Machine Learning for Identification and Optimal Control of Advanced Automotive Engines.

Abstract

The complexity of automotive engines continues to increase to meet increasing performance requirements such as high fuel economy and low emissions. The increased sensing capabilities associated with such systems generate a large volume of informative data. With advancements in computing technologies, predictive models of complex dynamic systems useful for diagnostics and controls can be developed using data based learning. Such models have a short development time and can serve as alternatives to traditional physics based modeling. In this thesis, the modeling and control problem of an advanced automotive engine, the homogeneous charge compression ignition (HCCI) engine, is addressed using data based learning techniques. Several frameworks including design of experiments for data generation, identification of HCCI combustion variables, modeling the HCCI operating envelope and model predictive control have been developed and analyzed. In addition, stable online learning algorithms for a general class of nonlinear systems have been developed using extreme learning machine (ELM) model structure.