• Corpus ID: 165163772

Deep Model Predictive Control with Online Learning for Complex Physical Systems

  title={Deep Model Predictive Control with Online Learning for Complex Physical Systems},
  author={Katharina Bieker and Sebastian Peitz and Steven L. Brunton and J. Nathan Kutz and Michael Dellnitz},
The control of complex systems is of critical importance in many branches of science, engineering, and industry. Controlling an unsteady fluid flow is particularly important, as flow control is a key enabler for technologies in energy (e.g., wind, tidal, and combustion), transportation (e.g., planes, trains, and automobiles), security (e.g., tracking airborne contamination), and health (e.g., artificial hearts and artificial respiration). However, the high-dimensional, nonlinear, and multi… 

Figures from this paper

Constrained Physics-Informed Deep Learning for Stable System Identification and Control of Unknown Linear Systems

The empirical results demonstrate the effectiveness of the unifying framework for constrained optimal control of linear systems to provide stability guarantees of the learned dynamics, robustness to uncertainty, and high sampling efficiency.

Differentiable Predictive Control: An MPC Alternative for Unknown Nonlinear Systems using Constrained Deep Learning

Beyond improved control performance, the DPC method scales linearly compared to exponential scalability of the explicit MPC solved via multiparametric programming, hence, opening doors for applications in nonlinear systems with a large number of variables and fast sampling rates which are beyond the reach of classical explicitMPC.

Embedded Implementation of Deep Learning-based Linear Model Predictive Control

Results show that the proposed DNN-MPC performs faster and with less memory footprints while retaining the controller performance, and is compared with traditional MPC.

Learning based Model Predictive Control (LBMPC)

This thesis work devises a novel deep learning-based model predictive control (DNN-MPC) that uses recurrent neural network (RNN) to accurately predict the future output states based on the previous training data and performs faster and with less memory footprints while retaining the controller performance.

A Hybrid Learning Method for System Identification and Optimal Control

The method is designed for systems for which only historical data under closed-loop control are available and where historical control commands exhibit low variability, and generates stable functional controllers that outperform on comfort and energy benchmark rule-based controllers.

Non-diverging neural networks supported tube model predictive control

A safe neural network-supported learning tube model predictive control scheme is proposed, which allows to bound the worst-case performance in case of a malfunctioning of the machine learning component, yet enables decreasing the conservatism.

Stochastic spatio-temporal optimization for control and co-design of systems in robotics and applied physics

A novel sampling-based stochastic optimization framework based entirely in Hilbert spaces suitable for the general class of semi-linear SPDEs which describes many systems in robotics and applied physics which is utilized for simultaneous policy optimization and actuator co-design optimization.

State-space models for building control: how deep should you go?

This work systematically investigates whether using RNNs for building control provides net gains in MPC and compares the representation power and control performance of two architectures: an RNN architecture and a linear state-space model with a nonlinear regressor to estimate energy consumption.

Deep Learning Based Model Predictive Control for a Reverse Osmosis Desalination Plant

An NMPC for a RO desalination plant that utilizes an LSTM as the predictive model will be presented and it will be tested to maintain a given permeate flow rate and keep the permeate concentration under a certain limit by manipulating the feed pressure.



Sparse identification of nonlinear dynamics for model predictive control in the low-data limit

The recent sparse identification of nonlinear dynamics (SINDY) modelling procedure is extended to include the effects of actuation and it is demonstrated that the resulting SINDY-MPC framework has higher performance, requires significantly less data, and is more computationally efficient and robust to noise than NN models, making it viable for online training and execution in response to rapid system changes.

DeepMPC: Learning Deep Latent Features for Model Predictive Control

This work presents DeepMPC, an online real-time model-predictive control approach designed to handle complex nonlinear dynamics tasks, using a novel deep architecture and learning algorithm, learning controllers for complex tasks directly from data.

Deep Dynamical Modeling and Control of Unsteady Fluid Flows

The proposed approach, grounded in Koopman theory, is shown to produce stable dynamical models that can predict the time evolution of the cylinder system over extended time horizons and is able to find a straightforward, interpretable control law for suppressing vortex shedding in the wake of the cylinders.

Model Predictive Control — Status and Challenges

Learning to Fly like a Bird

It is argued that machine learning will play an important role in the control design process for agile flight by building data-driven approximate models of the aerodynamics and by synthesizing high-performance nonlinear feedback policies based on these approximate models and trial-and-error experience.

Deep Learning and Model Predictive Control for Self-Tuning Mode-Locked Lasers

This work demonstrates the first integration of a deep-learning architecture with model predictive control (MPC) in order to self-tune a mode-locked fiber laser and builds a dynamical model of the laser and appropriate control law for maintaining robust, high-energy pulses despite a stochastically drifting birefringence.

Cluster-based feedback control of turbulent post-stall separated flows

The objective of the present work is not necessarily to suppress flow separation but to minimize the desired cost function to achieve enhanced aerodynamic performance.

Human-level control through deep reinforcement learning

This work bridges the divide between high-dimensional sensory inputs and actions, resulting in the first artificial agent that is capable of learning to excel at a diverse array of challenging tasks.