Response-Based Forecasting of Vessel Motion Shifts to Data-Driven Methods

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper OTC 35614, “Response-Based Forecasting of Vessel Motion: From Physics-Based to Data-Driven Methods,” by Ian Milne, SPE, The University of Western Australia and Woodside Energy; Lachlan C. Astfalck, The University of Western Australia; and Matthew Zed, Woodside Energy, et al. The paper has not been peer reviewed. Copyright 2025 Offshore Technology Conference. _ The conventional approach used to assess the operability of floating facilities typically has been based on wave height rather than wave-induced responses. During the past decade, this paradigm has been challenged through the development and provision of response-based forecasting. The availability of reliable measured motion data from vessels has enabled probabilistic rather than deterministic predictions through the application of statistical methods. The complete paper explores the motivation and development of response-based forecasting from the authors’ perspective, reviewing examples and processes that have served as validation and led to modeling refinement. Fundamentals of Response-Based Forecasts and Sources of Error Physics-Based Computation of Response Statistics. The theoretical framework for estimating the statistics of vessel response from wave spectra by means of linear hydrodynamic theory is well-established. Further, second-order effects such as slow drift, or contributions from winds and currents, are excluded. Because of the finite resolution of the wave spectra and response amplitude operators (RAOs), application of the significant heave amplitude (Eq. 1 of the complete paper) necessitates a numerical approximation of the integral and, in general, interpolation of either the wave spectra or RAO data, which may contribute significant finite-sample error. This also requires consideration of the fact that the hydrodynamic resonances are narrow-banded and that, at low frequencies, the wave spectral bins may be relatively wide. Interpolating nonlinear data with respect to frequency and direction is not straightforward and is a potential source of error. Numerical Weather Prediction (NWP) and Wave Spectra Forecasts. NWP, based on mathematical simulators of the atmosphere and oceans, is fundamental to operational wave forecasts. The outputs from NWP forecasts provide primary forcing to a numerical wave model by defining the transfer of energy from the atmosphere to the ocean surface. In a numerical wave model, the evolution of the wave spectra is modeled, solving the random phase spectral action density balance equation for wave-number-direction spectra. This considers wave generation, propagation, shoaling, refraction, energy transfer between waves, and wave dissipation. Specifically, third-generation models, which are currently industry standard, parameterize the physical sources and sinks of energy. Such models form the basis of industry forecasting products. However, many other forecast products are available that leverage these wave models. The industry has, however, emphasized that due consideration should be given to the fact that forecasts can differ greatly in quality. The authors note that operational forecasts now exist that have been built and modified to various degrees by machine-learning (ML) techniques. Through validation studies, the authors have found examples in which these are not of sufficient quality for use in response-based forecasting, likely because of the quality of particular ML algorithms; however, this is expected to change with time as the field continues to mature.