Small-Signal Security Region (SSSR)

• Small-Signal Security Region (SSSR) is defined as the set of operating points or injection vectors where the reduced Jacobian’s eigenvalues lie in the left-half-plane, ensuring asymptotic stability.
• The concept extends to various parameter spaces, including conventional synchronous generators and wind power injections, using projections, robust SDP certificates, and data-driven surrogates.
• Applications of SSSR include secure stability assessment in differential–algebraic models, cyber-physical security frameworks, and inverter-dominated grid operations with reliability margins.

Small-Signal Security Region (SSSR) denotes the set of operating points, parameters, or injection vectors for which a power-system equilibrium exists and the associated small-signal model is asymptotically stable. In the cited literature, the concept is defined on parameter space or power injection space rather than only at a single operating point, and it is characterized by the left-half-plane location of the eigenvalues of the reduced Jacobian together with regularity of the algebraic subsystem. In this sense, “security” is the operational interpretation of the underlying stability region rather than a sharply distinct mathematical object (Pan et al., 2015). Subsequent work broadens the concept toward robust inner approximations in DAE state space, projection-based admissible regions under uncertain wind injections, empirical boundary-sampling procedures, and data-driven or online surrogates for secure-region membership (Pareek et al., 2018).

1. Classical formulation in differential-algebraic models

A standard starting point is the parameterized power-system DAE

$$\begin{cases} \dot{x} = F(x,y,p) \ 0 = G(x,y,p), \end{cases}$$

where $x\in\mathbb{R}^n$ are state variables, $y\in\mathbb{R}^m$ are algebraic variables, and $p\in\mathbb{R}^l$ is a parameter vector (Pan et al., 2015). Linearization at an equilibrium gives

$$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$

and, when $D$ is nonsingular, reduction to state space yields

$$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$

The eigenvalues $\lambda_i$ of $\tilde A$ provide the classical small-signal criterion: the equilibrium is small-signal stable when $\Re(\lambda_i) < 0$ for all modes (Pan et al., 2015).

On this basis, the SSSR is defined as

$x\in\mathbb{R}^n$0

Its boundary is

$x\in\mathbb{R}^n$1

The cited formulation associates $x\in\mathbb{R}^n$2 with Hopf bifurcation or saddle-node bifurcation, and $x\in\mathbb{R}^n$3 singularity with singularity induced bifurcation (Pan et al., 2015).

This definition makes SSSR a region-wise generalization of pointwise eigen-analysis. Instead of asking whether one operating point is stable, it asks which parameter vectors or injection patterns produce equilibria that remain small-signal stable. In the conventional generation-injection setting, the parameter vector may be taken as

$x\in\mathbb{R}^n$4

with active-power balance

$x\in\mathbb{R}^n$5

so that the SSSR becomes a region in synchronous-generator injection space (Pan et al., 2015).

2. Geometric interpretations and region variants

The classical SSSR is a subset of parameter or injection space, but the literature develops several related geometric views. In the wind-integration setting, the SSSR is extended from conventional synchronous-generator injections $x\in\mathbb{R}^n$6 to an augmented injection vector

$x\in\mathbb{R}^n$7

which includes both synchronous-generation injections and wind power injections (WPIs). Under AGC participation factors $x\in\mathbb{R}^n$8, the paper defines an extended $x\in\mathbb{R}^n$9-SSSR in this higher-dimensional injection space and then projects it onto WPI space to obtain the admissible region of wind generation considering small-signal stability (SSAR): $y\in\mathbb{R}^m$0 Geometrically, the extended SSSR is the stable set in joint SG+wind injection coordinates, while SSAR is its projection onto wind-injection coordinates (Pan et al., 2015).

