Fast nonlinear model predictive control of gasoline engines

Nonlinear Model Predictive Control (NMPC) has proven successful in many applications, especially in chemical and process engineering where sampling times are usually in the order of seconds or minutes. Combustion engines show highly nonlinear behaviour and require sampling times of few milliseconds which poses a challenging control problem. We aim at tracking a desired torque profile of a gasoline engine without increasing the emissions, using the throttle and the exhaust gas recirculation valve as manipulated variables. In our simulations, it was possible to reduce the torque error considerably compared to controls taken from an optimal static map. Employing Bock's direct multiple shooting method and the so-called real-time iteration scheme the average runtime needed for one re-optimisation was reduced to below 10ms. These preliminary results show that NMPC might soon become a viable technique for demanding control problems within the automotive area.