Nonlinear predictive control based on local model networks for air management in diesel engines

Air management process in a turbocharged diesel engine is a multivariable, highly coupled nonlinear system with fast dynamics. Because of this, control algorithms with reasonably low computation times (enabling real-time application) must be used. Furthermore, testing new algorithms on a real engine is expensive. Therefore, a detailed nonlinear engine simulator based on a first-principles model is developed (Virtual Engine). This platform provides an extensive range of possibilities in the experimentation field, because complex and innovative algorithms can be tested in a nondestructive way. Identification and control schemes based on model predictive control and local model networks are proposed. The controller is implemented using the explicit solution based on the formulation introduced by Bemporad et al. [(2001). The explicit linear quadratic regulator for constrained systems. Automatica 38 (1), 3–20]. Finally, some results of these algorithms implemented on the engine simulator are offered, and compared with those obtained by applying standard predictive control.