Adaptive Robust Data-driven Building Control via Bi-level Reformulation: an Experimental Result

Abstract

In the era of digitalization, utilization of datadriven control approaches to minimize energy consumption of residential/commercial building is of far-reaching significance. Meanwhile, 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 addresses the key noise-free assumption, and extends these data-driven control schemes to adaptive building control with measured process noise and unknown measurement noise via a robust bilevel 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 a numerical example and a real-world experiment on a lecture hall on EPFL campus.

Data-Driven Prediction with Stochastic Data: Confidence Regions and Minimum Mean-Squared Error Estimates

“…Subspace predictor [10], [19]: the solution of the algorithm is exactly the same as that for the noise-free case (7). However, the interpretation here is different.…”

“…Wasserstein distance minimization [7]: this algorithm finds the vector g that minimizes the Wasserstein distance between the stochastic distribution of y ini and that of Y p g:…”

“…This framework is especially suitable for predictive control, where multiple data-driven algorithms have been developed, including subspace predictive control [4], data-enabled predictive control [5], and behavioral input-output parametrization [6]. Successful applications have been described in [7], [8].…”

“…Recently, the extension of the fundamental lemma to stochastic data from a system identification point of view has been drawing increasing interest [9]. Such work includes model predictive control based on the prediction error method [10], maximum likelihood signal matrix model [11], [12], and a Wasserstein distance minimization approach [7].…”

Experimental Data-Driven Model Predictive Control of a Hospital HVAC System During Regular Use

“…Although defining twolevel temperature envelopes is common for residences and offices (see e.g. [34,35]), some employees still make use of the surgery center spaces during night hours and, thus, the indoor temperature has to stay below T max even then. Furthermore, this is to be done while minimizing the chiller energy consumption as dictated by its coefficient of performance (COP) curve and the building thermal load.…”