Frequency Nadir in Power Systems

Updated 18 April 2026

Frequency nadir is defined as the maximum absolute frequency deviation at a bus following a disturbance, serving as a key indicator for transient security and triggering protection measures.
Analytical models using swing equations, modal decomposition, and data-driven surrogates accurately predict nadir behavior even in complex, multi-machine grid systems.
Mitigation strategies, including grid-forming controls and dynamic droop methods, actively counteract nadir effects to enhance system stability and operational reliability.

Frequency nadir is a central concept in the stability and security assessment of electrical power systems and is also used as a descriptive term for a lower-bound threshold in other domains (such as radio astronomy). In power systems, frequency nadir is rigorously defined as the largest absolute frequency deviation at a specified location (bus) following a disturbance, and is a key criterion for transient security, grid protection, and operational reserve requirements. This article comprehensively surveys the definition, modeling, prediction techniques, mitigation strategies, computational challenges, and applications of frequency nadir, with technical details anchored in recent arXiv literature.

1. Formal Definition and System-Theoretic Context

Frequency nadir in power systems is defined as the most extreme magnitude of frequency deviation (dip or peak) at a bus following a system disturbance. For bus $i$ , the frequency nadir is

$\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$

where $\omega_i(t)$ denotes the frequency deviation (in p.u. or Hz) at bus $i$ , and $t=0$ is the disturbance onset (Jiang et al., 2024). If nadir exceeds certain operational thresholds (e.g., $\pm$ 0.01–0.016 p.u.), it may trigger underfrequency relays or protection schemes, leading to undesirable events including generator tripping, load shedding, or cascading failure (Jiang et al., 2024, You, 2020). The worst-case bus nadir,

$\max_{i} \max_{t} |\omega_i(t)|$

is generally more conservative and operationally relevant than the inertia-weighted center-of-inertia (COI) nadir, especially in networks with asynchronous resources or where modal coherency is weak.

The concept is also used in other contexts, such as the lowest frequency at which a phenomenon is detected, e.g., the "frequency nadir" for fast radio bursts is the lowest frequency at which emission is observed (Pilia et al., 2020).

2. Mathematical Models and Nadir Calculation Frameworks

Frequency nadir is inherently a transient metric that arises from the system’s dynamic response to a disturbance. The foundational models are based on the swing equation and extensions:

Classical Power System Model

$2H \cdot \frac{d\Delta f(t)}{dt} + D \cdot \Delta f(t) = P_{\text{imbalance}}(t)$

where $H$ is the system inertia constant, $D$ the load frequency sensitivity (damping), and $\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 0 the net (disturbance minus reserves) (Rajabdorri et al., 2023, Liu et al., 2021). Extensions incorporate generator governor/turbine response (typically first-order lag):

$\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 1

for aggregate droop constant $\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 2 and time constant $\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 3 (Liu et al., 2021).

For multi-machine or networked settings, the full LTI model is

$\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 4

with $\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 5 stacking angles and frequencies, $\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 6 given by the aggregation of swing and network susceptance, and $\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 7 the disturbance profile (Jiang et al., 2024).

Historically, nadir prediction used average system frequency (ASF) models that treat all machines as a single equivalent block. However, such reductions are often inaccurate in the presence of pronounced oscillatory modes, non-uniform inertia, or network-induced modal behavior (Zelaya-Arrazabal et al., 4 Feb 2025). Modal decomposition offers a refined approach:

Linearize the system DAEs and perform Kron reduction to obtain $\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 8, the system matrix.
Eigen-decompose $\text{nadir}_i := \max_{t \ge 0} |\omega_i(t)|$ 9 to obtain the slow modes (eigenvalues/eigenvectors with largest time constants).
COI frequency trajectory is then written as a sum over (typically 2–4) dominant modes:

$\omega_i(t)$ 0

where the nadir is computed by maximizing $\omega_i(t)$ 1 over $\omega_i(t)$ 2 using either closed-form expressions or a brief root-finding step (Zelaya-Arrazabal et al., 4 Feb 2025).

For single-area or reduced-order models, classical second-order analysis yields the time to nadir $\omega_i(t)$ 3 and nadir value:

$\omega_i(t)$ 4

with $\omega_i(t)$ 5 denoting the undamped natural frequency and damping ratio (Liu et al., 2021).

Step, proportional, or derivative (synthetic inertia) fast frequency responses—particularly from inverter-based resources—are included explicitly in recent models (Dong et al., 2022).

4. Computational Techniques and Efficient Algorithms

Nadir assessment is computationally demanding in large grids with many disturbance scenarios. Recent research contributes several advancements:

Oscillation-aware algorithms: Efficient eigendecomposition and modal superposition techniques—using the system Laplacian and weighted modal basis—yield a fast, closed-form computation of the worst-case nadir for arbitrary disturbance budgets. The inner maximization over disturbances admits analytic dual-norm solutions, reducing problem complexity from $\omega_i(t)$ 6 (for $\omega_i(t)$ 7 time-domain simulations) to $\omega_i(t)$ 8 (for $\omega_i(t)$ 9 buses and $i$ 0 time discretizations) (Jiang et al., 2024).
Data-driven and learning-based surrogates: High-fidelity data is used to fit linear or piecewise-linear constraints—via regression, support vector machines, or extreme learning machines—which approximate the mapping from scheduling variables to nadir. These constraints are tractable within mixed-integer (MILP) frameworks for unit commitment and scheduling (Rajabdorri et al., 2023, Rajabdorri et al., 2022, Liu et al., 2021). ELM and SVM-based nadir surrogates achieve sub-1% error versus full dynamic simulations and accelerate UC solution times by two orders of magnitude (Rajabdorri et al., 2022, Liu et al., 2021).

