Handwritten Word Recognition Using Markov Models

Many industrial processes are non-linear, but in a certain range they can be considered linear. The objective of this work is to show the use of sub-space identification methods and prediction error methods, applied to a fluidized catalytic cracking unit. This unit is a complex operating equipment, non-linear, multivariable with many couplings, complex bifurcations and stability problems. In this simulated study, three discrete-time identification algorithms are applied to obtain an approximate model in state space, with multiple inputs and multiple outputs, around a given operating point, with the system operating in open loop, excited by multilevel random signals. The performance of those algorithms is compared employing quality criteria, considering cross validation. The selected model describes the complex dynamics of the system quite well.