MODEL PREDICTIVE CONTROL AND MODEL PREDICTIVE Q-LEARNING FOR STRUCTURAL VIBRATION CONTROL

This paper describes the relationship between Model Predictive Control (MPC) and Q-Learning, and formulates an algorithm called Model Predictive Q-Learning that integrates the two concepts. As a unifying theme, the paper explains how the Linear Quadratic Regulator (LQR), MPC, Q-Learning, and Model Predictive Q-Learning solve the same structural vibration control problem, and how the Q-Learning approach naturally handles both continuous and discrete-action inputs. The relationship between Model Predictive Q-Learning and standard Q-Learning is analogous to the relationship between MPC and LQR.