On Direct vs Indirect Data-Driven Predictive Control

Krishnan, Vishaal; Pasqualetti, Fabio

doi:10.1109/cdc45484.2021.9683187

Cited by 43 publications

(32 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Proof: It can be easily seen that both solutions satisfy linear equations (10). In addition, for the former SPC solution, it holds that…”

Section: A Generalized Fundamental Lemma and Data-driven Optimal Pred...mentioning

confidence: 95%

“…This assures consistency, that is, under perfect data there is no bias in the regularized solution to (13). In contrast, heterogeneous solutions to (10) will be penalized by norm-based regularizers, which is an unwanted effect from an identification perspective. In addition, it was pointed out by [23] that problem (13) with R proj (•) can be regarded as a convex relaxation of the SPC problem (12) for λ sufficiently small.…”

Section: Remedies For Stochastic Lti Systemsmentioning

confidence: 99%

“…As compared to (10) in SPC, the system of linear equations (29) contains two extra blocks of constraints, which encode innovation information and strengthen our understanding of data-driven control of stochastic systems. On the one hand, it is conceptually beneficial to add more hard/soft constraints to mitigate the noise effect.…”

Section: A Generalized Fundamental Lemma and Data-driven Optimal Pred...mentioning

confidence: 99%

“…Thus, both the cumbersome identification of parametric models and the construction of an online state observer are no longer needed. Besides its easy implementation, there is an additional subtlety that the bias induced by inadequate model fitting can be effectively circumvented [10], leading to competitive performance with model predictive control (MPC). Thanks to these merits, DeePC has found miscellaneous applications across domains, such as power systems [11]- [14], motion control [15]- [17], and smart buildings [18].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Data-Driven Control of Stochastic Systems: An Innovation Estimation Approach

Wang¹,

Chen²,

Huang³

2022

Preprint

View full text Add to dashboard Cite

Recent years have witnessed a booming interest in the data-driven paradigm for predictive control. However, under noisy data ill-conditioned solutions could occur, causing inaccurate predictions and unexpected control behaviours. In this article, we explore a new route toward data-driven control of stochastic systems through active offline learning of innovation data, which gives an answer to the critical question of how to derive an optimal data-driven model from a noise-corrupted dataset. A generalization of the Willems' fundamental lemma is developed for non-parametric representation of input-outputinnovation trajectories, provided realizations of innovation are precisely known. This yields a model-agnostic unbiased output predictor and paves the way for data-driven receding horizon control, whose behaviour is identical to the "oracle" solution of certainty-equivalent model-based control with measurable states. For efficient innovation estimation, a new low-rank subspace identification algorithm is developed. Numerical simulations show that by actively learning innovation from input-output data, remarkable improvement can be made over present formulations, thereby offering a promising framework for data-driven control of stochastic systems.

show abstract

“…Proof: It can be easily seen that both solutions satisfy linear equations (10). In addition, for the former SPC solution, it holds that…”

Section: A Generalized Fundamental Lemma and Data-driven Optimal Pred...mentioning

confidence: 95%

Section: Remedies For Stochastic Lti Systemsmentioning

confidence: 99%

Section: A Generalized Fundamental Lemma and Data-driven Optimal Pred...mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Data-Driven Control of Stochastic Systems: An Innovation Estimation Approach

Wang¹,

Chen²,

Huang³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…In both cases, we compare the performance of the data-driven MHE to the performance of a model-based MHE where the model is identified using the same (offline) input, output and state measurements that are used in the offline phase of the data-driven MHE formulation. In the context of controller design, some recent works compare the performance of direct data-driven control frameworks to indirect data-driven control frameworks, where in the latter the offline data are used to identify a model of the system, which is then used for model-based control, see [38], [39]. The main outcome of these works is that the indirect approach typically performs better if the offline data are only corrupted by measurement noise.…”

Section: Application To Four-tank Systemmentioning

confidence: 99%

Robust Data-Driven Moving Horizon Estimation for Linear Discrete-Time Systems

M.¹,

Lopez²,

Müller³

2022

Preprint

View full text Add to dashboard Cite

In this paper, a robust data-driven moving horizon estimation (MHE) scheme for linear time-invariant discrete-time systems is introduced. The scheme solely relies on offline collected data without employing any system identification step. First, robust global exponential stability is proven under standard assumptions for a nominal case where the offline collected data are noise-free but the online measured outputs are corrupted by some non-vanishing measurement noise. Second, practical robust exponential stability is shown for the case where, in addition to the measurement noise in the online phase, the offline collected data are corrupted by some non-vanishing and bounded noise. The behavior of the novel robust data-driven MHE scheme is illustrated by means of a simulation example and compared to a standard model-based MHE, where the model is identified using the same offline data as for the data-driven MHE.

show abstract