5. Mitigation, Control, and Nadir Elimination Strategies

Nadir mitigation is a primary objective in system design. Several advanced methods exist:

Grid-forming frequency shaping: Inverter controls are designed to force the aggregate coherent mode dynamics to be exactly first-order,

$i$ 1

such that the frequency response is monotonic and nadir is eliminated ( $i$ 2). Parameter choices allow tight tuning of RoCoF and steady-state error (Jiang et al., 2020, Jiang et al., 2020).

Dynamic droop (iDroop) and frequency shaping via storage: ESS implement a proportional-plus-derivative feedback law ( $i$ 3). The closed-loop becomes strictly first-order, fully eliminating nadir and allowing algebraic RoCoF and nadir tuning. ESS sizing (peak power, energy) requirements decrease by up to 40% compared to virtual inertia approaches (Jiang et al., 2019, Jiang et al., 2020).
Learning-augmented regulation: Monotone neural network mappings shape the input-output behavior of secondary frequency controllers, with loss functions directly penalizing nadir excursion. Data-driven optimization improves response speed and reduces nadir while maintaining provable stability (Yu et al., 10 Mar 2026).
Optimal wind turbine active power control: Trajectory optimization (via Gauss-pseudospectral transcription) finds the time-profile of wind power injection that maximizes nadir across all credible events. The reverse-engineered law then implements this “ideal” frequency trajectory in real systems, outperforming all known emulations and providing universal security (Zhang et al., 30 Mar 2026).

6. Integration with Power System Operations and Scheduling

Nadir constraints are now routinely integrated into operational scheduling, particularly unit commitment (UC):

Continuous-time UC with nadir constraints: Bernstein polynomials represent hour-to-hour generator trajectories. Exact (but nonlinear) nadir constraints are replaced with high-fidelity linear surrogates trained on simulation data, resulting in reliable and efficient enforcement of nadir limits in MILP-based UC (Rajabdorri et al., 2023).
Machine learning surrogates: Logistic regression and SVM classifiers predict the feasibility of a candidate dispatch with respect to enforced nadir thresholds. These models—trained on simulated or measured outcomes—enable real-time, certified security within large-scale UC problems with minimal MILP overhead (Rajabdorri et al., 2022, Liu et al., 2021).
Frequency response characteristic (FRC) methodology: Real-time tracking of the FRC curve provides an immediate graphical tool for estimating the nadir following various disturbances, and for operationally validating security across fast-changing system conditions (You, 2020).

7. Nadir Beyond Power Systems: Signal Detection Thresholds

The term “frequency nadir” is sometimes used in other scientific fields to denote a lower detection bound or minimum frequency of a phenomenon. Notably, in fast radio burst (FRB) astronomy, the frequency nadir refers to the lowest frequency at which FRB emission is found. For example, SRT detected periodic FRB 180916 at 328 MHz, establishing a new frequency nadir for such events and providing constraints on propagation effects and source models (Pilia et al., 2020).

References

(Jiang et al., 2024) Oscillations-Aware Frequency Security Assessment via Efficient Worst-Case Frequency Nadir Computation
(Zelaya-Arrazabal et al., 4 Feb 2025) A Modal-Based Approach for System Frequency Response and Frequency Nadir Prediction
(Balla-Elliott, 2023) Data-Driven Continuous-Time Framework for Frequency-Constrained Unit Commitment
(You, 2020) Frequency Response Characteristic (FRC) Curve and Fast Frequency Response Assessment in High Renewable Power Systems
(Yu et al., 10 Mar 2026) Learning-Augmented Primal-Dual Control Design for Secondary Frequency Regulation
(Jiang et al., 2019) Dynamic Droop Approach for Storage-based Frequency Control
(Jiang et al., 2020) Storage-Based Frequency Shaping Control
(Rajabdorri et al., 2022) Inclusion of Frequency Nadir constraint in the Unit Commitment Problem of Small Power Systems Using Machine Learning
(Liu et al., 2021) An Extreme Learning Machine-Based System Frequency Nadir Constraint Linearization Method
(Pilia et al., 2020) The lowest frequency Fast Radio Bursts: Sardinia Radio Telescope detection of the periodic FRB 180916 at 328 MHz
(Dong et al., 2022) A Unified Analytical Method to Quantify Three Types of Fast Frequency Response from Inverter-based Resources
(Zhang et al., 30 Mar 2026) A System-View Optimal Additional Active Power Control of Wind Turbines for Grid Frequency Support