Uso de técnicas avanzadas para la modelización de la evolución de variables de proceso en una fábrica de cristal

Abstract

This work is enclosed in the developments achieved by the European Network of Excellence on Intelligent Technologies for Smart Adaptive Systems (EUNITE). It proposes the modeling of the evolution of a glass melting tank. This problem is especially complex because dependent variables must be considered as temporal series whose cycles, tendency and dependencies are unknown. The final mission is come together the conclusions of the different selection, filtering and modeling techniques. The advance in the solution takes advantage of synergy between the members of the excellence network. The GAPI team proposes a solution based on the data filter using projection techniques and relative distances between data, and a selection of the most relevant variables based on the APIMARS use. The estimation of the dependant variables future performance is made using a mix of the extracted information of the variable’s previous values and the static modeling using the independent variables. The solution includes three different models. For short term predictions, 8 hours (it must be considered a data was captured every quarter), the model gives much relevance to the previous values of the output variable. For predictions of medium range, 3 days, the model must give a greater relevance to the independent variables. And for predictions of long term, the model is based exclusively on independent variables. This approximation lets improve the short range fitting.