M. C. Deo

Forecasting wind with neural networks

Abstract Wind forecasts over a varying period of time are needed for a variety of applications in the coastal and ocean region, like planning of construction and operation-related works as well as prediction of power output from wind turbines located in coastal areas. Such forecasting is currently done by adopting complex atmospheric models or by using statistical time-series analysis. Because occurrence of wind in nature is extremely uncertain no single technique can be entirely satisfactory. This leaves scope for alternative approaches. The present work employs the technique of neural networks in order to forecast daily, weekly as well as monthly wind speeds at two coastal locations in India. Both feed forward as well as recurrent networks are used. They are trained based on past data in an auto-regressive manner using back-propagation and cascade correlation algorithms. A generally satisfactory forecasting as reflected in its higher correlation and lower deviations with actual observations is noted. The neural network forecasting is also found to be more accurate than traditional statistical time-series analysis.

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.

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.

On-line wave prediction

Operational prediction of wave heights is generally made with the help of complex numerical models. This paper presents alternative schemes based on stochastic and neural network approaches. First order auto regressive moving average and auto regressive integrated moving average type of models along with a three-layered feed forward network are considered. The networks are trained using three different algorithms to make sure of the correct training. Predictions over intervals of 3, 6, 12 and 24 h are made at an offshore location in India where 3-hourly wave height data were being observed. Comparison of model predictions with the actual observations showed generally satisfactory performance of the chosen tools. Neural networks made more accurate predictions of wave heights than the time series schemes when shorter intervals of predictions were involved. For long range predictions both the stochastic and neural approaches showed similar performance. Small interval predictions were made more accurately than the large interval ones.

Neural networks in ocean engineering

Abstract The soft computing technique of neural network is being extensively used across all disciplines of ocean engineering, namely, offshore, coastal, and deep-ocean engineering including marine engineering. This paper takes a stock of the research studies reported so far in these areas. It is found that, in general, neural networks provide a better alternative, either substitutive or complementary, to traditional computational schemes of statistical regression, time series analysis, pattern matching, and numerical methods. The relative advantages of the neural network schemes proposed by various investigators are improved accuracy, lesser complexity in modeling and hence smaller computational effort and time, reduced data requirement in some cases, and so on. Neural networks have a very high degree of freedom, and that comes as handy while training it with examples. Exploration of more areas of application, implementation of advanced and hybrid forms of networks together with interpretation of the information contained in a trained network should receive more focus in future. Similarly the current difficulties in dealing with very large variations in the input, large warning times, extreme value predictions, and extrapolation beyond the observed range would have to be addressed in the near future.

Alternative neural networks to estimate the scour below spillways

Artificial neural networks (ANNs) are associated with difficulties like lack of success in a given problem and unpredictable level of accuracy that could be achieved. In every new application it therefore becomes necessary to check their usefulness vis-a-vis the traditional methods and also to ascertain their performance by trying out different combinations of network architectures and learning schemes. The present study was oriented in this direction and it pertained to the problem of scour depth prediction for ski-jump type of spillways. It evaluates performance of different network configurations and learning mechanisms. The network architectures considered are the usual feed forward back propagation trained using the standard error back propagation as well as the cascade correlation training schemes, relatively less used configurations of radial basis function and adaptive neuro-fuzzy inference system. The network inputs were characteristic head and discharge intensity over the spillways while the output was the predicted scour depth at downstream of the bucket. The performance of different schemes was tested using error criteria of correlation coefficient, average error, average absolute deviation, and mean square error. It was found that the traditional formulae of Veronese, Wu, Martins and Incyth as well as a new regression formula derived by authors failed to predict the scour depths satisfactorily and that the neuro-fuzzy scheme emerged as the most satisfactory one for the problem under consideration. This study showed that the traditional equation-based methods of predicting design scour downstream of a ski-jump bucket could better be replaced by one of the soft computing schemes.

Suitability of different neural networks in daily flow forecasting

Alternative forms of neural networks have been applied to forecast daily river flows on a continuous basis with the purpose of understanding how recent architectures like ANFIS, GRNN and RBF compare with traditional FFBP when monsoon-fed rivers involving significant statistical bias are involved. The forecasts are made at a location called Rajghat along river Narmada in India. Division of yearly data into four seasons and development of separate networks accordingly was found to be more useful than a single network applicable for the entire year. When a variety of error criteria were viewed together the most satisfactory network for all seasons was the radial basis function, which showed better performance then FFBP, GRNN and ANFIS. The FFBP network was found to be equally acceptable as the RBF in seasons other than the monsoon. Generally the peak flows were more satisfactorily modeled by the RBF than FFBP, GRNN and ANFIS. The relatively simpler handling of data non-linearity in FFBP was more attractive than complex ones of ANFIS and GRNN. The representative statistical model, namely response surface method, yielded highly unsatisfactory results compared to any ANN model involved in this study, confirming that the complexity of ANNs is really necessary to model daily river flows.

Prediction of breaking waves with neural networks

Abstract 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.

Estimation of scour below spillways using neural networks

Information on the depth of the scour hole formed downstream of a ski-jump bucket type of energy dissipator is necessary in determining the safety of dams and adjoining structures. Traditional formulae available to predict scour depth suffer from many limitations including one arising out of the technique of data analysis commonly employed, namely, statistical regression. This paper presents an alternative to regression in the form of neural networks. A feed forward network is developed to predict the depth of the scour hole below ski-jump spillways from the specified values of head and discharge intensity. Field measurements collected from a variety of published literature are used to train the network. The validation of the developed network using observations that were not involved in the training indicated the usefulness of the neural network approach for the prediction problem under consideration. Network-yielded values are found to be more accurate than those given by the traditional equations.A matrix of weights and bias values for general use in any location is specified

Tide Prediction Using Neural Networks

Prediction of tides at a subordinate station located in the interior of an estuary or a bay is normally done by applying an empirical correction factor to observations at some standard or reference station. This paper presents an objective way to do so with the help of the neural network technique. In complex field conditions this approach may look more attractive to apply. Prediction of high water and low water levels as well as that o fcontinuous tidal curves is made at three different locations. The networks involved are trained using alternative training algorithms. Testing of the networks indicated satisfactory reproduction of actual observations. This was further confirmed by a high value of the accompanying correlation coefficient. Such a correlation was better than the one obtained through use of the statistical linear regression model. The training algorithm of cascade correlation involved the lowest training time and hence is found to be more suitable for adaptive training purpose.