Large-Load Demand Flexibility as Virtual Storage

arXiv:2607.04564 · eess.SY, cs.SY · Submitted 2026-07-06 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: We're looking at the paper: "Large-Load Demand Flexibility as Virtual Storage".

Title and authors: Rosa: So we’re talking about the paper "Large-Load Demand Flexibility as Virtual Storage," and the authors are Chandan Chaudhary, Mohammed Benidris, and Joydeep Mitra from Michigan State University. They’re trying to show how these massive industrial loads can be modeled in a way that connects them directly to battery storage behavior.

Dev: They introduce this virtual storage concept as the bridge between two different ways of looking at flexibility—the load side and the battery side—and they argue this equivalence is exact within their base model, meaning no approximation is needed there.

Taro: So, what does that equivalence actually mean for us when we’re thinking about how these systems operate in real life?

Rosa: It means that every single feasible trajectory for a large flexible load under those power bounds and energy window constraints is mathematically identical to the charge trajectory of a virtual storage device operating with unity accounting efficiency in the grid power balance.

Dev: That's a big statement because it implies that if you know how to model the load, you automatically have a model for storage, and they can then optimize both at once.

The paper's summary: Rosa: The main thing they summarize is that this VS equivalence lets them move away from having two separate formulations that don't share any common mathematical structure. Instead, they use this connection to create a single linear program where you can jointly optimize the virtual storage and the actual battery energy storage system dispatch.

Dev: They also talk about how if you have a portfolio of N large loads sitting next to one BESS, they can use something called a Minkowski-sum aggregate representation.

Taro: So, how does that aggregation help simplify things for someone trying to run this kind of optimization?

Rosa: It reduces the constraint count from O of N times T to just O of T, which is a huge reduction when you have a really large number of loads in your portfolio. This allows one single linear program to jointly optimize both resources, and its optimality conditions give you a co-dispatch price for both.

Dev: And they mention that this aggregation is exact as long as all the loads share the same normalized energy windows across the whole group.

The paper's improvements: Rosa: The paper points out that while their base model is good, there are ways to extend it for more real-world fidelity. For instance, they look at technologies like electrolyzers or aluminum potlines and show how the base model fits those well, but then they add extensions for things like hot standby modes or rolling thermal-energy constraints.

Dev: It highlights that while the base model captures the main intertemporal coupling for these loads, fidelity often requires adding more technology-specific constraints to capture things like service level agreements or cooling headroom.

Taro: So, they’re saying the base model is a conservative lower bound, and you need those extra bits to get closer to reality when dealing with different types of industrial flexibility.

Rosa: They also discuss how the outer approximation they use for the aggregate flexibility set Fouter might be a strict superset of the true aggregate feasibility set Fagg precisely because the three loads they tested have structurally dissimilar flexibility signatures.

Conclusion: Dev: To wrap up, this paper establishes that you can treat any large flexible load's curtailment feasibility set as a virtual storage device charge trajectory, which allows for joint optimization with a BESS.

Rosa: And the result is that co-dispatch outperforms BESS-only dispatch on procurement cost savings, and the VS efficiency advantage is significant, delivering three hundred sixty-two point seven megawatt hours per day in their tested case.

Taro: I'm interested in what they say about the shadow price of the curtailment budget remaining stable over the study period and tracking the peak-price band onset instead of just a daily peak price.

Dev: That’s important because it suggests that this shadow price, lambda D bar, could actually be used as a settlement signal for day-ahead curtailment contracts.

Rosa: Exactly, and they point out that the capacity for this virtual storage comes entirely from production commitments already held by industrial operators, so it doesn't require any new grid capital investment.

Taro: So it seems like the implication is that this growth complements physical storage investment rather than replacing it.

Dev: That’s right, and matching those normalized flexibility signatures across different loads tightens that outer approximation, suggesting a way to design portfolios for better results in the future.

