NONLINEAR MODEL PREDICTIVE CONTROL BASED ON SEQUENTIAL MONTE CARLO STATE ESTIMATION

Abstract

Formulating the nonlinear model predictive control (NMPC) problem using nonlinear differential equations has been gaining attention recently, with the promise of improved performance. NMPC requires a knowledge of the states, which are rarely available directly. Hence, the role of a state estimator is crucial to provide state information from noisy process measurements. Earlier attempts to combine variants of the Kalman filter with NMPC met with limited success due to debilitating effects of linearization. Currently, moving horizon estimation (MHE) is the most popular choice since it is seen as a dual to the control problem. However, MHE typically makes simplifying assumptions about the nature of stochastic variables and lacks an efficient recursive formulation. Most importantly, MHE is an optimization burden in addition to the regulation problem to be solved online. We propose using the sequential Monte Carlo (SMC) filter for state estimation in NMPC since it is significantly faster and at least as accurate as MHE. More accurate and fast estimation results in faster control optimization for realtime use of NMPC and improves the performance. In this paper a comparison of NMPC performance is detailed with MHE and SMC state estimation in a nonlinear CSTR simulation study.