The assessment of wave conditions that may lead to potential infrastructure failure is a critical step in the design not only of wave energy converters (WECs), but marine and coastal infrastructure in general. Overly conservative estimates can considerably increase the costs associated with infrastructure development whilst underestimates can lead to unacceptable rates of failure and potential loss of property and life. This is particularly the case for marine energy, where designing for the correct conditions is critical. Present approaches to estimate design conditions involve fitting a population of extreme events to a probability distribution. Extreme events, by their definition are rare and identifying the population of extreme events involves several somewhat arbitrary decisions on a threshold wave height or wave period above which an event is defined as extreme in addition to a time interval between events for them to be considered independent. It has been demonstrated that estimated design conditions are sensitive to these decisions, introducing some degree of subjectivity in the estimation of extreme conditions.

Deep neural networks present an alternative data-driven method that does not rely on subjective assumptions.  These emerging approaches have the potential for significant application across all areas of marine engineering, from wave renewable energy design, to offshore oil and gas and coastal engineering design. Existing extreme value analysis approaches provides a single point in time estimate of the wave conditions, whereas complex failure modes may be associated with the sequencing of conditions across time. This presentation will outline initial investigations on the use of auto-regressive deep learning methods for wave extreme value analysis and extreme multivariate timeseries generation (for wave height and wave period).