About

Constance Crozier is an assistant professor at Georgia Tech. Her research group focuses on understanding future electrified systems, and exploiting their flexibility.

Research topics

• Electrical engineering
• Engineering
• Business
• Geography
• Physics
• Economics
• Economic growth
• Environmental economics
• Automotive engineering
• Natural resource economics

Selected publications

• A Sensitivity Analysis of Flexibility from GPU-Heavy Data Centers

  arXiv (Cornell University) · 2026-03-29

  articleOpen access

  The rapid growth of GPU-heavy data centers has significantly increased electricity demand and creating challenges for grid stability. Our paper investigates the extent to which an energy-aware job scheduling algorithm can provide flexibility in GPU-heavy data centers. Compared with the traditional first-in first-out (FIFO) baseline, we show that more efficient job scheduling not only increases profit, but also brings latent power flexibility during peak price period. This flexibility is achieved by moving lower energy jobs, preferentially executing jobs with lower GPU utilization and smaller node requirements, when the electricity price is high. We demonstrate that data centers with lower queue length and higher variance in job characteristics such as job GPU utilization and job size, offer the greatest flexibility potential. Finally we show that data center flexibility is highly price sensitive, a 7% demand reduction is achieved with a small incentive, but unrealistically high prices are required to achieve a 33% reduction.

  PublisherOA PDF

• A Sensitivity Analysis of Flexibility from GPU-Heavy Data Centers

  arXiv (Cornell University) · 2026-03-29

  preprintOpen access

  The rapid growth of GPU-heavy data centers has significantly increased electricity demand and creating challenges for grid stability. Our paper investigates the extent to which an energy-aware job scheduling algorithm can provide flexibility in GPU-heavy data centers. Compared with the traditional first-in first-out (FIFO) baseline, we show that more efficient job scheduling not only increases profit, but also brings latent power flexibility during peak price period. This flexibility is achieved by moving lower energy jobs, preferentially executing jobs with lower GPU utilization and smaller node requirements, when the electricity price is high. We demonstrate that data centers with lower queue length and higher variance in job characteristics such as job GPU utilization and job size, offer the greatest flexibility potential. Finally we show that data center flexibility is highly price sensitive, a 7% demand reduction is achieved with a small incentive, but unrealistically high prices are required to achieve a 33% reduction.

  PublisherDOI

• Learning a Local Trading Strategy: Deep Reinforcement Learning for Grid-scale Renewable Energy Integration

  Proceedings of the ... Annual Hawaii International Conference on System Sciences/Proceedings of the Annual Hawaii International Conference on System Sciences · 2025-01-01 · 1 citations

  articleOpen accessSenior author

  Variable renewable generation increases the challenge of balancing power supply and demand. Grid-scale batteries co-located with generation can help mitigate this misalignment. This paper explores the use of reinforcement learning (RL) for operating grid-scale batteries co-located with solar power. Our results show RL achieves an average of 61% (and up to 96%) of the approximate theoretical optimal (non-causal) operation, outperforming advanced control methods on average. Our findings suggest RL may be preferred when future signals are hard to predict. Moreover, RL has two significant advantages compared to simpler rules-based control: (1) that solar energy is more effectively shifted towards high demand periods, and (2) increased diversity of battery dispatch across different locations, reducing potential ramping issues caused by super-position of many similar actions.