Michigan State University

eess.SY, cs.SY

Submitted: 2026-07-06

Updated: 2026-11-16

Comments: Submitted for 24th National Power System Conference NPSC 2026, 12-14 December, 2026

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 79/100

The gist: Large loads are scheduled through per-interval power bounds and horizon energy windows, whereas co-located battery energy storage systems (BESS) operate under state-of-charge dynamics…

Key concepts

Virtual Storage (VS) Equivalence
This establishes that any feasible schedule for a large flexible load under power and energy constraints is mathematically identical to the charge trajectory of a virtual storage device. This means the load's flexibility can be treated as an equivalent storage asset in grid balancing calculations, operating with unity efficiency.
Minkowski-sum Aggregate Representation
This technique combines the flexibility constraints of multiple loads (a portfolio) into a single, simpler set of constraints. It reduces the constraint count from O(NT) to O(T), allowing a single linear program to handle many loads efficiently without needing separate optimization for each one.
Curtailment Shadow Price Signal
The shadow price associated with curtailment (the cost of not using excess power) is shown to track the onset of the peak-price band rather than the daily peak. This provides a stable, systematic signal ($\lambda D_{\bar{}}$) that can be used in day-ahead contracts, reflecting after-peak structure.

Terminology

Summary

Large loads are scheduled through per-interval power bounds and horizon energy windows, whereas co-located battery energy storage systems (BESS) operate under state-of-charge dynamics <ref:2607. p. 0045641,Abstract—Water electrolysis plants, hyperscale data centers, and aluminum potlines represent gigawatts of demand-side flexibility for bulk power system balancing, operational planning, and procurement services <ref:2607. p. 004564#pg2> The two formulations share no common mathematical structure, and the joint procurement value of co-located loads and storage goes unrealized <ref:2607. p. 0 p. 004564#pg2> This paper establishes the connection between the two formulations through a virtual storage (VS) equivalence <ref:2607. p. 0 p. 004564#pg2>.

The Core Equivalence

Every feasible largeload trajectory under power-bound and energy-window constraints is a valid charge trajectory of a VS device that operates at unity accounting efficiency in the grid power balance <ref:2607. p. 004564#pg2> The curtailment feasibility set of any large flexible load with a positive process floor, a rated power ceiling, and a horizon energy window is identical to the charge trajectory set of a virtual storage (VS) device with parameters derived from the load’s physical bounds <ref:2607. p. 004564#pg2> The equivalence is exact within the base model and requires no approximation <ref:2607. p. 2 p. 204564#pg2>.

Portfolio Aggregation

For a portfolio of N large loads co-located with a BESS, a Minkowski-sum aggregate representation reduces the load-side constraint count from O(NT) to O(T) <ref:2607.04564#pg2> A single linear program jointly optimizes VS and BESS dispatch, and its optimality conditions yield a co-dispatch price for both resources <ref:2607. p. 104564#pg2> Aggregation is exact when all loads share the same normalized energy windows <ref:2607. p. 504564#pg2>.

Joint Optimization Formulation

The joint scheduling problem for N large loads and one co-located BESS over horizon T is formulated as a single linear program <ref:2607.04564#pg2> The cost-minimization problem is min δagg, uc, ud, e X T t=1 h c t p¯tot−δ t agg−u t d+u t c +q t δagg aj∆t <ref:2607. p. 304564#pg2> The constraint reduction is exact under the proportionality condition, and otherwise yields a tractable outer approximation requiring ex-post disaggregation validation <ref:2607. p. 304564#pg2>.

Case Study Findings

Co-dispatch achieves a 29.89% procurement cost reduction against 3. p. 403% for BESS-Only <ref:2607. p. 004564#pg2> The VS efficiency advantage is 362. p. 47 MWh/day <ref:2607. p. 004564#pg2> The shadow price of the curtailment budget remains stable over the study period and tracks the peak-price band onset rather than the daily peak price <ref:2607. p. 504564#pg2>.

Implications for Market Design

The shadow price of the curtailment budget is a candidate settlement signal <ref:2607. p. 5 p. 504564#pg2> VS capacity derives entirely from production commitments that industrial operators already hold, so no additional grid capital investment is required <ref:2607. p. 5 p. 5 p. 5 p. 504564#pg2> This property makes it a candidate signal for day-ahead curtailment settlement <ref:2607. p. 504564#pg2>. The outer approximation Fouter is a strict superset of the true aggregate feasibility set Fagg precisely because the three loads have structurally dissimilar flexibility signatures <ref:2607. p. 504564#pg2> A portfolio in which all loads share the same normalized energy windows would satisfy the proportionality condition, collapse the outer approximation to an exact representation, and eliminate the disaggregation LP entirely <ref:2607. p. 5 p. 5 p. 504564#pg2>. The case study therefore identifies a portfolio design criterion <ref:2607. p. 504564#pg2>.

