Recent Articles and Multimedia
Paper Title: Integrating Reinforcement Learning and Model Predictive Control for Mixed-logical Dynamical Systems
Authors: Caio Fabio Oliveira da Silva, Azita Dabiri, and Bart De Schutter
The use of model predictive control for mixed-logical dynamical systems entails the solution of mixed-integer programs, which are typically intractable for real-time operation. The authors of this paper address this challenge by training a reinforcement learning policy to determine the discrete decision variables, thereby simplifying the control problem to a continuous optimization problem that is significantly more tractable.
READ MORE ABOUT THIS PAPER HERE
A depiction of the proposed control scheme that integrates reinforcement learning into an MPC framework for mixed-logical dynamical system. The agentβs goal is to maximize its long-term reward. It learns to adapt its policy by repeatedly interacting with the environment, that is, by sending a discrete action π_d(π) and by receiving the extended state π(π) and immediate reward π(π). The MPC controller, which is lumped in the environment, receives this discrete action π_d(π) and then solves an optimization problem to determine the continuous action π_c(π). Finally, the input π(π) is fed to the system, and the next state is computed.