This projection view is significant because uncertain WPIs do not merely perturb one equilibrium; they relocate equilibria, alter $y\in\mathbb{R}^m$1, and therefore change $y\in\mathbb{R}^m$2 and its eigenvalues. The cited wind paper treats this effect primarily through equilibrium relocation while neglecting WTG dynamics in the main development, assumes algebraic balancing through AGC participation factors, and uses the projected region to quantify how much uncertain wind generation can be accommodated without breaking small-signal stability (Pan et al., 2015). It also assumes exciter parameters are in proper ranges so that no holes exist in the relevant extended SSSR.

A distinct geometric variant appears in DAE-based robust certification. There, the constructed object is not the exact SSSR in the classical necessary-and-sufficient sense, but a robustly stable inner subset in the state-algebraic variable space $y\in\mathbb{R}^m$3. The intended interpretation is a region of feasible equilibria or operating points embedded in the DAE state-algebraic space, not an arbitrary transient-state invariant set. This shifts the region concept from parameter space toward state space while remaining tied to equilibrium feasibility and small-signal stability (Pareek et al., 2018).

3. Exact spectral tests and conservative robust certificates

For semi-explicit DAEs

$y\in\mathbb{R}^m$4

the reduced Jacobian is

$y\in\mathbb{R}^m$5

and the exact benchmark remains

$y\in\mathbb{R}^m$6

The DAE-based robust framework constructs a bridge between the reduced model and the original DAE Jacobian

$y\in\mathbb{R}^m$7

through generalized reduced and unreduced Jacobians

$y\in\mathbb{R}^m$8

with $y\in\mathbb{R}^m$9, $p\in\mathbb{R}^l$0, and $p\in\mathbb{R}^l$1 (Pareek et al., 2018).

The key surrogate is the logarithmic norm

$p\in\mathbb{R}^l$2

with

$p\in\mathbb{R}^l$3

The central lemma gives

$p\in\mathbb{R}^l$4

and the resulting sufficient condition is

$p\in\mathbb{R}^l$5

Thus, negativity of the matrix measure of a generalized unreduced Jacobian certifies small-signal stability of the DAE system, even though it is only a sufficient condition and not equivalent to the exact eigenvalue test (Pareek et al., 2018).

For $p\in\mathbb{R}^l$6, the certificate is written as a BMI

$p\in\mathbb{R}^l$7

with $p\in\mathbb{R}^l$8 and $p\in\mathbb{R}^l$9. Because this BMI is bilinear in $$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$0 and $$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$1, hence nonconvex and $$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$2-hard, the tractable step is to fix a certificate matrix $$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$3 at a base equilibrium $$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$4, solve

$$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$5

and then certify a box uncertainty set

$$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$6

through the robust SDP

$$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$7

If the optimal $$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$8, all feasible equilibria in $$\begin{bmatrix} \Delta \dot{x} \ 0 \end{bmatrix} = \begin{bmatrix} A & B \ C & D \end{bmatrix} \begin{bmatrix} \Delta x \ \Delta y \end{bmatrix},$$9 are certified stable. The cited work explicitly characterizes this as an inner approximation of the BMI-defined region and therefore also an inner approximation of the true eigenvalue-based SSSR (Pareek et al., 2018).

In the reported 2-bus study, the BMI-certified stable area was always contained in the true stable area and was close to it in a two-dimensional state plane; the robust SDP/LMI also produced convex stable regions in transformed variable planes. This suggests a conservative but optimization-friendly approximation to SSSR-like sets under operating-point variability, especially when repeated reduced-Jacobian eigenvalue checks are undesirable.

4. Boundary construction, sampling, and security-margin formulations

Recent boundary-oriented work operationalizes the secure region as the intersection of AC feasibility and small-signal security margin constraints. In that setting, an operating point is feasible if it satisfies steady-state operational constraints, stable if it is small-signal stable, and secure if it is both feasible and small-signal stable. The controllable operating-point vector is

$D$0

for non-slack generator active powers, generator-bus voltage magnitudes, and load-bus complex powers. The practical security boundary is then defined jointly by the AC-feasible region boundary and a minimum damping-ratio boundary, with case studies using

$D$1

for the least damped mode (Giraud et al., 16 Jan 2025).

