Non-linear model-based predictive control of gasoline engine air-fuel ratio

Abstract

Control developments allowing accurate regulation of air-fuel ratio in gasoline engines are critical if legislative emissions levels are to be adhered to early in the next century. However, the task is far from straightforward with severe non-linearities and long/variable dead times challenging even the most sophisticated control algorithms. The availability of accurate models of the system can aid in overcoming these hurdles. Neural networks offer one modelling approach which enables rapid and accurate model formulation from system performance data. Whilst neural network models may provide the required accuracy, they do not easily fit within a control framework, particularly when there is a requirement for a rapid sampling frequency. This paper shows how a neural network model may be built and incorporated within a model predictive control framework and, with some approximations, may be implemented on a system requiring frequent sampling. Application to a simulation of a sophisticated car engine serves to demonstrate the potential of the approach.