Dear M. J. Poustini,

Predictive learning-based control (LBMPC) is a recent technique that rigorously combines statistics and learning with control techniques, while providing assurance levels of security, robustness and convergence. . Indeed the aim is to reach the goal by distinguishing the concept of security and robustness of the concept of performance. An approximate model of a system with knowledge of its uncertainty provides security, and statistical and learning techniques refine the model and improve performance.

The beneficial effect of introducing learning to the MPC control is precisely to enrich the level of representation of the model in order to obtain the performances with great precision.

For more details and information in subject, I suggest you to see links and attached files in topics.

- Distributed MPC Using Reinforcement Learning Based ... - Springer Link

https://link.springer.com/chapter/10.1007/978-94-007-7006-5_32

- Distributed MPC using reinforcement learning based ... - ResearchGate

https://www.researchgate.net/.../284550989_Distributed_MPC_using_reinforcement_lea... WyS

-Reinforcement Learning Versus Model Predictive Control - IEEE Xplore

ieeexplore.ieee.org/iel5/3477/4802395/04717266.pdf

-What is the difference between Model Predictive Control (MPC ...

https://www.quora.com/What-is-the-difference-between-Model-Pre...

-Learning-based model predictive control for Markov decision processes

www.dcsc.tudelft.nl/~bdeschutter/pub/rep/04_021.pdf

-Learning-Based Tuning of Supervisory Model Predictive ... - CSIC

digital.csic.es/bitstream/10261/96417/1/Learning-based.pdf

-Distributed MPC Using Reinforcement Learning Based ... - Springer Link

https://link.springer.com/content/pdf/.../978-94-007-7006-5_32.pd...

Best regards

Dear M. J. Poustini,

Predictive learning-based control (LBMPC) is a recent technique that rigorously combines statistics and learning with control techniques, while providing assurance levels of security, robustness and convergence. . Indeed the aim is to reach the goal by distinguishing the concept of security and robustness of the concept of performance. An approximate model of a system with knowledge of its uncertainty provides security, and statistical and learning techniques refine the model and improve performance.

The beneficial effect of introducing learning to the MPC control is precisely to enrich the level of representation of the model in order to obtain the performances with great precision.

For more details and information in subject, I suggest you to see links and attached files in topics.

- Distributed MPC Using Reinforcement Learning Based ... - Springer Link

https://link.springer.com/chapter/10.1007/978-94-007-7006-5_32

- Distributed MPC using reinforcement learning based ... - ResearchGate

https://www.researchgate.net/.../284550989_Distributed_MPC_using_reinforcement_lea... WyS

-Reinforcement Learning Versus Model Predictive Control - IEEE Xplore

ieeexplore.ieee.org/iel5/3477/4802395/04717266.pdf

-What is the difference between Model Predictive Control (MPC ...

https://www.quora.com/What-is-the-difference-between-Model-Pre...

-Learning-based model predictive control for Markov decision processes

www.dcsc.tudelft.nl/~bdeschutter/pub/rep/04_021.pdf

-Learning-Based Tuning of Supervisory Model Predictive ... - CSIC

digital.csic.es/bitstream/10261/96417/1/Learning-based.pdf

-Distributed MPC Using Reinforcement Learning Based ... - Springer Link

https://link.springer.com/content/pdf/.../978-94-007-7006-5_32.pd...

Best regards

Thanks a lot Prof. Lafifi for your complete answer. I really expected such an answer.

It seems that there is a delicate difference between Adaptive & LB MPC which is included in the Model for the predicting the future.

Since for control of a system in practice, we have to use some filtering Technics to estimate the state parameters, What is exactly been estimated in LB MPC that can improve the Model parameters?

Dear M.J. Poustini,

I am totally agree with you, there is a delicate difference between Adaptive & LB MPC and i add that historically MPC was developed with the experiment of Adaptive Control and Optimal Control.

For the other part of question is to estimated model parameters In order to obtain a continuously refined model.

Best regards

While this differs slightly from traditional LBMPC, another application of machine learning (in this case, reinforcement learning) with MPC is for selecting the weights in the objective function. For certain applications, it can be difficult to select these weights using first principles. Reinforcement learning can be used to link the selection of these weights with the desired performance. You can check this out in one of our papers recently published in the International Journal of Adaptive Control and Signal Processing:

"Adaptive predictive control of a differential drive robot tuned with reinforcement learning".

Article Adaptive predictive control of a differential drive robot tu...

If you would like to investigate this idea in greater depth, you can also read my PhD thesis from Queen's University, Canada:

"A Reinforcement Learning Approach to Predictive Control Design: Autonomous Vehicle Applications".

Thesis A Reinforcement Learning Approach to Predictive Control Desi...

...a continuation of my previous response to the question on LBMPC...

Robust MPC is concerned with providing performance, stability, and feasibility guarantees in the presence of uncertainty. This uncertainty can come from modeling error, measurement error, or noise. The following paper provides some background and is concerned with providing guarantees of obstacle avoidance for unmanned aerial vehicles:

“A Robust Model-Predictive Guidance System for Autonomous Vehicles in Cluttered Environments” in IEEE Systems Journal:

https://ieeexplore.ieee.org/document/8338390