The same work defines a high-information-content region

$D$2

where $D$3 is the security-boundary value of the chosen margin, $D$4 is the security index at $D$5, and $D$6 is a tolerance band. The computational pipeline combines optimization-based bound tightening, separating hyperplanes derived from a QC relaxation of AC-OPF, Hit-and-Run sampling in the remaining convex polytope, directed walks toward the boundary using damping-ratio sensitivity, and final AC-feasibility repair. The directed-walk step uses

$D$7

with adaptive step size $D$8 chosen according to distance-to-boundary tiers (Giraud et al., 16 Jan 2025).

This methodology does not derive an analytic SSSR manifold; it samples near a composite boundary empirically. In the reported PGLib-OPF 39-bus and 162-bus studies, the proposed method yielded $D$9 and $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$0 of samples in the HIC region, respectively. Decision-tree classifiers trained on these boundary-enriched datasets achieved better F1-scores on explicit boundary test sets than naïve Latin hypercube sampling or a Gaussian importance-sampling benchmark: $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$1 versus $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$2 and $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$3 on the 39-bus boundary set, and $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$4 versus $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$5 and $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$6 on the 162-bus boundary set (Giraud et al., 16 Jan 2025). This indicates that, in practical security assessment, preserving boundary geometry and balancing samples on both sides of the secure/insecure divide can materially improve learned region discrimination.

5. Data-driven and online approximations of secure-region membership

One line of work replaces explicit geometric construction by learned membership testing. A graph neural network formulation for real-time small-signal security assessment defines binary security under an $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$7 contingency criterion: output $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$8 if the operating point satisfies the small-signal security criterion for all contingencies considered, output $$\Delta \dot{x} = \tilde A\,\Delta x,\qquad \tilde A = A - B D^{-1} C.$$9 otherwise. Labels are generated by varying generator active powers and load active/reactive powers, solving OPF, simulating all single-line contingencies, applying a 3-phase fault at $\lambda_i$0 and clearing it at $\lambda_i$1, then checking whether the damping ratio

$\lambda_i$2

satisfies $\lambda_i$3 for every contingency. The learned classifier therefore can be interpreted as an implicit approximation of the indicator function of an SSSR-like secure set, although it does not explicitly construct a geometric region (Justin et al., 2024).

In that GNN approach, each bus carries node features

$\lambda_i$4

and graph filtering is defined by

$\lambda_i$5

The IEEE 68-bus and NPCC 140-bus case studies reported best accuracies of $\lambda_i$6 and $\lambda_i$7 for the proposed GNNs, versus $\lambda_i$8 and $\lambda_i$9 for CNN baselines; training times were $\tilde A$0 min and $\tilde A$1 min for the GNNs, versus $\tilde A$2 min and $\tilde A$3 min for the CNNs. Under $\tilde A$4, $\tilde A$5, and $\tilde A$6 missing data at unhighlighted buses, the reported GNN accuracy remained unchanged up to $\tilde A$7 missing data, whereas CNN performance degraded sharply (Justin et al., 2024). For SSSR use, this is best interpreted as fast secure/insecure membership testing rather than direct boundary computation.

A different data-driven direction estimates a confidence-qualified secure neighborhood around a stable equilibrium in state space. Starting from the reduced DAE Jacobian

$\tilde A$8

the cited method assumes $\tilde A$9 is Hurwitz at the current operating point, constructs a converse Lyapunov function

$\Re(\lambda_i) < 0$0

learns it from trajectory data with a windowed online Gaussian process, and defines the estimated secure region

$\Re(\lambda_i) < 0$1

This object is not a classical SSSR; it is a probabilistic inner approximation of a region of attraction or dynamic secure set in state space. Still, it is closely related operationally because it tracks a changing secure neighborhood around a small-signal-stable equilibrium. The reported microgrid study used a $\Re(\lambda_i) < 0$2-bus, $\Re(\lambda_i) < 0$3-generator system, window width $\Re(\lambda_i) < 0$4, an RBF kernel, a $\Re(\lambda_i) < 0$5 confidence region, and computation time around $\Re(\lambda_i) < 0$6 seconds in Matlab 2017b (Zhai, 2019).

