Indirect Training of Gray-Box Models Using Ls-Svm and Genetic Algorithms

Gray-Box Models which combine a phenomenological model with a black box tool are useful for determining the values of not well known parameters of the model. In this work an indirect strategy for training these gray box models using least-square support vector machine and genetic algorithms is presented. The gray box model was tested in a Continuous Stirred Tank Reactor process with good results (Index of Agreement for the model output variable and the estimated time-varying parameter > 0.90). © 2016 IEEE.