Zone Model Predictive Control with Active Learning and Application to Cerebrospinal Fluid Dynamics

This paper presents a zone model predictive control (MPC) scheme with active dynamics learning for a class of nonlinear systems with time-varying and uncertain parameters. The goal of this MPC scheme is to provide a systematic trade-off between three competing objectives: regulation of the system state to a desired zone, identification of the system parameters up to a desired accuracy, and minimization of the rate of input changes. To address this challenge, we integrate the covariance propagation of a Kalman filter used for parameter estimation into a zone MPC scheme and control the accuracy of the parameter estimates to a desired level by introducing a suitable soft constraint on the predicted covariances. We highlight the potential of the active learning zone MPC for biomedical applications through simulations of cerebrospinal fluid dynamics, demonstrating how the proposed method can be used to improve the diagnosis and therapy of hydrocephalus.