6. Cybersecurity and inverter-dominated extensions

Cyber-physical work uses the same small-signal boundary concept in reverse: instead of constructing the stable set, it asks whether constrained perturbations can force an operating point to leave it. In a false-data-injection framework, the operating point is altered through attacked load measurements, AC power flow is re-solved, and the resulting linearized model is checked for instability via the criterion

$\Re(\lambda_i) < 0$7

The attack synthesis problem couples AC network equations, operating constraints, and small-signal eigenvalue constraints. Although no full SSSR is explicitly computed, the method studies attack-induced crossing of a small-signal stability boundary. On the WSCC 3-machine 9-bus system, $\Re(\lambda_i) < 0$8 uniformly random attack vectors produced only one destabilizing case, for an empirical success rate of about $\Re(\lambda_i) < 0$9, whereas an informed attacker could still drive the system unstable even with limited measurement access (Jafari et al., 2021). In SSSR terms, the work is about reachability of the unstable complement of the stable region under attack constraints.

In inverter-dominated systems, the SSSR has been reformulated explicitly for switched grid-following/grid-forming control. For a parameter vector $x\in\mathbb{R}^n$00 and eigenvalues $x\in\mathbb{R}^n$01 of the small-signal state matrix $x\in\mathbb{R}^n$02, the region is defined as

$x\in\mathbb{R}^n$03

with numerical boundary approximation

$x\in\mathbb{R}^n$04

The cited paper develops full-order $x\in\mathbb{R}^n$05th-order GFL and $x\in\mathbb{R}^n$06th-order GFM subsystem models, fits subsystem SSSR boundaries by a hyperplane-approximation algorithm with $x\in\mathbb{R}^n$07 and $x\in\mathbb{R}^n$08, and states that the overall switched-system SSSR is the union of the constituent subsystem SSSRs (Lai et al., 20 Mar 2026).

That inverter work also introduces internal stability margin distributions (ISMDs), where the margin $x\in\mathbb{R}^n$09 is the orthogonal distance from the rightmost eigenvalue to the imaginary axis, and a comprehensive stability index

$x\in\mathbb{R}^n$10

with $x\in\mathbb{R}^n$11. The margin term is derived from a GMM regression, the sensitivity term from the Euclidean norm of $x\in\mathbb{R}^n$12, and the boundary-distance term from the minimum Euclidean distance to the SSSR boundary. The reported parameter studies found that GFL is more stable under larger SCR, GFM is more secure under smaller SCR, GFL is more sensitive to $x\in\mathbb{R}^n$13, and GFM has comparable sensitivity to both $x\in\mathbb{R}^n$14 and $x\in\mathbb{R}^n$15. EMT simulations further showed that a CSI-based adaptive switching policy can outperform a conventional strategy that switches mainly on SCR, because the proposed policy avoids operation near fragile portions of the SSSR (Lai et al., 20 Mar 2026).

Across these developments, the SSSR remains the central abstraction for organizing small-signal security: as an exact eigenvalue-defined region in parameter or injection space, as a projected admissible set under renewable uncertainty, as a conservative DAE-based robust inner approximation in state-algebraic space, as a composite AC-feasibility-plus-margin boundary for dataset generation, as an implicitly learned secure-set indicator in real-time assessment, and as a switching-design object in inverter-dominated grids. The cited literature collectively suggests that modern use of SSSR increasingly combines exact spectral definitions with conservative certificates, projected or composite boundaries, and data-driven surrogates when direct region construction becomes difficult.