristics of the generating wind, a large scope still exists to improve on the existing models or to provide alternatives to them. This paper explores the possibility of employing the relatively recent technique of neural networks for this purpose. A simple 3-layered feed forward type of network is developed to obtain the output of significant wave heights and average wave periods from the input of generating wind speeds. The network is trained with different algorithms and using three sets of data. The results show that an appropriately trained network could provide satisfactory results in open wider areas, in deep water and also when the sampling and prediction interval is large, such as a week. A proper choice of training patterns is found to be crucial in achieving adequate training.     

Using an artificial neural network to model seasonal changes in beach profiles

An artificial neural network (ANN) was applied to predict seasonal beach profile evolution at various locations along the Tremadoc Bay, eastern Irish Sea. The beach profile variations in 19 stations for a period of about 7 years were studied using ANN. The model results were compared with field data. The most critical part of constructing ANN was the selection of minimum effective input data and the choice of proper activation function. Accordingly, some numerical techniques such as principal component analysis and correlation analysis were employed to detect the proper dataset. The geometric properties of the beach, wind data, local wave climate, and the corresponding beach level changes were fed to a feedforward backpropagation ANN. The performance of less than 0.0007 (mean square error) was achieved. The trained ANN model results had very good agreement with the beach profile surveys for the test data. Results of this study show that ANN can predict seasonal beach profile changes effectively, and the ANN results are generally more accurate when compared with computationally expensive mathematical model of the same study region. The ANN model results can be improved by the addition of more data, but the applicability of this method is limited to the range of the training data.     

Real time wave forecasting using wind time history and numerical model

Operational activities in the ocean like planning for structural repairs or fishing expeditions require real time prediction of waves over typical time duration of say a few hours. Such predictions can be made by using a numerical model or a time series model employing continuously recorded waves. This paper presents another option to do so and it is based on a different time series approach in which the input is in the form of preceding wind speed and wind direction observations. This would be useful for those stations where the costly wave buoys are not deployed and instead only meteorological buoys measuring wind are moored. The technique employs alternative artificial intelligence approaches of an artificial neural network (ANN), genetic programming (GP) and model tree (MT) to carry out the time series modeling of wind to obtain waves. Wind observations at four offshore sites along the east coast of India were used. For calibration purpose the wave data was generated using a numerical model. The predicted waves obtained using the proposed time series models when compared with the numerically generated waves showed good resemblance in terms of the selected error criteria. Large differences across the chosen techniques of ANN, GP, MT were not noticed. Wave hindcasting at the same time step and the predictions over shorter lead times were better than the predictions over longer lead times. The proposed method is a cost effective and convenient option when a site-specific information is desired.     

Calibration and verification of a spectral wind–wave model for Lake Okeechobee

A spectral wind wave model SWAN (Simulation WAves Nearshore) that represents the generation, propagation and dissipation of waves was applied to Lake Okeechobee. This model includes the effects of refraction, shoaling, and blocking in wave propagation. It accounts for wave dissipation by whitecapping, bottom friction, and depth-induced wave breaking. The wave–wave interaction effect also is included in this model. Measurements of wind and wave heights were made at different stations and different time periods in Lake Okeechobee. Significant wave height values were computed from the recorded data. The correlation between wind stress and significant wave height also was analyzed. A 6-day simulation using 1989 data was conducted for model calibration. Another 6-day simulation using 1996 data was conducted for model verification. The simulated significant wave heights were found to agree reasonably well with measured significant wave heights for calibration and verification periods. Agreement between observed and simulated values was based on graphical comparisons, mean, absolute and root mean square errors, and correlation coefficient. Comparisons showed that the model reproduced both general observed trends and short term fluctuations.     

基于多变量LSTM神经网络模型的PDO指数预测研究

利用1921–2020年的海平面气压、海平面高度、热含量数据以及海冰密集度作为太平洋年代际振荡（Pacific Decadal Oscillation, PDO）指数的预报要素,建立了关于PDO指数时间序列预测的多变量长短期记忆（Long Short Term Memory, LSTM）神经网络模型,对比分析了2011–2020年不同时间序列预测模型的PDO指数预测结果,最后利用多变量LSTM神经网络模型实现了2021–2030年的PDO指数预测。结果显示,多变量LSTM神经网络模型的预测值与观测值经过交叉验证后的平均相关系数和均方根误差分别为0.70和0.62;PDO未来10年将一直处于冷位相,PDO神经网络指数出现两次波动,于2025年出现最小值。相比于其他时间序列预测模型,本文采用的多变量LSTM神经网络模型预测结果误差小、拟合效果好,可以作为一种新型的预测PDO指数的手段。     

基于BP神经网络的摇臂式波浪发电平台取能效率预测

取能效率是衡量波浪发电装置设计合理与否的重要参考标准。文章首先介绍了摇臂式波浪发电平台,接着对BP神经网络的原理和算法进行了描述,最后以水池试验过程中收集的数据为样本数据,在Matlab平台上运用BP神经网络对实海况下摇臂式波浪发电平台的取能效率作了仿真预测。仿真结果表明：实海况下摇臂式波浪发电平台的取能效率达到了预期目标,进一步说明BP神经网络成功训练出可靠的网络,在此基础上预测的数据具有一定的参考价值。     

Calibration of visual observations of wave period

Visual observations of wave properties are an important source of statistical information needed for the prediction of design and operational conditions of ocean structures. In particular, mean wave periods are important parameters for predicting the response of ocean structures. The existing calibration studies were based on data sets with a poor correlation between observations and measurements. The data set analysed here show a good correlation indicating the feasibility of collecting further good data sets to provide an adequate calibration to the existing statistical compilations of wave data.