Design sea state assessment via IFORM-based environmental contour approach derived with Gaussian mixture model

Environmental contour (EC) plays a crucial role in the design of marine structures. It has been widely used to pinpoint extreme sea states and estimate potential structural damages. ECs can be obtained through the Inverse First Order Reliability Method (IFORM) in combination with joint probability techniques for metocean variables. This paper aims to capture the intricate dependence structure within metocean data by employing a suitable and versatile statistical model. The Gaussian mixture distribution was adopted to model both the marginal and bivariate distributions of environmental variables. The optimal components number was ascertained using the Bayesian Information Criterion, and the expectation-maximization algorithm was harnessed to estimate the numerous unknown parameters in the mixture models. The validity of the joint mixture model was then evaluated through statistical tests. Subsequently, the ECs derived using the Gaussian mixture model were compared with those constructed by the commonly used conditional Weibull-Lognormal model. The robustness of the mixture model was verified by its satisfied performance in fitting wave data from diverse sites and handling contours with highly complex probabilistic characteristics. Those refined ECs showed great potential for evaluating the most probable extreme responses of marine structures, thereby enhancing the safety and reliability of marine engineering designs. • A Gaussian mixture model is formulated to characterize the joint distribution of waves. • The multiple unknown parameters are solved using the Expectation-Maximization algorithm. • The effectiveness and flexibility are verified via error metrics and environmental contours. • The mixture model is contrasted against the commonly applied conditional model.