Aggregate Flexibility Envelope

• Aggregate Flexibility Envelope is a mathematical framework that characterizes all feasible aggregate power or energy trajectories from distributed flexible devices under device-level and system constraints.
• It leverages convex analysis, polyhedral geometry, and robust optimization to construct vertex-based inner approximations that enable tractable, scalable, and implementable scheduling.
• The framework supports operational decisions such as cost and peak power minimization by ensuring any aggregate trajectory can be efficiently disaggregated into feasible device-level instructions.

An aggregate flexibility envelope is a mathematical construct that compactly characterizes the set of all feasible aggregate power or energy trajectories that can be delivered by an ensemble of flexible devices—such as batteries, thermostatically controlled loads (TCLs), or electric vehicles (EVs)—over a given time horizon, subject to device-level and system constraints. These envelopes are essential for system operators, aggregators, and markets to internalize and schedule distributed flexibility, while enabling guarantees that any schedule within the envelope can be disaggregated to feasible device-level instructions. The construction and utilization of flexibility envelopes are governed by convex analysis, polyhedral geometry, and robust optimization, and advanced recent work focuses on scalable, accurate representations and tractable optimization and disaggregation.

1. Exact Formulation of the Aggregate Flexibility Envelope

Let $N$ denote the number of flexible devices (e.g., batteries), each indexed by $i$. Over a discrete time horizon of $d$ periods, the feasible trajectories of device $i$ are described by a polytope $F_i = \{ x \in \mathbb{R}^d \mid A_i x \le b_i \}$, where $x_i$ denotes the power profile and $(A_i, b_i)$ encode operational constraints such as power, energy, or state-of-charge limits.

The aggregate flexibility envelope $F_{\rm agg}$ is defined as the Minkowski sum of these polytopes: $$F_{\rm agg} = \sum_{i=1}^N F_i = \left\{ x = \sum_{i=1}^N x_i \mid x_i \in F_i \right\}$$ This set describes all aggregate power trajectories achievable by combining any feasible individual schedules. However, for general $N$ and $d$, $F_{\rm agg}$ is a high-dimensional polytope whose exact facet or vertex enumeration is computationally intractable (“curse of dimensionality”). This motivates the computation of tractable inner- or outer-approximations $\mathcal{A} \subseteq F_{\rm agg}$ with explicit disaggregation guarantees (Öztürk et al., 2023).

2. Vertex-Based Inner Approximation and the Envelope Construction

Öztürk et al. (Öztürk et al., 2023) propose an objective-agnostic inner-approximation based on extremal-vertex enumeration:

• For each device $i$ and each sign pattern $j = (j_1,\ldots,j_d) \in \{-1,+1\}^d$, construct a vertex candidate $y_i^j \in F_i$ by greedily maximizing along the coordinate directions:
  • $$y_{i,1}^j = j_1 \max \{ j_1 x \mid (x,0,\ldots,0)\in F_i \}$$

  • For $t=2,\ldots,d$,

    $$y_{i,t}^j = j_t \max \left\{ j_t x \mid (y_{i,1}^j,\ldots,y_{i,t-1}^j,x,0,\ldots,0) \in F_i \right\}$$

• The vector of all $y_i^j$ across $i$ is summed to form an aggregate vertex: $v^j = \sum_{i=1}^N y_i^j$.
• Proposition III.4: Each $v^j$ is a vertex of $F_{\rm agg}$. The inner-approximation is defined as the convex hull:

  $$\mathcal{A} = \mathrm{Conv}\left\{ v^j: j \in \{-1,1\}^d \right\} \subseteq F_{\rm agg}$$

• In practical computation, one may sample a subset $g\le 2^d$ of sign patterns to trade off between accuracy and scalability.

The polytope $\mathcal{A}$ resides explicitly as a convex combination of the $v^j$, and its description scales linearly in $N$ and at most exponentially (in the worst case) in $d$. For most applications, far fewer than $2^d$ patterns suffice to recover the essential shape of the envelope.

3. Optimization over the Envelope and Accuracy Metrics

The constructed envelope $\mathcal{A}$ is intended to be used as an admissible proxy for the exact aggregate flexibility set in system-level optimization:

• Cost minimization: $\min_{x\in \mathcal{A}} c^\top x$, with $c_t$ as (e.g.) the price vector.
• Peak power minimization: $$\min_{x\in \mathcal{A}} \| x \|_\infty = \min \{ \max_t |x_t| \mid x\in \mathcal{A} \}$$.

