Data-enabled Predictive Control
Data-enabled Predictive Control (DeePC) is a data-driven non-parametric algorithm for combined identification (learning) and control of dynamical systems. This algorithm avoids an explicit representation of the system dynamics and leverages the behavioural systems framework to implicitly describe the system trajectories using previously collected input-output data. Its receding horizon implementation is shown to be equivalent to an MPC formulation for the case of deterministic linear time-invariant (LTI) systems, while its application to nonlinear and stochastic systems is shown to be possible via different regularization techniques [1]-[3]. The cornerstones of our research are grounded in behavioral system theory, contemporary methods from statistical and machine learning, recent stochastic optimization methods, and classic control theory.
Research
Motivation
Control is the science of interfacing physical processes in feedback information technology. The design of control systems crucially relies on a model of the process, especially if controllers should be deployed in safety-critical environments, minimize the cost of operation, or perform robustly in face of uncertainty. The most expensive and time-consuming task in model-based control is actually identifying the model. An alternative approach entirely bypasses models and learns controllers directly from data. This data-driven approach is favored in many application domains, driven by advances in machine learning, and promises exciting developments such as the pixels-to-control paradigm enabling end-to-end automation in autonomous driving. However, to date neither the machine learning nor the control community know how to design data-driven controllers that perform reliably in safety-critical and real-time environments and are tractable in terms of computation and sample efficiency.
Theory
Our research focuses on Data-enabled Predictive Control (DeePC). DeePC is a method for deriving controllers directly from raw data that exploits a matrix of previously collected trajectories of the system which acts as an underlying implicit non-parametric model [1]. We have shown that our controller performs very well even in the presence of constraints, nonlinearity, and stochasticity, at an extremely low algorithmic complexity, and it outperforms a sequential identification and model-based control. Our goal is to (1) lay out rigorous theoretical foundations of data-driven and reliable control, (2) design robust numerical algorithms, and (3) deploy them in challenging and safety-critical systems. The cornerstones of our research are grounded in behavioral system theory, contemporary methods from statistical and machine learning, recent stochastic optimization methods, and classic control theory.
Applications
DeePC provides an effective way to optimally control complex systems without any modeling or identification step, and thus can be applied in many scenarios. In the power system context, we use DeePC to perform model-free damping control for large-scale power grids. Conventional damping control is model-based, which requires system modeling and may have inferior performance because the grid is ever-changing. By comparison, DeePC can potentially be adaptive to the grid condition and guarantee the damping performance in practice. Furthermore, DeePC has been successfully applied to real-world quadcopters for trajectory tracking, and to control an inverted pendulum.
"Data-Enabled Predictive Control (DeePC) of Quadcopters"video will follow.
Publications
[1] Coulson, J., Lygeros, J. and Dörfler, F., 2019,
June. Data-enabled predictive control: In the shallows of the DeePC. In 2019 18th European Control Conference (ECC) (pp. 307-312). IEEE.
[2] Coulson, J., Lygeros, J. and Dörfler, F., 2019,
December. Regularized and distributionally robust data-enabled predictive control. In 2019 IEEE 58th Conference on Decision and Control (CDC) (pp. 2696-2701). IEEE.
[3] Huang, L., Coulson, J., Lygeros, J. and Dörfler, F., 2019,
December. Data-enabled predictive control for grid-connected power converters. In 2019 IEEE 58th Conference on Decision and Control (CDC) (pp. 8130-8135). IEEE.
[4] Huang, L., Coulson, J., Lygeros, J. and Dörfler, F., 2019.
Decentralized Data-Enabled Predictive Control for Power System Oscillation Damping. arXiv preprint arXiv:1911.12151.
[5] Coulson, J., Lygeros, J. and Dörfler, F., 2020.
Distributionally Robust Chance Constrained Data-enabled Predictive Control. arXiv preprint arXiv:2006.01702.
[6] Elokda, E., Coulson, J., Beuchat, P., Lygeros, J. and Dörfler, F., 2019.
Data-Enabled Predictive Control for Quadcopters.
[7] Carlet, P.G., Favato, A., Bolognani, S. and Dörfler, F., 2020, October.
Data-driven predictive current control for synchronous motor drives. In 2020 IEEE Energy Conversion Congress and Exposition (ECCE) (pp. 5148-5154). IEEE.
[8] Alpago, D., Dörfler, F. and Lygeros, J., 2020.
An Extended Kalman Filter for Data-enabled Predictive Control. IEEE Control Systems Letters.
People
Professors
-John Lygeros (Head of Automatic Control Laboratory)
-Florian Dörfler (Associate Professor)
Postdoc
-Linbin Huang
-Mathias Hudoba de Badyn
external page-Soroosh Shafieezadeh Abadeh
external page-Jianzhe Zhen
PhD-candidate
-Jermey Coulson
-Francesco Micheli
Presentations
• Jeremy Coulson, “Data-Enabled Predictive Control: In the Shallows of the DeePC”, ECC 2019
•Jeremy Coulson, “Data-Enabled Predictive Control: In the Shallows of the DeePC”, ABB Annual Meeting 2019.
•Jeremy Coulson, “Distributionally Robust Data-Enabled Predictive Control”, INFORMS 2019.
•Jeremy Coulson, “Regularized and Distributionally Robust Data-Enabled Predictive Control”, CDC 2019.
•Jeremy Coulson, “Regularized and Distributionally Robust Data-Enabled Predictive Control”, Peking University Control Seminar 2020.
•Jeremy Coulson, “Regularized and Distributionally Robust Data-Enabled Predictive Control”, Robust Optimization Webinar 2020.
•Jeremy Coulson, “Regularized and Distributionally Robust Data-Enabled Predictive Control”, Sparse Learning Workshop 2020. Video, Slides
•John Lygeros, “Data Enabled Predictive Control: Stochastic Systems and Implicit Dynamic Predictors”, L4DC Keynote 2020. external pageVideo
•Florian Dörfler, “Data Enabled Predictive Control”, CDC 2020. Video and Slides
•Linbin Huang, “Data-Enabled Predictive Control for Grid-Connected Power Converters”, CDC, 2020.
•Linbin Huang, “Data-Enabled Predictive Control for Grid-Connected Power Converters”, ABB Annual Meeting 2019. external pageSlides
•Linbin Huang, “Data-Enabled Predictive Control for Grid-Connected Power Converters”, Peking University Control Seminar 2020.
Research supported by
NCCR and ETH funds