Conclusion

This paper has established that the curtailment feasibility set of any large flexible load under power-bound and energywindow constraints is identical to the charge-trajectory set of a VS device <ref:2607.04564#pg2> Co-dispatch outperformed BESSalone dispatch on procurement cost <ref:2607. p. 504564#pg2> VS delivered the larger share of savings because it carries no round-trip conversion loss <ref:2607. p. 504564#pg2>. The curtailment shadow price remained stable over the study period and tracked the peak-price band onset rather than the daily peak price <ref:2607. p. 504564#pg2>. VS capacity derives entirely from production commitments that industrial operators already hold, so no additional grid capital investment is required <ref:2607.04564#pg2>. The savings therefore quantified price-taking economic value rather than endogenous networkconstrained dispatch value <ref:2607. p. 504564#pg2>.

--- Page 1 ---

Large loads are scheduled through per-interval power bounds and horizon energy windows, whereas co-located battery energy storage systems (BESS) operate under state-of-charge dynamics <ref:2607.04564#pg2> The two formulations share no common mathematical structure, and the joint procurement value of co-located loads and storage goes unrealized <ref:2607.04564#pg2> This paper establishes the connection between the two formulations through a virtual storage (VS) equivalence <ref:2607.04564#pg2>. Every feasible largeload trajectory under power-bound and energy-window constraints is a valid charge trajectory of a VS device that operates at unity accounting efficiency in the grid power balance <ref:2607.04564#pg2> Production and service-level costs lie outside this abstraction and enter the dispatch through curtailment opportunity costs <ref:2607. p. 004564#pg2>. For a portfolio co-located with a BESS, aggregation reduces the constraint count from O(NT) to O(T) and yields a co-dispatch price for both resources <ref:2607. p. 004564#pg2> Validation on the IEEE RTS-GMLC with three representative load classes shows that virtual storage delivers the dominant share of joint procurement savings <ref:2607. p. 004564#pg2>. In the tested case, savings are additive because the two resources dispatch to nonoverlapping intervals, and the curtailment shadow price tracks the peak-price band onset rather than the daily peak price <ref:2607. p. 004564#pg2>.

--- Page 2 ---

Research on grid-scale storage has concentrated on optimizing charge and discharge decisions against uncertain DER output <ref:2607.04564#pg2> On the demand side, flexibility models for residential, commercial, and industrial loads have been surveyed in [4] <ref:2607. p. 004564#pg2> The aggregate flexibility of thermostatically controlled loads can be represented as a convex polytope in powertime space [5] <ref:2607. p. 004564#pg2> Flexibility envelopes have also been proposed as a planning tool that integrates supply-side and demand-side contributions [7] <ref:2607. p. 004564#pg2> However, none of these works establishes a formal equivalence between largeload curtailment trajectories and storage charge trajectories, nor do they provide a unified formulation to dispatch flexible loads jointly with co-located BESS <ref:2607. p. 004564#pg2> Large industrial and commercial loads occupy a position in this landscape that neither strand of prior literature addresses <ref:2607. p. 004564#pg2> What these loads share, and what prior work has not formalized, is a curtailment cycle that structurally mirrors the charging phase of a BESS <ref:2607. p. 004564#pg2> The mathematical formulation to achieve this in a unified form does not yet exist <ref:2607. p. 004564#pg2> This paper establishes the mathematical connection that enables coordinated operation of large flexible loads and colocated BESS in day-ahead scheduling and planning contexts <ref:2607. p. 004564#pg2>. The main contributions are as follows <ref:2607. p. 004564#pg2> 1) The curtailment feasibility set of any large flexible load with a positive process floor, a rated power ceiling, and a horizon energy window is identical to the chargetrajectory set of a virtual storage (VS) device with parameters derived from the load’s physical bounds <ref:2607. p. 004564#pg2> 2) For a portfolio of N large loads co-located with a BESS, a Minkowski-sum aggregate representation reduces the load-side constraint count from O(NT) to O(T), independent of portfolio size <ref:2607. p. 004564#pg2> 3) The method is evaluated on the IEEE RTS-GMLC [20] with three large-load classes <ref:2607. p. 104564#pg2>

--- Page 3 ---

A. Physical Structure of Large Flexible Loads

Three features distinguish large flexible loads from classical demand response <ref:2607.04564#pg2> First, power must remain above a strictly positive process floor <ref:2607. p. 104564#pg2> Second, total energy over the horizon must satisfy a minimum throughput requirement tied to the facility’s production commitment <ref:2607. p. 104564#pg2> Third, consumption is freely adjustable within the admissible power range <ref:2607. p. 104564#pg2>

