Data-driven estimation of air mass using Gaussian mixture regression

Abstract

The modelling and calculation of charge cycles with conventional intake manifold pressure based extensions is difficult to implement in real-time for combustion engines with extra actuators in valve train (VVT - variable valve timing) on current control units. Additionally, there is a high parametrization effort due to a variety of engine characteristics of this approach. In this paper we will analyse a cycle based calculation of the air mass with regard to an applicability for estimation in real time on the engine unit as well as varying options of actuators and sensor equipment components of combustion engines. We present a physical based, zero-dimensional model and the problem of its real-time realization is discussed. Furthermore, we will introduce a data-driven alternative for estimation of air mass using Gaussian Mixture Regression (GMR). The GMR allows a flexible data-driven modelling with a high input space dimensions together with a perspective of possibilities of adaption and local optimisation. Subsequently, the proposed method will be applied to a current Volkswagen (VW) Otto engine and the results discussed.