Devised Methodology \labelprobMod
The fast and accurate modeling of complex systems is a relevant target nowadays. Modeling techniques allow designers to estimate the effects of variations in the performance of a system. Complex systems present non-linear characteristics as well as a high number of potential variables. Also, the optimal set of features that impacts the system performance is not well known as many mathematical relationships can exist among them.
Hence, we propose a methodology that considers all these factors by combining the benefits of both GE algorithms and classical \emphlasso regressions. This technique provides a generic and effective modeling approach that could be applied to numerous problems regarding complex systems, where the number of relevant variables or their interdependence are not known.
Figure \reffig:diagram shows the proposed methodology approach for the optimization of system modeling problem. Detailed explanations of the different phases are summarized in the following subsections.
[width=0.70]Diagram_GE_LASSO.png
\thesubsection GE feature selection
Given an extensive set of parameters that may cause an effect on system performance, FE selects the optimal set that best describes the system behavior. Also, this technique, which is provided by GE, avoids the inclusion of irrelevant features while incorporating correlations and combinations of representative variables. The input to our approach consists of a vector of initial data that includes the entire set of variables extracted from the system.