[Paper Review] Data-driven modeling, control and tools for cyber-physical energy systems

Madhur Behl, Achin Jain+1 more|arXiv (Cornell University)|Apr 11, 2016

Smart Grid Energy Management17 references23 citations

TL;DR

This paper proposes a data-driven control framework, mbCRT, for demand response (DR) in large buildings using regression trees to predict baselines and synthesize optimal control actions. It achieves 17% better performance than rule-based DR, reducing curtailment by 380 kW and saving over $45,000, and is integrated into the open-source DR-Advisor tool with 92.8–98.9% prediction accuracy across 8 buildings.

ABSTRACT

Demand response (DR) is becoming important as the volatility on the grid continues to increase. Current DR approaches are either completely manual or involve deriving first principles based models which are extremely cost and time prohibitive to build. We consider the problem of data-driven DR for large buildings which involves predicting the demand response baseline, evaluating fixed DR strategies and synthesizing DR control actions. We provide a model based control with regression trees algorithm (mbCRT), which allows us to perform closed-loop control for DR strategy synthesis for large buildings. Our data-driven control synthesis algorithm outperforms rule- based DR by 17% for a large DoE commercial reference building and leads to a curtailment of 380 kW and over $45,000 in savings. Our methods have been integrated into an open source tool called DR-Advisor, which acts as a recommender system for the building's facilities manager and provides suitable control actions to meet the desired load curtailment while maintaining operations and maximizing the economic reward. DR-Advisor achieves 92.8% to 98.9% prediction accuracy for 8 buildings on Penn's campus. We compare DR-Advisor with other data driven methods and rank 2nd on ASHRAE's benchmarking data-set for energy prediction.

Motivation & Objective

• Address the high cost and time required to build first-principles models for demand response (DR) in large buildings.
• Develop a scalable, data-driven approach to DR that enables closed-loop control without relying on manual or physics-based modeling.
• Improve DR performance beyond rule-based strategies by synthesizing optimal control actions using historical data.
• Create an open-source tool, DR-Advisor, to assist facilities managers with real-time, economically optimal DR decisions.

Proposed method

• Propose a model-based control with regression trees (mbCRT) algorithm that learns from historical building data to predict baseline energy consumption.
• Use regression trees to model complex, nonlinear relationships between building operations and energy use for accurate baseline prediction.
• Integrate mbCRT into a closed-loop control framework that dynamically synthesizes DR control actions based on real-time feedback.
• Develop DR-Advisor as an open-source, recommender system that provides actionable control strategies to facilities managers.
• Train and validate the model on real data from 8 buildings at the University of Pennsylvania, using a data-driven approach to avoid reliance on detailed building physics.
• Benchmark performance against rule-based DR and other data-driven methods using ASHRAE’s standard benchmarking dataset.

Experimental results

Research questions

• RQ1 Can a data-driven control synthesis method outperform rule-based DR strategies in terms of economic savings and curtailment efficiency?
• RQ2 To what extent can regression trees accurately predict baseline energy use in large commercial buildings without first-principles modeling?
• RQ3 How effective is the mbCRT algorithm in enabling closed-loop control for demand response in real-world building operations?
• RQ4 What is the predictive accuracy and economic impact of the DR-Advisor tool across diverse building types on a real campus network?
• RQ5 How does the proposed method compare to other data-driven approaches on standardized benchmarking datasets?

Key findings

• The mbCRT-based data-driven control approach outperforms rule-based DR by 17% in economic performance for a large DoE commercial reference building.
• The method achieves a load curtailment of 380 kW while generating over $45,000 in savings, demonstrating significant cost efficiency.
• DR-Advisor achieves 92.8% to 98.9% prediction accuracy across 8 buildings on the University of Pennsylvania campus, indicating strong generalization to real-world conditions.
• The tool ranks second on ASHRAE’s benchmarking dataset for energy prediction, validating its robustness against other data-driven methods.
• The integration of mbCRT into DR-Advisor enables real-time, economically optimal control decisions while maintaining building operations and occupant comfort.
• The open-source deployment of DR-Advisor enables scalable, low-cost DR implementation across large building portfolios without detailed modeling.