Since the last century, mathematical models have been development under the context of modelling Time Series, as well as their forecasts. And with the development of new technologies, new computer programming languages e applications of these models, a new sight of forecast and Time Series modelling are been used. Such fact can be seen by this papers approach, which considers not to use mathematical-heuristics criterion for explanatory variable selection, significance comparing p-values, but to use a method with optimization, AdaLasso, that can be more trustworthy and robust, and can even be used as an initialization technique for the Structural Model, because he depends on an starting point for its non-linear optimization method. Based on the study on this work, the combination of methods and techniques presented are able to stabilize the forecast of the Structural Model, and also robustly select explanatory variables.