Sparse Identification and Nonlinear Model Predictive Control for Diesel Engine Air Path System

This paper presents a sparse identification of nonlinear dynamic systems (SINDy) for a diesel engine air path system and nonlinear model predictive control (NMPC) with the SINDy model to attain good control performance. The air path system control is well known as a challenging problem, and many studies have been presented such as traditional model-based control design and machine learning. However, these conventional approaches still have some difficulties including the control performance and design costs. In this paper, we obtain the model of the air path system in a data-driven manner using the SINDy algorithm and construct the offset-free NMPC with the SINDy model. SINDy is a suitable modeling method for controlling a complicated air path system, owing to its characteristics of high computational efficiency, high learning efficiency, high modeling accuracy, and applicability to complex systems. Additionally, NMPC provides high control performance under constraints. The proposed offset-free NMPC with the SINDy model is verified through the simulations. The results show that the coefficient of determination of the SINDy model provided over 90%, and the controller performance of the NMPC was better than that of the traditional robust controller and satisfied the constraints.