Modeling and Control of Engine Knock Using Gaussian Process Regression

Abstract

Knock limit control is necessary for high efficient operation of automotive engines. In this paper, we propose an identification and control method of the engine knock using a Gaussian process. Feedforward input is designed by the Gaussian process model approximating knock probability, and its model is updated online by a feedback of the knock intensity value. As modeling methods using the Gaussian process, we apply a heteroscedastic Gaussian process and a recursive Gaussian process. The effectiveness of the proposed method is demonstrated by a numerical simulation.