  PublisherOA PDFDOI

• Battery Swapping Stations for Long Haul Freight Charging Considering an Electrified Supply Chain

  2025-10-26

  articleSenior author

  Electrification of heavy goods vehicles currently represents the most promising pathway to decarbonizing long-haul freight. The larger battery sizes compared to light-duty vehicles mean that charging times are likely to slow down freight operation significantly. Battery swapping stations, where the empty battery is swapped with another pre-charged battery, represent a promising alternative. In this paper we develop a model to determine the optimal logistics system, including the purchase of trucks, battery swapping stations, chargers, and warehouse space. By taking this integrated approach, we can understand the trade-offs of between battery swapping and other components of the supply chain. Using a case study involving seven cities, we show that battery swapping stations are installed when battery costs account for less than 40% of the truck cost. We show that the inclusion of swappable batteries results in a smaller truck fleet. The number of surplus batteries required depends strongly on the charging speed; with 500 MW (the current fastest chargers), around 1.2 batteries per truck were selected. We find that battery swapping increases utilization of renewable energy by less than 1%; the high utilization of trucks and batteries does not provide much energy flexibility in the considered case.

  PublisherDOI

• Least-regrets approach to evaluate inter-hemispheric HVDC connections

  SSRN Electronic Journal · 2025-01-01

  preprintOpen access
  PublisherDOI

• Distribution grids may be a barrier to residential electrification

  Cell Reports Sustainability · 2025-09-17 · 1 citations

  articleOpen access

  <h2>Summary</h2> Replacing fossil-fueled appliances and vehicles with electric alternatives can reduce greenhouse gas emissions and air pollution in many settings. However, electrification can also raise electricity demand beyond the safe limits of electrical infrastructure. This can increase the risk of blackouts or may require grid reinforcement that is often slow and expensive. Here, we estimate the physical and economic impacts on distribution grids of electrifying all housing and personal vehicles in each county of the lower 48 states of the United States. We find that space heating is the main driver of grid impacts, with the coldest regions seeing demand peaks up to five times higher than today's peaks. Accommodating electrification of all housing and personal vehicles is estimated to require 600 GW of distribution grid reinforcement nationally, at a cost of $350–$790 billion, or $2,800–$6,400 per household (95% confidence intervals). However, demand-side management could eliminate over two-thirds of grid reinforcement costs.

  PublisherOA PDFDOI

• On the Potential of Electrified Supply Chains to Provide Long Duration Demand Flexibility

  ArXiv.org · 2025-05-09

  preprintOpen access

  Demand flexibility can offset some of the variability introduced on the supply-side by variable renewable generation. However, most efforts (e.g. control of residential vehicle charging) focus on short durations -- typically on the scale of minutes to hours. This paper investigates whether a fully electrified supply chain (transport and manufacturing) could provide demand flexibility over longer durations, exploiting the latency that typically exists between the processing of raw material to the delivery of finished product. Using a case study of the cement industry along the East Coast of the United States, we demonstrate that electrified supply chains could shift gigawatt-hours (GWh) of electricity demand for durations of more than a week, largely following wind power variability. Furthermore, we show that this occurs using low levels of carbon taxing (below $50/tn), at which battery storage is not economically viable. A sensitivity analysis shows potential to provide flexibility in all considered cost scenarios, although where the flexibility comes from can change (e.g. transport vs manufacturing). We show that today's cost of electrified heavy goods vehicles are the most significant parameter -- with substantially lower costs yielding a more demand-flexible supply chain.

