Multilevel Predictive Models of IC Engine for Model Predictive Control Implementation

The paper deals with model based predictive control of combustion engines. Nonlinear black-box predictive models based on neuro-fuzzy approach are utilized. The structure of the models is optimized within an identification process. The nonlinear models are locally linearized and consequently used for the efficient on-line computation of forthcoming control actions. In desire to respect a fact that the speed of input-output response may vary significantly for different input/output groups, multilevel predictive models are proposed. Predictive control is again applied to approximate the desired behavior of chosen output variables. Potential algebraical constraints between different prediction layers are involved in the control algorithm using quadratic programming. The control scheme is optimized using simplified fast simulation model.