Adaptive Ensemble of Multi-kernel Gaussian Process Regressions based on Heuristic Model Screening for Nonparametric Modeling of Ship Maneuvering Motion

Abstract

Gaussian process regression (GPR) is a commonly used approach for establishing the nonparametric models of ship maneuvering motion, and its performance depends on the selection of the kernel function. However, no single kernel function can be universally applied to all nonparametric models of ship maneuvering motion, which may compromise the robustness of GPR. To address this issue, an adaptive ensemble of multi-kernel GPRs based on heuristic model screening (AEGPR-HMS) is proposed in this paper. In the proposed method, four kernel functions are involved in constructing the ensemble model. The HMS method is introduced to determine the weights of individual-based GPR models, which can be adaptively assigned according to the baseline GPR model. To determine the hyper-parameters of these kernel functions, the genetic algorithm is also employed to compute the optimal values. The KVLCC2 tanker provided by the SIMMAN 2008 workshop is used to validate the performance of the proposed method. The results demonstrate that the AEGPR-HMS is an efficient and robust method for nonparametric modeling of ship maneuvering motion.