Não informado
It is well-known that Thermoelectric Generators (TEG) can be employed as energy harvesting devices for power supply in sensor networks over the internet of things frameworks. However, proper models for these generators need to be devised to allow the suitable design of the associated electronic system. Many metrics can be used for this purpose, but the task can become tedious and yet the performance of the model may be far below to the desired one. Recently, through the evolutionary computation concept, new methodologies have been discovered that allowed us to obtain amenable equations to represent phenomena that were very complex to be explained mathematically by the use of conventional modelling techniques. This paper proposes a novel methodology to establish new governing equations for thermoelectric devices from a given input/output data set. One component of the method executes Genetic Programming (GP) ensemble based on symbolic regression with tree evolution restrictions, to select among thousands of structures the most parsimonious one along with its parameters. Resulted expressions are calculated numerically based on the time series of the state variables using models of the Nonlinear Auto-Regressive eXogenous (NARX) type, that match with the original nonlinear dynamical system representing the dynamical TEG behavior. To demonstrate the potential of the proposed methodology, we present four different new structures for the same system under investigation, which were obtained from just a given time series data.