Power Bounds and Energy Window

Let the planning horizon consist of T intervals given by T = 1,..., T, each of duration ∆t hours <ref:2607. p. 104564#pg2> At every interval, the load must operate between its process floor p > 0 and its rated power p¯: p ≤ p t ≤ p¯ ∀ t ∈ T <ref:2607. p. 104564#pg2> Over the full horizon, the total energy consumed must satisfy the energy window constraint Emin ≤ X T t=1 p t ∆t ≤ Emax <ref:2607. p. 104564#pg2> The load feasibility set is FL = p ∈ R T: (1) and (2) hold <ref:2607. p. 104564#pg2>

Deviation Variable and Curtailment Energy Bounds

The operator-relevant quantity is the deviation from rated consumption <ref:2607.04564#pg2> Substituting p t = ¯p−δ t into (2) converts the energy window into a curtailment energy bound: D ≤ X T t=1 δ t ∆t ≤ D, ¯ (6) <ref:2607. p. 104564#pg2>

--- Page 4 ---

D. Battery Energy Storage System

A co-located BESS is characterized by charge power u t c, discharge power u t d, and state of charge e t MWh at the end of interval t <ref:2607.04564#pg2> The state of charge evolves as e t = e t−1 + ηc∆t u tc − ∆t ηd u td, (10) <ref:2607. p. 104564#pg2>

Virtual Storage Device

The virtual storage (VS) device is defined by this charge-only trajectory, with initial state s 0 = 0, power rating ∆P, capacity CVS = D¯, minimum terminal energy D, and unity accounting efficiency <ref:2607.04564#pg2> The equivalence is exact within the base model and requires no approximation <ref:2607.04564#pg2>

Aggregate Flexibility for a Portfolio of Loads

For a portfolio of N large flexible loads with individual deviation sets Fδ,i parametrized by (∆Pi, Di, D¯ i), the aggregate flexibility is the Minkowski sum Fagg = LN i=1 Fδ i <ref:2607.04564#pg2> The additive aggregate parameters ∆Pagg = X i ∆Pi, Dagg = X i Di, D¯ agg = X i D¯ i (15) define the outer set Fouter = 0 ≤ δ t ≤ ∆Pagg, Dagg ≤ 1⊤δ∆t ≤ D¯ agg <ref:2607. p. 204564#pg2>

Joint Co-Dispatch Formulation

The cost-minimization problem is min δagg, uc, ud, e X T t=1 h c t p¯tot−δ t agg−u t d+u t c +q t δagg aj∆t <ref:2607.04564#pg2> Constraint Reduction: Replacing the disaggregated load constraints with the aggregate outer representation reduces the load-flexibility block from 2NT + 2N constraints to 2T + 2 = O(T) <ref:2607. p. 304564#pg2>

--- Page 5 ---

Fig. 1. Net load (MW) and LMP is for three scheduling scenarios, July 5 <ref:2607. p. 404564#pg2> Co-Dispatch superimposes both resources <ref:2607. p. 404564#pg2>

Dispatch Analysis

When q t = 0 and prices are positive, VS is preferred over BESS at the same interval in identical conditions, since it carries no conversion penalty <ref:2607.04564#pg2> Co-dispatch superimposes both resources <ref:2607. p. 404564#pg2> Curtailment and BESS discharge coincide at hours 18–19 to achieve the maximum procurement reduction <ref:2607. p. 404564#pg2>

Study Summary

The co-dispatch procurement cost savings total 1,850,094 over 14 days against a baseline of 5,475,478 <ref:2607.04564#pg2> The mean daily savings is 33. p. 49% <ref:2607. p. 004564#pg2> Disaggregation is infeasible on all 14 days <ref:2607. p. 404564#pg2>.

