[Paper Review] Adaptive Robust Data-driven Building Control via Bi-level Reformulation: an Experimental Result

Ying Zhao Lian, Jicheng Shi+2 more|arXiv (Cornell University)|Jun 10, 2021

Control Systems and Identification43 references20 citations

TL;DR

This paper proposes a robust bi-level data-driven control framework for building HVAC systems using Willems' fundamental lemma, enabling adaptive, noise-robust predictive control without prior system modeling. It achieves 18.4% lower energy consumption than an industry-standard controller in a real-world 20-day experiment while ensuring occupant comfort and constraint satisfaction through active excitation and trajectory prediction under noisy measurements.

ABSTRACT

Data-driven control approaches for the minimization of energy consumption of buildings have the potential to significantly reduce deployment costs and increase uptake of advanced control in this sector. A number of recent approaches based on the application of Willems' fundamental lemma for data-driven controller design from input/output measurements are very promising for deterministic LTI systems. This paper \change{proposes a systematic way to handle unknown measurement noise and measurable process noise}, and extends these data-driven control schemes to adaptive building control via a robust bi-level formulation, whose upper level ensures robustness and whose lower level guarantees prediction quality. Corresponding numerical improvements and an active excitation mechanism are proposed to enable a computationally efficient reliable operation. The efficacy of the proposed scheme is validated by \change{a multi-zone building simulation} and a real-world experiment on a single-zone conference building on the EPFL campus. The real-world experiment includes a 20-day non-stop test, where, without extra modeling effort, our proposed controller improves 18.4\% energy efficiency against an industry-standard controller, while also robustly ensuring occupant comfort.

Motivation & Objective

• To develop a practical, adaptive, and robust data-driven control scheme for building HVAC systems that handles unknown measurement noise and measurable process disturbances like outdoor temperature and solar radiation.
• To extend data-driven control based on Willems’ fundamental lemma to time-varying systems under uncertainty, ensuring robustness and prediction accuracy.
• To enable online, computationally efficient operation through active excitation and efficient matrix inversion updates for real-time deployment.
• To validate the controller’s performance in both simulation and real-world experiments, demonstrating energy savings and robustness without additional modeling.

Proposed method

• Formulates a bi-level optimization problem: the upper level ensures robustness against measurement noise via a Wasserstein distance-based uncertainty set, while the lower level ensures accurate trajectory prediction using input/output data.
• Introduces a novel trajectory prediction method that minimizes an upper bound on the Wasserstein distance between the true system trajectory and noisy measurements.
• Applies the Willems’ fundamental lemma to represent system behavior from historical input/output data, enabling model-free controller design.
• Incorporates an active excitation mechanism (Algorithm 1) to maintain persistent excitation and ensure data quality for online Hankel matrix updates.
• Uses regularization and convex relaxation techniques to handle noise in the data-driven system identification process, linking it to distributional robustness.
• Employs efficient online matrix inversion updates to maintain computational feasibility during real-time operation.

Experimental results

Research questions

• RQ1 Can a data-driven controller based on Willems’ fundamental lemma be made robust to unknown measurement noise and measurable process disturbances in building control?
• RQ2 How can the bi-level formulation be structured to simultaneously ensure robustness and prediction accuracy in time-varying building systems?
• RQ3 What computational and data quality mechanisms are required to enable reliable, real-time deployment of data-driven controllers in buildings?
• RQ4 To what extent can a data-driven controller outperform a standard industrial controller in terms of energy efficiency and constraint satisfaction?

Key findings

• The proposed controller achieved 18.4% lower energy consumption compared to the default industrial controller in a four-day real-world experiment under similar weather conditions.
• The controller ensured zero constraint violations over a 20-day continuous test, maintaining indoor temperature within comfort bounds despite significant outdoor temperature variations (14.6°C to 29.5°C).
• The active excitation scheme successfully maintained persistent excitation, ensuring data quality and enabling stable online updates of the Hankel matrix.
• The controller demonstrated strong adaptivity to changing weather conditions, including both heating and cooling periods, without requiring model re-estimation or re-tuning.
• The robust bi-level formulation enabled reliable performance under noisy measurements, with the upper-level optimization effectively bounding uncertainty via the Wasserstein distance.
• The numerical improvements, including efficient matrix inversion and regularization, allowed the controller to operate in real time with low computational overhead.