COHORT: Coordination of Heterogeneous Thermostatically Controlled Loads for Demand Flexibility

Bingqing Chen, Jonathan Francis, Marco Pritoni, Soummya Kar, Mario Bergés

TL;DR

This paper introduces COHORT, a novel distributed control framework for coordinating a large population of heterogeneous thermostatically controlled loads (TCLs) like air conditioners. COHORT enables these devices to collectively provide grid services by optimizing a grid-level objective while respecting individual comfort constraints, achieving significant peak load reduction and improved grid flexibility.

What Problem Does This Paper Address?

The increasing demand for electricity, coupled with the growing integration of renewable energy sources, presents a significant challenge for grid operators. Balancing supply and demand in real-time is becoming more complex. Thermostatically controlled loads (TCLs), such as air conditioners, refrigerators, and water heaters, represent a substantial portion of electricity consumption and possess inherent flexibility due to their thermal inertia. However, effectively harnessing this flexibility for grid services is hindered by two main challenges: (1) the sheer scale of coordinating millions of individual devices, leading to a massive state-action space, and (2) the combinatorial and non-convex nature of individual TCL control (on/off states), which makes traditional optimization difficult. Existing methods often rely on simplified aggregate models with low fidelity or centralized approaches that raise privacy concerns and are computationally expensive.

Methods

COHORT employs a distributed control framework that leverages the Alternating Direction Method of Multipliers (ADMM) to coordinate a population of TCLs. The core idea is to decompose the grid-level optimization problem into smaller, manageable subproblems that can be solved in parallel at each TCL. This approach preserves privacy as each TCL only needs to share its local information.

Key methodological contributions include:

  • Convex Relaxation of TCL Flexibility: The paper addresses the non-convexity of TCL constraints by characterizing the set of all admissible power profiles for each TCL as a convex set through relaxation. This makes the optimization problem computationally tractable, even for long planning horizons.
  • ADMM for Distributed Optimization: ADMM is used to coordinate the solutions of individual TCL subproblems to find a global optimum for a grid-level objective. The paper utilizes a more efficient variant of ADMM that operates on the mean of decision variables.
  • Pulse Width Modulation (PWM) for Actuation: After the continuous power trajectory is determined through optimization, a low-frequency PWM strategy is used to translate these continuous values back into discrete on/off signals for the TCLs. This method is inspired by previous work and allows for quasi-continuous output from on/off actuators.
  • Generalizable Framework: COHORT is designed to be generalizable to various grid objectives, such as generation following, minimizing ramping, and peak load curtailment, by simply reformulating the grid-level objective function.

Datasets

  • CAISO Load Profiles: Used for simulation studies in the generation following and minimizing ramping use cases.
  • PJM Load Profiles: Used for the peak load curtailment use case in the hardware-in-the-loop (HIL) simulation.
  • ecobee's Donate Your Data dataset: Utilized to model residential ACs for the HIL simulation, providing realistic thermodynamics and population heterogeneity.

Key Insights and Ideas

  • Scalability through Distribution: The distributed nature of COHORT, enabled by ADMM, is crucial for managing millions of TCLs. The computational burden is shared, and privacy is maintained, overcoming limitations of centralized approaches.
  • Convexity is Key for Long Horizons: Relaxing the TCL flexibility to a convex set is a critical step that allows for computationally efficient optimization over extended planning horizons (e.g., 24 hours), which is essential for tasks like load shifting.
  • PWM Bridges the Gap: The PWM strategy effectively translates the continuous, optimized power signals into practical on/off commands for TCLs, making the framework directly applicable to real-world devices.
  • Real-World Validation Matters: The HIL simulation, including a real ecobee smart thermostat and a physical AC unit, provides strong evidence of COHORT's practical applicability and performance beyond purely simulated environments.
  • Comfort is Achievable: The experiments demonstrate that significant grid benefits can be achieved without compromising occupant comfort, as evidenced by occupant satisfaction in the real-world testbed.

Key Results

  • Peak Load Reduction: In the HIL simulation over a 15-day period, COHORT reduced daily peak loads by an average of 12.5% (with a 2.9% standard deviation), assuming TCLs constitute 20% of the total load.
  • Ramping Reduction: In the simulation study for minimizing ramping, COHORT achieved a 23.1% reduction compared to a baseline scenario.
  • Generation Following: The framework demonstrated comparable performance to prior work in tracking generation following signals.
  • Comfort Maintained: Occupants in the real-world testbed reported no discomfort and could not distinguish between COHORT operation and standard on/off control.
  • Scalability: The number of ADMM iterations for convergence was observed to be largely independent of population size, indicating good scalability.

Limitations and Future Work

  • Number of Iterations: While ADMM is efficient, a significant number of iterations can still be required for convergence, especially for long planning horizons. This could raise concerns about communication costs.
  • Initialization for Faster Convergence: The authors suggest that initializing COHORT with an approximate solution from imitation learning could reduce the number of iterations needed for consensus.
  • Extension to Other Loads: The framework is generalizable and could be extended to coordinate other types of flexible loads beyond TCLs, with tailored tracking strategies.
  • Real-World Heterogeneity: While the HIL simulation included a real AC and simulated ACs based on real data, further validation with a larger and more diverse population of real-world devices would be beneficial.

Summary updated

Abstract

Demand flexibility is increasingly important for power grids. Careful coordination of thermostatically controlled loads (TCLs) can modulate energy demand, decrease operating costs, and increase grid resiliency. We propose a novel distributed control framework for the Coordination Of HeterOgeneous Residential Thermostatically controlled loads (COHORT). COHORT is a practical, scalable, and versatile solution that coordinates a population of TCLs to jointly optimize a grid-level objective, while satisfying each TCL's end-use requirements and operational constraints. To achieve that, we decompose the grid-scale problem into subproblems and coordinate their solutions to find the global optimum using the alternating direction method of multipliers (ADMM). The TCLs' local problems are distributed to and computed in parallel at each TCL, making COHORT highly scalable and privacy-preserving. While each TCL poses combinatorial and non-convex constraints, we characterize these constraints as a convex set through relaxation, thereby making COHORT computationally viable over long planning horizons. After coordination, each TCL is responsible for its own control and tracks the agreed-upon power trajectory with its preferred strategy. In this work, we translate continuous power back to discrete on/off actuation, using pulse width modulation. COHORT is generalizable to a wide range of grid objectives, which we demonstrate through three distinct use cases: generation following, minimizing ramping, and peak load curtailment. In a notable experiment, we validated our approach through a hardware-in-the-loop simulation, including a real-world air conditioner (AC) controlled via a smart thermostat, and simulated instances of ACs modeled after real-world data traces. During the 15-day experimental period, COHORT reduced daily peak loads by an average of 12.5% and maintained comfortable temperatures.
BibTeX
@inproceedings{chen2020cohort,
  author = {Bingqing Chen and Jonathan Francis and Marco Pritoni and Soummya Kar and Mario Bergés},
  title = {{COHORT: Coordination of Heterogeneous Thermostatically Controlled Loads for Demand Flexibility}},
  booktitle = {BuildSys},
  year = {2020},
  doi = {10.1145/3408308.3427980},
}
Figure