Efficiency of OWC wave energy converters: A virtual laboratory
Introduction
Wave energy stands out among the different renewable energy sources not only for its high potential—which, according to the International Energy Agency can reach up to 80,000 TWh/year [1]—but also for its high energy density, the highest of all renewables [2]. This explains why wave energy is one of the renewable energy sources that is currently receiving greatest attention and R&D efforts [[3], [4], [5], [6]].
Oscillating water column (OWC) systems are one of the most popular technologies for wave energy conversion [[7], [8], [9]]. They consist of a partially submerged chamber with an underwater opening on its front and an air turbine. Waves impinging on the device cause the water column inside the chamber to oscillate, which gives its name to the system. As a result of these oscillations, the water column acts like a piston, forcing the air in the upper part of the chamber to flow alternatively out of the chamber and into it, driving the turbine in the process.
OWC converters present two main advantages over other wave energy converters (WECs). First, their simplicity—they consist exclusively of the two aforementioned elements, the chamber and the air turbine. Second, their low maintenance cost relative to other WECs, which are a result of both their simplicity and the absence of mechanical elements in direct contact with seawater.
The chamber and turbine are, therefore, the two essential elements of an OWC converter. Two main types of self-rectifying turbines are used: Wells turbines or impulse turbines [10,11]. As regards the chamber, a number of works were carried out with the aim of studying and optimising the design of the chamber [[12], [13], [14], [15]]. It is worth noting that, in most of the studies carried out so far, these two elements of an OWC converter, the air turbine and the chamber, are investigated separately—in spite of the fact that the coupling between both plays a fundamental role in the performance of the system [16]. In effect, the turbine should ideally provide the pneumatic damping (pressure drop through the turbine) for the chamber to work at, or near, resonant conditions, and the chamber, in turn, should provide the amount of pneumatic power that maximises the turbine output.
This work is concerned with a novel method based on artificial intelligence (in particular, artificial neural networks, ANNs) that enables to predict the pneumatic efficiency of an OWC converter with a given geometry under different wave conditions, taking into account the influence of the turbine (in other words, the turbine-chamber coupling) and the tidal level. ANNs have been widely used in coastal engineering, e.g. in connection with coastal structures [17,18], water level forecasting [[19], [20], [21]], long wave prediction [22], wind wave prediction [23,24], floating booms [25] or beach planform [[26], [27], [28]]. Moreover, several works have been developed on applications of ANN to wave energy conversion [[29], [30], [31]]. In this case, the problem at hand is certainly complex, and therefore various network architectures were implemented and their performances compared. The data for training and eventually validating the ANNs were obtained from laboratory tests of a model OWC converter carried out under different conditions of incident waves, tidal level and turbine damping. For this study, only regular waves were considered. Owing to the 2D nature of the tests, three-dimensional effects were not modelled.
Once the best architecture for the ANN model had been chosen, its results were analysed in depth, leading to the conclusion that the model was able to successfully generalise the information contained in the training cases. In sum, the ANN model thus developed constitutes a virtual laboratory that enables to predict the performance of an actual ANN prototype under the conditions of incident waves, tidal level and turbine damping of interest.
