Machine Learning Enhanced Engine Model for Real-Time Model Predictive Control Application 2019-01-1289
Model predictive engine control has attracted research interests recently for its ability to reduce calibration effort and improve transient engine response. Performance of the Model Predictive Control (MPC) relies heavily on the accuracy of the model used to predict the future system behavior. This paper introduces a hybrid engine modeling approach combining physics-based methods and machine learning techniques. It preserves the generalization and robustness of the physics model, as well as the accuracy of data-driven models. Advantages of applying the proposed model with MPC are discussed. Issues encountered during this research are reported, and potential solutions are provided. While this paper focuses on the Artificial Neural Network approach, conclusions derived from this research can be applied to general machine learning techniques. The combination between MPC and the machine learning enhanced model was validated in both GTPower simulation and real-time dynamometer tests. The proposed control system successfully managed the investigated engine with tractable computational load, making it feasible for future Engine Control Units (ECUs).
Daniel Egan, Rohit Koli, Qilun Zhu, Robert Prucka