  PublisherOA PDFDOI

• Fast and robust strategies for large-scale mixed-integer SCOPF

  2025-02-28

  reportOpen accessSenior author

  This project develops scalable, computationally efficient algorithms to solve realistic large-scale power system optimization problems, including systems with more than 8,000 buses, as part of a larger series of competitions run by ARPA-E. These problems are critical because the secure and reliable operation of the power grid is becoming increasingly challenging, especially under conditions of increased uncertainty and variability. The economic feasibility of our methods is high, given that they are purely software-based solutions designed to operate power grids more efficiently. The technical effectiveness balances heuristics and approximations to provide a trade-off between speed and accuracy. Advancements in software for power grid operations are essential as operational reliability is increasingly threatened. Better computational tools can benefit the public by providing access to reliable energy in a robust manner. Our team’s approach leverages multiple techniques, including reasonable approximations of physics and data-driven methods, to reduce the complexity of the problem and uncover patterns in the chosen operational strategies. To address these challenges, we propose breaking the grid operation problem into two sequential sub-problems: the DC and AC modules. The DC module simplifies the problem using certain assumptions (discussed in our recent paper ,2) to speed up computation, optimizing binary and continuous variables. These binary results are then fixed to the derived value for the next step, the AC module. In the second step, we incorporate critical AC constraints to represent the power grid (see our paper for more detials1, ). Our two-pronged approach enforces critical limits, like ramping constraints, to prevent issues such as abrupt device shutdowns. The DC model (module 1 in Figure 1) is solved using the mixed-integer linear solver provided by the Gurobi. The nonlinearities and non-convexities of the AC model (second module of Figure 1) are tackled using the IPOPT solver. Figure 1 illustrates an abstract representation of considered constraints. The two sub-problems are designed to manage the computational workload of the underlying complex problem provided by ARPA-E. We use the following techniques to reformulate the original problem, eliminating quadratic terms by reformulation, relaxing nonconvex constraints, linearizing convex constraints, managing uncontrollable loads, and post-simulation calculation of power reserves. Additionally, our solution benefits from the structural representation of Jacobian and Hessian matrices, utilizing vectorized forms, and efficient storage methods for sparse matrices in a coordinated format. We have noticed that decomposition and specialized solver approach, alongside reformulation techniques, enhance both computational effectiveness and overall efficiency.

  PublisherOA PDFDOI

• Combined Bound Tightening on McCormick Relaxations of AC Optimal Power Flow

  2025-10-26

  article

  This paper investigates the effectiveness of combining valid inequalities and feasibility-based bound tightening (FBBT) to strengthen linear programming (LP) relaxations for the AC Optimal Power Flow (ACOPF) problem. We focus on reformulations using McCormick envelopes applied to the Second-Order Conic Programming (SOCP) relaxation of ACOPF. Several classes of valid inequalities, including ring cuts, reverse cone envelopes, and arctangent envelopes, are integrated into the relaxation. A combined tightening framework is proposed, beginning with bound tightening via SOCP relaxations, followed by adding valid inequalities and successive refinement of the McCormick envelopes with FBBT. Experimental results show that integrating arctangent envelopes with FBBT yields a tighter and more scalable relaxation. This approach significantly reduces the average SOCP gap from 8.14% to 4.11%, thereby improving the practical performance of convex relaxations for ACOPF.

  PublisherDOI

• Scalable Solutions for Security-Constrained Optimal Power Flow With Multiple Time Steps

  IEEE Transactions on Industry Applications · 2025-01-23 · 6 citations

  article

  This work introduces an innovative approach to scaling security-constrained optimal power flow problems to large power grids with multi-timestep, addressing the significant challenges associated with managing millions of continuous and integer optimization variables as well as nonlinear and nonconvex constraints. Through a strategic combination of problem reformulation, linearization methods, constraint-relaxation techniques, and sequential optimization, the complexities inherent to large power grid optimization are effectively navigated. The proposed methodology enables the resolution of complex power grid models with strict time constraints while attaining high-quality solutions. Demonstrating remarkable robustness, the novel approach consistently surpasses established benchmark methods.

  PublisherDOI

Frequent coauthors

• Malcolm McCulloch

  34 shared

• Kyri Baker

  30 shared

• Thomas Morstyn

  University of Oxford

  22 shared

• Dimitra Apostolopoulou

  Oxford Institute for Energy Studies

  15 shared

• Matthew Deakin

  Newcastle University

  10 shared

• John Montagu

  Applied Mathematics (United States)

  9 shared

• Cristina Torres-Machí

  Georgia Institute of Technology

  9 shared

• S. Curtis

  University of Colorado Boulder

  9 shared

Labs

• Constance CrozierPI

Awards & honors

• Anderson-Interface PhD Student Fellowship for Research Excel…
• PSERC grant
• Sustainability Next Seed Grant