Disturbance-Adaptive Data-Driven Predictive Control: Trading Comfort Violations for Savings in Building Climate Control

TL;DR

The paper addresses improving HVAC energy efficiency while preserving occupant comfort by introducing Disturbance-Adaptive Data-Driven Predictive Control (DAD-DPC). It combines Willems' Fundamental Lemma-based data-driven predictive control with conformal-prediction-derived disturbance bounds and an online adaptation mechanism to guarantee an asymptotic average comfort-violation bound $\lim_{t\to\infty} \frac{1}{t}\sum_{i=1}^t v_i \le \alpha$ without requiring knowledge of disturbance distributions. A backup controller and a large operating set ensure feasibility, and the framework is instantiated for buildings with a practical design, including offline RB-DPC and online updates via Algorithm 1. The method is validated in four simulation cases using BOPTEST and in an occupied campus building (Polydome), achieving substantial energy savings at a modest target violation (e.g., $\alpha=5\%$) and demonstrating robust, low-commissioning data-driven control for real-world building climate control.

Abstract

Model Predictive Control (MPC) has demonstrated significant potential in improving energy efficiency in building climate control, outperforming traditional controllers commonly used in modern building management systems. Among MPC variants, Data-driven Predictive Control (DPC) offers the advantage of modeling building dynamics directly from data, thereby substantially reducing commissioning efforts. However, inevitable model uncertainties and measurement noise can result in comfort violations, even with dedicated MPC setups. This paper introduces a Disturbance-Adaptive DPC (DAD-DPC) framework that ensures asymptotic satisfaction of predefined violation bounds without knowing the uncertainty and noise distributions. The framework employs a data-driven pipeline based on Willems' Fundamental Lemma and conformal prediction for application in building climate control. The proposed DAD-DPC framework was validated through four building cases using the high-fidelity BOPTEST simulation platform and an occupied campus building, Polydome. DAD-DPC successfully regulated the average comfort violations to meet pre-defined bounds. Notably, the 5%-violation DAD-DPC setup achieved 30.1%/11.2%/27.1%/20.5% energy savings compared to default controllers across four cases. These results demonstrate the framework's effectiveness in balancing energy consumption and comfort violations, offering a practical solution for building climate control applications.

Figures (17)

• Figure 1: Illustration of the DAD-DPC. Offline phase: Historical io data are used to construct a disturbance bound estimator and a DPC controller. Online phase: The disturbance bound is adaptively updated based on the current violation condition, and the DPC then utilizes the updated bound and the latest io measurement to compute the optimal input, which is applied to the building system.
• Figure 2: Illustration of the switching rule \ref{['eqn:dad_mpc_u']} in Algorithm \ref{['alg:dad_dpc']}. Colored segments mark which input is activated along three example trajectories.The backup controller $\pi^B$ is activated either when $y_t$ remains within $\mathcal{Y}_{\text{lim}}$ but the constraint is violated for too long such that $\bar{\alpha}_t = 0$, or when $y_t$ leaves $\mathcal{Y}_{\text{lim}}$. The set $\mathcal{Y}_{\text{lim}} \supseteq \mathcal{Y}_t$ defines the precondition under which $\pi^B$ operates according to Property 1, and its design methods are discussed in Remark \ref{['remark:assum2&3']}.
• Figure 3: Illustration of five zones in Case 3 deru2008doe.
• Figure 4: Case 1: trajectories of indoor temperature, relative energy consumption, average violations, relative disturbance bound (sub-figures from top to bottom).
• Figure 5: Case 2: trajectories of indoor temperature, relative energy consumption, average violations, relative disturbance bound (sub-figures from top to bottom).

Theorems & Definitions (10)

• Remark 1
• Lemma 1
• proof
• Lemma 2
• proof
• Theorem 3
• proof
• Lemma 4
• Remark 2: Computational complexity
• Remark 3: Other data-driven methods