An interpretative and adaptive MPC for nonlinear systems

TL;DR

This paper introduces an interpretative and adaptive MPC approach for nonlinear systems that simplifies implementation and improves online model updating using a modified EKF and a matrix-free algorithm, demonstrating effectiveness in benchmarks.

Contribution

The paper presents a novel IA-MPC method that transforms a linearized model into an ARX form and updates it online with a modified EKF, enabling easier deployment on embedded systems.

Findings

IA-MPC outperforms traditional nonlinear MPC in benchmarks.

The method reduces computational complexity and deployment difficulty.

Experimental results validate the effectiveness of the proposed approach.

Abstract

Model predictive control (MPC) for nonlinear systems suffers a trade-off between the model accuracy and real-time computational burden. One widely used approximation method is the successive linearization MPC (SL-MPC) with EKF method, in which the EKF algorithm is to handle unmeasured disturbances and unavailable full states information. Inspired by this, an interpretative and adaptive MPC (IA-MPC) method, is presented in this paper. In our IA-MPC method, a linear state-space model is firstly obtained by performing the linearization of a first-principle-based model at the initial point, and then this linear state-space model is transformed into an equivalent ARX model. This interpretative ARX model is then updated online by the EKF algorithm, which is modified as a decoupled one without matrix-inverse operator. The corresponding ARX-based MPC problem are solved by our previous construction-free, matrix-free and library-free CDAL-ARX algorithm. This simple library-free C-code implementation would significantly reduce the difficulty in deploying nonlinear MPC on embedded platforms. The performance of the IA-MPC method is tested against the nonlinear MPC with EKF and SL-MPC with EKF method in four typical nonlinear benchmark examples, which show the effectiveness of our IA-MPC method.