Modelling complex dynamical systems plays a crucial role to understand several phenomena in different domains such as physics, engineering, biology and social sciences. In this paper, a genetic programming ensemble method is proposed to learn complex dynamical systems' underlying mathematical models, represented as differential equations, from systems' time series observations. The proposed method relies on decomposing the modelling space based on given variable dependencies. An ensemble of learners is then applied in this decomposed space and their output is combined to generate the final model. Two examples of complex dynamical systems are used to test the performance of the proposed methodology where the standard genetic programming method has struggled to find matching model equations. The empirical results show the effectiveness of the proposed methodology in learning closely matching structure of almost all system equations.
Unfortunately, ACM prohibits us from displaying non-influential references for this paper.
To see the full reference list, please visit http://dl.acm.org/citation.cfm?id=3036336.