Author(s): S. Velickov

Linked Author(s):

Keywords: Nonlinear dynamical systems; Deterministic chaos; Mixture of models; Data-driven modelling framework

Abstract: driven modelling framework, termed as Hidden Markov Mixture of Models (experts) –HMMMs. The framework aims at separating the seemingly complex global nonlinear dynamics into couple of local sub-dynamics that can be modelled by separate models (experts). The separate local models through a competition specialise on modelling different parts of the reconstructed phase space of the dynamical system where the gating procedure between the models is described with a dynamic Bayesian network expressed as hidden Markov model. We firstly test this framework on synthetic data generated by known dynamical systems. Finally, we apply the mixture of models framework for forecasting surges at Hoek van Holland tidal station in the North Sea. The elaborated models (experts) are multivariate local models incorporating neighbouring statistics on the meteorological forcing, tide-surge interaction and the different tidal phases dynamics, showing reliable and accurate short-term forecasting performance of the surges. HMMMs experiments showed improved predictive performances in comparison with other nonlinear global data-driven modelling techniques, such as neural networks and fuzzy inference systems. Abstract: In this paper we propose, mathematically elaborate and apply a novel data-

DOI:

Year: 2003