Air–fuel ratio prediction and NMPC for SI engines with modified Volterra model and RBF network

The dynamics of air manifold and fuel injection in spark ignition (SI) engines are modelled with both a modified Volterra series and a radial basis function (RBF) network. In a Volterra model-based model predictive control (MPC) the global optimal control can always be solved by a linear optimization, but the model accuracy is low. On contrast, the RBF model provides more accurate prediction but its nonlinearity makes optimization nonlinear and non-convex, and therefore not always solvable. In this paper, the two models are combined so that linear optimization is enabled and the prediction accuracy is compensated, and therefore, both reliability and accuracy are achieved in a relative low computing cost. Using the developed method, the nonlinear MPC (NMPC) of the air/fuel ratio is implemented and evaluated by computer simulation. A real-time simulation using d-SPACE is also conducted to assess the real-time execution of the software. The simulation results show that the RBF compensated MPC outperformed over the Volterra model or RBF model based control.