Section snippets
Artificial neural networks
An artificial neural network (ANN) consists of an interconnected group of simple elements (neurons) working in parallel [[32], [33], [34]]. ANNs emulate the behaviour of biological neural networks and, like them, learn on the basis of experience, of training-hence the term ‘artificial intelligence’. For this work, multilayer feedforward neural networks were chosen. They yield excellent results in tasks such as function approximation (nonlinear regression) and pattern recognition or
Implementation of a neural network model of OWC performance
The first step of the model implementation is establishing its input and output variables. The output, or dependent, variable was the capture factor defined by Eq. (10). As regards the input, or independent, variables, they were chosen taking into account the parameters on which the performance of an OWC system depends. They were related, on the one hand, to the conditions under which the converter operates: the wave height (H), period (T) and water depth (h); and, on the other hand, to the
Comparative study of network architectures
The average MSE of the 101 training and testing cases for the different architectures with one hidden layer is presented in Fig. 4. It may be seen that the error is greater for the architectures with a lower number of neurons. As the number of neurons increases the test error decreases, up to the [4-15-1] architectures, for which the tendency is inversed: the training error continues to decrease as the number of neurons increases but, most importantly, the test error increases slightly, which
Neural network model
Having selected the best performing neural network architecture for the model, this section presents its results in greater detail. To this end, of the 101 training runs, the one corresponding to the median MSE value is chosen. A detailed description of this network can be found in Table 1. The linear regression graph for the training dataset (Fig. 8) shows an excellent agreement between the values of the capture factor computed with the model and those obtained from the laboratory tests, with R
Conclusions
In this work artificial neural networks were applied to an important topic in wave energy: the prediction of the pneumatic efficiency of OWC wave energy converters. To train and test the neural networks, laboratory tests of a model OWC system were carried out with different wave conditions (wave heights and periods), tidal levels and damping values. Multilayer feedforward neural networks were selected on the grounds of their excellent generalisation capabilities. In order to choose the network
Acknowledgments
This research is part of the project DPI2009-14546-C02-02 supported by Spain's Ministry of Science and Innovation (Ministerio de Ciencia e Innovación). I. Lopez is indebted to the Spain's Ministry of Education, Culture and Sport (Ministerio de Educación, Cultura y Deporte) for the funding provided in the FPU Program (AP2010-4690).
References (49)
- et al.
Wave energy in Europe: current status and perspectives
Renewable and Sustainable Energy Reviews
(2002) - et al.
Wave resource in El Hierro—an island towards energy self-sufficiency
Renewable Energy
(2011) - et al.
Wave energy and nearshore hot spots: the case of the SE Bay of Biscay
Renewable Energy
(2010) - et al.
National-scale wave energy resource assessment for Australia
Renewable Energy
(2010) - et al.
Offshore and nearshore wave energy assessment around the Korean Peninsula
Energy
(2011) - et al.
Wave energy extraction by an oscillating water column with a gap on the fully submerged front wall
Applied Ocean Research
(2012) - et al.
Air-water two-phase flow modelling of hydrodynamic performance of an oscillating water column device
Renewable Energy
(2012) - et al.
Air turbines
- et al.
Analytical and experimental investigation on the hydrodynamic performance of onshore wave-power devices
Ocean Engineering
(2002) - et al.
Modeling and optimization of the chamber of OWC system
Energy
(2011)
Effect of turbine section orientation on the performance characteristics of an oscillating water column device
Experimental Thermal and Fluid Science
A virtual laboratory for stability tests of rubble-mound breakwaters
Ocean Engineering
Neural network modelling of wave overtopping at coastal structures
Coastal Engineering
Predicting sea level variations with artificial neural networks at Hillarys Boat Harbour, Western Australia
Estuarine, Coastal and Shelf Science
Application of artificial neural networks in tide-forecasting
Ocean Engineering
Predicting sea-level variations at the Cocos (Keeling) Islands with artificial neural networks
Computers & Geosciences
Artificial intelligence for estimating infragravity energy in a harbour
Ocean Engineering
Derivation of coastal wind and wave parameters from offshore measurements of TOPEX satellite using ANN
Coastal Engineering
Real-time wave forecasts off the western Indian coast
Applied Ocean Research
Artificial intelligence applied to floating boom behavior under waves and currents
Ocean Engineering
Neural network modelling of planform geometry of headland-bay beaches
Geomorphology
Headland-bay beach planform and tidal range: a neural network model
Geomorphology
Using an artificial neural network to model seasonal changes in beach profiles
Ocean Engineering
Automatic early stopping using cross validation: quantifying the criteria
Neural Networks
Cited by (73)
Optimization of the chamber of OWC to improve hydrodynamic performance
2023, Ocean EngineeringData-driven and physics-based approach for wave downscaling: A comparative study
2023, Ocean EngineeringA novel Geno-fuzzy based model for hydrodynamic efficiency prediction of a land-fixed oscillating water column for various front wall openings, power take-off dampings and incident wave steepnesses
2022, Renewable EnergyCitation Excerpt :Especially for complex geometries, efficiency assessments are carried out via physical experimental campaigns. Consequently, core concepts associated with developing more efficient WECs such as geometric optimization and tuning of the PTO induced damping are identified or at least validated by performing model scale tests [12–15]. However, experimental studies are expensive, time-consuming and high labor force requirement restricts its applicability.