A study on automatic adjustment of the HCCI engine controller using machine learning

Abstract

An automatic adjustment method of the model-based controller of an HCCI engine was designed in this study. As modeling errors are inevitable, feedback control is usually introduced to reduce the effect of the modeling errors. However, especially under transient conditions, the control performance may deteriorate, because this is a control with the information of the previous cycle. The transient control performance is thought to be improved by taking modeling errors into consideration. Therefore, an algorithm to adjust the feedback input based on the prediction of the modeling error was developed. The modeling error was learned and predicted by ReOS-ELM (Regularized Online Sequential Extreme Learning Machine), which is a method of online machine learning with low computational load. The modeling error learning was conducted every engine cycle. The feedback input was adjusted so that the output prediction by the engine model including its modeling error prediction coincided with the output reference. The reference tracking performance was improved by the proposed method.