Probabilistic Interval Prediction of Ship Roll Motion Using Multi-Resolution Decomposition and Non-Parametric Kernel Density Estimation

Abstract

The prediction of ship roll motion is important for the safety of marine vessels and the efficiency of offshore operations. This study aims to address the issues of low prediction accuracy and incapability to illustrate the uncertainty of ship roll motion in traditional deterministic prediction models. Thus, a probabilistic interval prediction scheme for ship roll motion is established utilizing the novel data preprocessing strategy and non-parametric kernel density estimation (KDE). The novel data preprocessing method based on multi-resolution decomposition and reconstruction strategy is used to reduce the non-stationarity of the time series and number of mode components. This method is also utilized to extract trends from the mode components for preserving more effective information. Accordingly, the dynamic recurrent neural network is adopted to enable the identification and deterministic prediction of reconstruction data for ship roll motion. The deterministic prediction values of ship motion status can be obtained by reconstructing the predicted outcomes of the mode components. Non-parametric KDE is applied to predict the fluctuation range of ship roll status by combining deterministic prediction results. The feasibility and effectiveness of the probability prediction scheme are exhibited through the measured data of M.V. YuKun. Results show that the deterministic prediction accuracy of the proposed scheme is superior to those of traditional deterministic prediction models, with maximum improvements of 22.7% and 27.5% for single-and multi-step prediction accuracies, respectively. The minimum prediction interval coverage probability at the 85% confidence level is 98%. The reliable fluctuation range of ship roll can be realized through the non-parametric KDE, which can offer more effective reference information for ship operators. The proposed model not only satisfies the requirements of deterministic prediction accuracy but also produces reliable uncertainty estimates of ship roll motion status.

Keywords

Ship roll motion probabilistic prediction / Data preprocessing strategy / Neural network / Non-parametric kernel density estimation

RIGHTS & PERMISSIONS

Harbin Engineering University and Springer-Verlag GmbH Germany, part of Springer Nature