Approximate Optimal Control of Discrete I/O Systems with C/GMRES

This paper proposes an optimal control approach to control systems described by discrete time I/O models. In particular, the C/GMRES method, which is a curse-of- dimensionality free and fast approximative optimal control method is adapted to this purpose. Comparing this approach with conventional industry standard methods, the benefits are twofold: First, easy and intuitive tuning of the control via the quadratic cost function of the optimization problem, reducing the necessary parameterization and tuning effort typical in industrial applications. Second, direct use of discrete time I/O models, typically the result of data based modeling efforts, which themselves can reduce development time drastically, compared to classical first principles and parameterization methods. To demonstrate the possible advantages of this method, it is applied to a problem, where tedious control parametrization and the resulting tuning efforts are state of the art: The control of the Diesel engine airpath is a MIMO system tracking problem, where the achieved tracking error of the outputs are decisive of the main emissions of the Diesel engine.