--- Page 6 ---

The virtual storage budget is set by industrial production economics, not by grid investment <ref:2607.04564#pg2> BESS capacity is a capital decision priced in /kWh <ref:2607. p. 4 p. 404564#pg2> The shadow price of the curtailment budget is a candidate settlement signal <ref:2607. p. 5 p. 504564#pg2> VS capacity derives entirely from production commitments that industrial operators already hold, so no additional grid capital investment is required <ref:2607.04564#pg2> This growth complements physical storage investment <ref:2607. p. 5 p. 504564#pg2>. The shadow price of the curtailment budget is a candidate settlement signal <ref:2607.04564#pg2> λD¯ equals the LMP at the curtailmentonset hour, not the daily peak price <ref:2607. p. 5 p. 504564#pg2> Because the VS exhausts its budget on every study day, λD¯ reflects the systematic afternoon peak-band structure <ref:2607. p. 5 p. 504564#pg2> A curtailment contract priced at λD¯ would be self-triggering on any standard summer day <ref:2607. p. 5 p. 504564#pg2>.

--- Page 7 ---

Disaggregation is infeasible on all 14 days <ref:2607.04564#pg2> The outer approximation Fouter is a strict superset of the true aggregate feasibility set Fagg precisely because the three loads have structurally dissimilar flexibility signatures <ref:2607. p. 5 p. 504564#pg2> A portfolio in which all loads share the same normalized energy windows would satisfy the proportionality condition, collapse the outer approximation to an exact representation, and eliminate the disaggregation LP entirely <ref:2607.04564#pg2> The case study therefore identifies a portfolio design criterion <ref:2607. p. 5 p. 504564#pg2> Matching normalized flexibility signatures across constituent loads tightens the aggregate approximation <ref:2607. p. 5 p. 504564#pg2>. Although the exogenous nodal LMPs reflect marginal congestion conditions from the underlying market solution, the present co-dispatch model does not feed VS and BESS decisions back into network flows <ref:2607. p. 5 p. 504564#pg2> The reported savings therefore represent price-taking economic value rather than a networkconstrained feasibility result <ref:2607. p. 5 p. 504564#pg2>.

--- Page 8 ---

The virtual storage budget is set by industrial production economics, not by grid investment <ref:2607.04564#pg2> BESS capacity is a capital decision priced in /kWh <ref:2607.04564#pg2> The shadow price of the curtailment budget is a candidate settlement signal <ref:2607.04564#pg2> VS capacity derives entirely from production commitments that industrial operators already hold, so no additional grid capital investment is required <ref:2607.04564#pg2> This growth complements physical storage investment <ref:2607.04564#pg2>. The shadow price of the curtailment budget is a candidate settlement signal <ref:2607.04564#pg2> λD¯ equals the LMP at the curtailmentonset hour, not the daily peak price <ref:2607.04564#pg2> Because the VS exhausts its budget on every study day, λD¯ reflects the systematic afternoon peak-band structure <ref:2607.04564#pg2> A curtailment contract priced at λD¯ would be self-triggering on any standard summer day <ref:2607.04564#pg2>.

--- Page 9 ---

Disaggregation is infeasible on all 14 days <ref:2607.04564#pg2> The outer approximation Fouter is a strict superset of the true aggregate feasibility set Fagg precisely because the three loads have structurally dissimilar flexibility signatures <ref:2607.04564#pg2> A portfolio in which all loads share the same normalized energy windows would satisfy the proportionality condition, collapse the outer approximation to an exact representation, and eliminate the disaggregation LP entirely <ref:2607.04564#pg2> The case study therefore identifies a portfolio design criterion <ref:2607.04564#pg2> Matching normalized flexibility signatures across constituent loads tightens the aggregate approximation <ref:2607.04564#pg2>. Although the exogenous nodal LMPs reflect marginal congestion conditions from the underlying market solution, the present co-dispatch model does not feed VS and BESS decisions back into network flows <ref:2607.04564#pg2> The reported savings therefore represent price-taking economic value rather than a networkconstrained feasibility result <ref:2607.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Virtual Storage Equivalence enables unified dispatch by establishing that Every feasible largeload trajectory under power-bound and energy-window constraints is a valid charge trajectory of a VS device. This allows for a single linear program to jointly optimize virtual storage and BESS, yielding a co-dispatch price for both resources.

  2. Constraint Reduction significantly improves tractability by replacing the disaggregated load constraints with the aggregate outer representation, reducing the load-flexibility block from O(NT) to O(T). This allows for a single linear program that is O(T) overall, making joint optimization feasible even for large portfolios where N is significant.

  3. Enhanced economic signaling allows for more stable market mechanisms by showing that The curtailment shadow price tracks the peak-price band onset rather than the daily peak price. This provides a candidate settlement signal of lambda D bar, which reflects the systematic after-peak structure and avoids transferring spike-day risk to load operators.

Related papers