Policy Gradient Reinforcement Learning for Uncertain Polytopic LPV Systems based on MHE-MPC

In this paper, we propose a learning-based Model Predictive Control (MPC) approach for the polytopic Linear Parameter-Varying (LPV) systems with inexact scheduling parameters (as exogenous signals bounds), where Time Invariant (LTI) models (vertices) captured by combinations of becomes wrong. We first to adopt Moving Horizon Estimation (MHE) scheme simultaneously estimate convex combination vector and unmeasured states based on observations model matching error. To tackle wrong LTI used in both MPC MHE schemes, then Policy Gradient (PG) Reinforcement Learning (RL) learn estimator controller so that best closed-loop performance is achieved. The effectiveness proposed RL-based MHE/MPC design demonstrated using an illustrative example.