A Framework for Offshore Load Environment Modeling

Abstract

In the present paper, we propose a novel decision analytical framework for systems modeling in the context of risk informed integrity management of offshore facilities. Our focus concerns the development of system models representing environmental loads associated with storm events. Appreciating that system models in general serve to facilitate the optimal ranking of decision alternatives, we formulate the problem of systems modeling as an optimization problem to be solved jointly with the ranking of decision alternatives. Taking offset in recent developments in structure learning and Bayesian regression techniques, a generic approach for the modeling of environmental loads is established, which accommodates for a joint utilization of phenomenological understanding and knowledge contained in databases of observations. In this manner, we provide a framework and corresponding techniques supporting the combination of bottom-up and top-down modeling. Moreover, since phenomenological understanding as well as analysis of databases may lead to the identification of several competing system models, we include these in the formulation of the optimization problem. The proposed framework and utilized techniques are illustrated on a principal example. The example considers systems modeling and decision optimization in the context of possible evacuation of an offshore facility in the face of an emerging storm event.

Bayesian probabilistic representation of complex systems: With application to wave load modeling

“…The seasonal variable is defined based on the four seasons, the directions variables are defined based on eight directions (N, NE, E …), and the locational variables are clustered into two groups (black, i.e., upper left, and gray, i.e., lower right), as shown in Figure 7. A similar binary discretization of the location variables is learned in Glavind and Faber (2020) using the entire hindcast data set (23 platforms).…”

“…Starting point is taken by defining BNs in Section 3.1, and we proceed to discuss learning of BNs in Sections 3.2 for both fully observed and partially observed data sets. The reader is referred to, for example, Glavind and Faber (2020) and Bishop (2013) for a general introduction to inference in BNs.…”

“…Following Glavind and Faber (2018, 2020), and Nielsen et al. (2019), a system model $$\mathcal {M}$$ in this context provides a mapping from input to output, conditional on a decision alternative $${a}$$, which is measured in terms of utility $$\mathbb {U}(a)$$.…”

“…Decision analysis including system choice: (a) a single system and (b) multiple possible systems (adapted from Glavind & Faber, 2020)…”

“…This study considers the North Sea ocean environment in an area located approximately 220 km off the west coast of Denmark, where the ocean depth is approximately 40 m. The metocean records for the site includes wind fields and corresponding wave hindcast simulator outputs as well as wave measurements for a period of 37 years from January 10, 1979 to December 30, 2015. The hindcasts are produced for 23 locations using the spectral wind‐wave model MIKE21 SW (Sørensen et al., 2004) and the hydrodynamic model MIKE21 HD (Danish Hydraulic Institute, 2017), with climate forecast system reanalysis (CFSR) wind fields as input (Saha et al., 2014), and corresponding incomplete observations of the wave environment are available for seven of the locations, see also Glavind and Faber (2020) and Jones et al. (2018).…”