The approximation quality is assessed via the Unused-Potential-Ratio (UPR), defined as: $$\mathrm{UPR} = 100 \times \frac{z_{\rm approx} - z_{\rm exact}}{z_{\rm noflex} - z_{\rm exact}}$$ where $z_{\rm approx}$ is the achieved objective with $\mathcal{A}$, $z_{\rm exact}$ is the true optimum with $F_{\rm agg}$, and $z_{\rm noflex}$ is the unmanaged baseline (e.g., zero flexibility). UPR values closer to zero indicate a tight approximation (Öztürk et al., 2023).

4. Disaggregation and Implementability

A critical feature of the vertex-based envelope is the efficient disaggregation property:

• Given any aggregate profile $\bar{x} \in \mathcal{A}$ expressed as $\bar{x} = \sum_j \alpha_j v^j$ (with $\sum_j \alpha_j = 1, \alpha_j \ge 0$), the individual device trajectories are recovered as:

  $$\bar{x}_i = \sum_{j\in\{-1,1\}^d} \alpha_j y_i^j, \quad i=1,\ldots,N$$

• Each $\bar{x}_i \in F_i$, and $\sum_i \bar{x}_i = \bar{x}$.
• This avoids the need for additional LP disaggregation and guarantees implementability for any aggregate trajectory within the envelope (Öztürk et al., 2023).

5. Computational Complexity and Benchmarking

The overall computational effort is governed by:

• Enumeration of $g$ sign-patterns: $O(N\,g\,d)$ time.
• Convex hull operations: efficient as the set is supplied in vertex form, and convex optimization over $\mathcal{A}$ can proceed by barycentric coordinates.
• In empirical studies, the vertex envelope approach outperformed ten state-of-the-art inner approximations—such as zonotope, ellipsoid, and homothet-based methods—in both computational complexity and accuracy (e.g., achieving UPRs below 8% for cost and peak minimization, vs. 30–100% for other techniques), and scaled to $N=500,\ d=96$ in under 6 minutes (Öztürk et al., 2023).

Problem size	Vertex method time	UPR (cost)	Competing methods time	UPR (cost)
$N=30,\ d=24$	$\sim$0.4 s	$<$8%	10–100s, or fail ($>$10 min)	30–100%
$N=500,d=96$	$<$6 min	7–34%	—	—

6. Trade-offs, Scalability, and Broader Relevance

• The number of extreme points $g$ can be tuned to balance accuracy against runtime, and $g=d^2$ is empirically sufficient for high fidelity in most cases.
• The method generalizes straightforwardly to other classes of polytopic or convex flexibility sets, allowing for broader applicability including aggregations of TCLs, EVs, and mixed DERs.
• By representing the dominant vertices relevant for operational objectives, the aggregate envelope is immediately amenable to day-ahead scheduling, market participation, and real-time dispatch, while offering provable implementability via the disaggregation mapping (Öztürk et al., 2023).

7. Context Within the Literature and Extensions

The vertex-based envelope framework addresses classical computational barriers in Minkowski sum calculation by combining algorithmic tractability, inner-approximation guarantees, and efficient disaggregation:

• Contrasted with zonotopic, ellipsoidal, and homothet-based approaches, which may suffer from conservativeness or scalability issues, as detailed in benchmarking (Öztürk et al., 2023).
• The method unifies several strands of aggregate flexibility approximation, and connects with parallel advances in generalized polymatroid approaches, hyper-rectangular inner/outer bounds, and projection-based relaxations in the broader literature.
• Practical extensions include sampling-based refinement, hybridization with convex-projection and scenario methods for uncertainty handling, and incorporation into market interfaces and network-constrained aggregations.

Conclusion:

The aggregate flexibility envelope is a mathematically grounded, scalable, and operationally robust construct for quantifying and deploying the joint flexibility of distributed devices. The vertex-based convex hull approach introduced by Öztürk et al. establishes new best-in-class trade-offs between computational cost and accuracy, while guaranteeing that any profile in the envelope can be feasibly implemented at device level—making it an essential tool for modern aggregator, system operator, and market architectures (Öztürk et al., 2023).