[Paper Review] Adaptive Data-Driven Prediction in a Building Control Hierarchy: A Case Study of Demand Response in Switzerland

Jicheng Shi, Yingzhao Lian+2 more|arXiv (Cornell University)|Jul 17, 2023

Smart Grid Energy ManagementEngineering3 citations

TL;DR

This paper proposes a data-driven predictive control framework based on Willems' Fundamental Lemma (DeePC) for adaptive building control in Switzerland’s Demand Response program. It integrates seamlessly into a three-layer hierarchical control system for Secondary Frequency Control, reducing operating costs by 28.74% over 52 days with minimal tuning and automatic model updates using only days of operational data.

ABSTRACT

By providing various services, such as Demand Response (DR), buildings can play a crucial role in the energy market due to their significant energy consumption. However, effectively commissioning buildings for such desired functionalities requires significant expert knowledge and design effort, considering the variations in building dynamics and intended use. In this study, we introduce an adaptive data-driven prediction scheme based on Willems' Fundamental Lemma within the building control hierarchy. This scheme offers a versatile, flexible, and user-friendly interface for diverse prediction and control objectives. We provide an easy-to-use tuning process and an adaptive update pipeline for the scheme, both validated through extensive prediction tests. We evaluate the proposed scheme by coordinating a building and an energy storage system to provide Secondary Frequency Control (SFC) in a Swiss DR program. Specifically, we integrate the scheme into a three-layer hierarchical SFC control framework, and each layer of this hierarchy is designed to achieve distinct operational goals. Apart from its flexibility, our approach significantly improves cost efficiency, resulting in a 28.74% reduction in operating costs compared to a conventional control scheme, as demonstrated by a 52-day experiment in an actual building. Our findings emphasize the potential of the proposed scheme to reduce the commissioning costs of advanced building control strategies and to facilitate the adoption of new techniques in building control.

Motivation & Objective

• To reduce the high commissioning cost and expert effort traditionally required to implement advanced building control strategies.
• To address the challenge of building model identification in Model Predictive Control (MPC) by replacing parametric models with data-driven alternatives.
• To enable flexible, user-friendly, and adaptive prediction for diverse control objectives in building energy systems.
• To validate the proposed method in a real-world building control setting for Secondary Frequency Control (SFC) within a Swiss Ancillary Services market.
• To demonstrate significant cost savings and improved operational efficiency through real-time data adaptation and minimal human intervention.

Proposed method

• Employs DeePC (Data-Enabled Predictive Control) based on Willems’ Fundamental Lemma to construct non-parametric, data-driven models from operational I/O data.
• Introduces a user-friendly tuning process with clear guidelines for selecting the Hankel matrix size T, reducing reliance on expert calibration.
• Implements an adaptive update pipeline that automatically refreshes the Hankel matrix using recent operational data, enabling real-time model adaptation.
• Integrates the DDP predictor into a three-layer hierarchical control architecture: day-ahead planner, predictive controller, and real-time actuator layer.
• Uses a bi-level DeePC formulation with a physical rule filter to ensure model consistency and prevent overestimation of flexibility.
• Applies quadratic programming (QP) and linear programming (LP) to optimize control actions under system constraints and market requirements.

Experimental results

Research questions

• RQ1 Can a data-driven prediction method based on DeePC significantly reduce the commissioning effort and tuning complexity of building control systems compared to traditional parametric modeling?
• RQ2 How effective is the adaptive update mechanism of DeePC in maintaining prediction accuracy over time with changing building dynamics?
• RQ3 To what extent can DeePC-based control reduce operating costs in real-world building operations for grid support services like Secondary Frequency Control?
• RQ4 How does the integration of physical constraints into the DeePC framework improve reliability and control performance in practice?
• RQ5 Can the proposed method achieve high performance with minimal historical data, making it suitable for rapid deployment in diverse building types?

Key findings

• The proposed DDP-based control scheme reduced operating costs by 28.74% compared to a conventional control scheme over a 52-day real-world experiment.
• The method required only 5 days of operational data for the predictive controller and 10 days for the day-ahead planner in 2022, significantly less than the 2 weeks and 1 week needed for parametric model identification in 2017.
• The adaptive update pipeline enabled continuous model refinement without human tuning, enhancing long-term prediction accuracy and system responsiveness.
• The user-friendly tuning guidelines for the Hankel matrix size T reduced the need for expert intervention, improving deployability across different buildings.
• The integration of physical rules into the bi-level DeePC framework prevented overestimation of flexibility and improved prediction reliability compared to unconstrained models.
• The method demonstrated strong performance in a real-world SFC application, successfully coordinating a building and energy storage system to provide grid services with high precision and cost efficiency.