Advances in Wave Energy Conversion with Data-Driven Models

With an estimated theoretical resource of over 30,000 TWh/yr [...]

With an estimated theoretical resource of over 30,000 TWh/yr [1], a high degree of predictability (hours/days) and an energy density that is one order of magnitude greater than offshore wind and solar [2], wave energy has the potential to expand the renewable energy mix of coastal countries and contribute towards ongoing green energy transition and decarbonisation processes. Since 2010, wave energy reached a cumulative installed capacity of nearly 25 MW globally, about half of which derived from Europe [3]. Aside from mainland electrical grid supply, numerous marine applications are available for wave energy, including power supply to desalination plants [4], wave buoys [5], offshore aquaculture [6] and seaports [7], to which the prospects of storage/reconversion via green hydrogen production are also added [8]. These markets, however, require technological maturity and the demonstration of wave energy converters (WECs), which implies significant R&D efforts and investment.
Standard WEC development implies, from conception to open sea testing, multistage studies that apply distinct yet complementary techniques, following a composite modelling (CM) premise. The main branches of WEC R&D are physical modeling (PM) and numerical modelling (NM). Both share a fundamental aspect: the reproduction of physicsdriven phenomena, either via reduced-scale physical models of WECs-PM-or using simulations of said devices within a numerical environment-NM. Nonetheless, there are limitations. For instance, PM requires specialized laboratory facilities for conducting the experiments, which implies significant investments, logistical challenges, and extensive testing campaigns. To these, scale and laboratory effects are added, which can influence the reliability of the attained results [9]. Regarding NM, which only requires access to suitable hardware and the necessary simulation software, there are two main branches [10]:

•
Computational fluid dynamics (CFD), which exhibits high-fidelity for reproducing wave-structure interactions and solving the Navier-Stokes equations, but requires substantial computational time; • Potential flow theory (PFT), which is less computationally demanding at the expense of model fidelity and accuracy limitations inherent to theoretical assumptions: from fluid properties (e.g., incompressible and irrotational) to higher-order, non-linear terms (e.g., viscous hydrodynamic damping).
PM and NM generally share noteworthy testing/simulating times but can produce useful, reliable and relatively large datasets for analysis. However, the advent of advanced data-driven models (DDM), including artificial intelligence (AI), can highly benefit the development of WECs via integration into the CM paradigm. Data-driven models can employ the aforementioned datasets from demanding physics-driven models to predict, classify and assist in decision-making and optimization processes at a much lower cost and timeframe so long as they are adequately trained. This requires dataset pre-processing, DDM configuration (e.g., tuning hyperparameters), data-splitting (training, validation and testing) and quality control (e.g., avoidance of overfitting, gradient vanishing or local optima convergence), among others. WEC developers can either resort to existing toolboxes (e.g., Python Tensorflow or Matlab Optimization) or prepare/adapt AI scripts for their case studies' specificities. DDMs tend to be considered "black-box" or, at best, "grey-box" tools, as they have limited to no capability of explicitly reproducing physical phenomena. Even so, many can "learn" and identify key patterns in a given dataset, some of which would be very difficult to detect using standard means [11].
A wide variety of algorithmic tools, including statistical techniques, are already being implemented alongside standard PM and NM studies. Existing options encompass artificial neural networks (e.g., recurrent, convolutional and deep), evolutionary algorithms (e.g., genetic algorithms and swarm intelligence), fuzzy logic, random forests, clustering and system identification, among others. Thus far, they have successfully supported wave climate assessment and forecasting [12], representative sea-state selection [13], data-gap filling [14], control systems [15], WEC design [16] and park layout optimization [17]. With ever increasing data, algorithmic tools and hardware capability being produced, DDMs and AI have the potential to generate, alongside PM and NM, a CM "trinity" that is capable of bringing WECs closer to the viable commercialization milestone.
This Special Issue seeks to bring together cutting-edge research that utilizes datadriven approaches to optimize the design, performance, control and operational aspects of WECs and their sub-components, from mooring systems to the power take-off. This Special Issue encompasses various aspects, including stochastic techniques, meta-heuristic methods, machine learning algorithms, optimization methods, and case studies showcasing successful applications. We invite high-quality research papers that contribute to the advancement of WEC development and highlight the potential of data-driven models in promoting WEC efficiency, reliability and commercial viability.
Funding: This research was funded by project ATLANTIDA-Platform for the monitoring of the North Atlantic Ocean and tools for the sustainable exploitation of the marine resources, with reference NORTE-01-0145-FEDER-000040, which is co-financed by Fundação para a Ciência e Tecnologia (FCT) and the European Regional Development Fund (ERDF), via NORTE2020. This research study was further funded by WECANet COST Action CA17105 via the 8th Short-Term Scientific Mission call-grant with reference E-COST-GRANT-CA17105-63e51bbb.