Adaptive Inertia and Damping Control

• Adaptive Inertia and Damping Control is a dynamic strategy that treats inertia and damping as time-varying, programmable parameters to balance stability and response speed.
• It leverages converter-interfaced power sources and virtual synchronous generators to emulate mechanical inertia and optimize damping for frequency support and oscillation suppression.
• Methodologies include time-domain modeling, feedback control, and data-driven optimization to adaptively manage resource constraints and disturbance profiles.

Adaptive inertia and damping control denotes a family of control strategies in which the effective inertia and damping of a dynamical system are not treated as fixed coefficients, but as quantities that may vary with time, operating point, disturbance magnitude, disturbance location, or measured state. In power systems, the concept is most closely associated with converter-interfaced power sources, virtual synchronous generators, and virtual synchronous machines, where programmable control replaces the fixed electromechanical inertia of synchronous machines and supplements or reshapes damping for frequency support and oscillation suppression (Wang et al., 2016, Li et al., 2021). Across this literature, a recurring theme is that emulated inertia and damping are not merely substitutes for mechanical parameters: they are control variables whose benefits depend on disturbance timing, network coupling, and resource constraints.

1. Physical basis and control objective

In renewable-rich power systems, converter interfacing decouples the natural kinetic inertia of wind and photovoltaic resources from the grid, reducing aggregate inertia and altering damping and frequency-control behavior (Fernández-Guillamón et al., 2020). The canonical virtual synchronous generator formulation reinterprets the swing equation for inverter control as

$$H\dot{\omega} = P_{\text{ref}} - P_o - k(\omega-\omega_{\text{ref}}),$$

where $H$ is the virtual inertia constant and $k$ is the damping factor. In the same review, the active-power loop is represented as

$$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$

with the reported qualitative trade-off that increasing $H$ slows the response, increases power overshoot, and decreases frequency overshoot, whereas increasing $k$ increases damping and reduces oscillation (Li et al., 2021).

The need for adaptation follows directly from these trade-offs. Fixed inertia and fixed damping cannot simultaneously optimize fast disturbance arrest, nadir containment, oscillation damping, and recovery speed under all operating conditions. The review literature therefore frames adaptive strategies as a means to exploit the programmability of converter-based resources, while also respecting state-of-charge, ramp-rate, and capacity limits of storage and renewable sources (Li et al., 2021).

A further mechanistic point is that inertia and damping cannot be evaluated independently. In a stochastic aggregate model of center-of-inertia frequency, the steady-state variance of frequency deviation is reported as

$$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$

and when $D_L=0$, the variance grows with time. This analysis supports the conclusion that increasing inertia alone does not tightly control long-term frequency fluctuations when load damping is weak; sufficient physical or synthetic damping is also required (Giudice et al., 2020). The same work reports that synthetic inertia can lead to a robust unimodal frequency distribution, but the best performance is observed when synthetic inertia and damping are both present (Giudice et al., 2020).

2. Time-dependent emulation and explicit equivalent parameters

A distinct line of work derives emulated inertia and damping directly from linearized converter-grid models. For converter-interfaced power sources switched into frequency-support mode once a dead-band threshold is exceeded, the active-power reference is written as

$$P_{\text{total,ref}}=P_{\text{ref}}+K_{dr}\Delta\omega+K_{ie}\Delta\dot{\omega},$$

with $K_{dr}$ and $H$0 denoting emulated damping and inertia coefficients, respectively (Wang et al., 2016). After linearization around an operating point, the converter dynamics are expressed as

$H$1

$H$2

while the grid is reduced to an equivalent single machine with swing equation

$H$3

Elimination of converter states yields

$H$4

and hence time-dependent equivalent parameters

$H$5

In this formulation, the emulated inertia and damping are explicit functions of time rather than constants, and their time variation is attributed to convolution terms involving both converter internal dynamics and disturbance profiles (Wang et al., 2016).

The same article further reports explicit tractable expressions under assumed disturbance shapes such as $H$6, making the time-domain contribution of the converter analytically available (Wang et al., 2016). This suggests that “adaptive” behavior can arise even before an additional supervisory optimizer is introduced: once converter dynamics and disturbance history are included, the effective inertia and damping seen by the grid become intrinsically time-varying.

A related but more nonlinear formulation appears in zero-inertia grids operated by grid-forming inverters. There, the virtual synchronous generator equations are modified by frequency- and phase-shift terms driven by an auxiliary state $H$7: $H$8

$H$9

$k$0

$k$1

Here the adaptive action is explicitly linked to proximity to inverter power limits, and the reported linearized Jacobian is negative semi-definite for proper positive choices of control constants (Khamisov et al., 2024).

3. Local adaptive laws in converter-based sources

At device level, adaptive inertia and damping control has been implemented through switching logic, continuous gain scheduling, output-speed feedback, and RoCoF-driven inertia dynamics. The reviewed virtual synchronous generator literature organizes these strategies into several families, including fast-damping stabilization with large inertia during acceleration and small inertia during deceleration, dual-adaptivity for both inertia and damping, process-based multi-phase tuning, and fuzzy or disturbance-adaptive schemes (Li et al., 2021).

Setting	Adaptive quantity	Representative relation
CIPS frequency support	Emulated damping and inertia in active-power reference	$k$2
Dual-adaptive VSG	$k$3 and $k$4	$k$5, $k$6
Output-speed-feedback VSG	$k$7 and damping via $k$8	$k$9
RoCoF-responsive adaptive inertia	$$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$0	$$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$1

In the dual-adaptivity strategy summarized in the review, inertia is switched between $$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$2 and $$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$3 depending on deviation magnitude, while damping is adapted similarly to balance damping versus response speed (Li et al., 2021). A separate process-optimal strategy divides the transient into three phases with parameter pairs $$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$4, $$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$5, and $$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$6 selected to minimize transient frequency response time under bounds on $$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$7 (Li et al., 2021).

An experimentally validated alternative replaces large excursions in inertia and droop damping by introducing output-speed feedback through a coefficient $$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$8. The modified swing equation is

$$G(s)=\frac{\Delta P}{\Delta P_{\text{ref}}}=\frac{1}{2Hs^2+ks+2},$$9

with reported natural frequency and damping ratio

$H$0

The paper states that $H$1 is adapted to maintain an over-damped condition with $H$2, while $H$3 is increased or decreased according to the sign and rate of frequency change (Ren et al., 2020). In the reported prototype experiment with an active-power step from 157W to 600W, the proposed method suppressed power overshoot, kept frequency deviation within $H$4Hz, shortened the adjustment period, and required a peak $H$5 of only $H$6 instead of $H$7 (Ren et al., 2020).

Another local law makes inertia itself a dynamic state: $H$8 This scheme assigns high inertia immediately after a large RoCoF event and lets it decay back to a minimum baseline as the disturbance settles, with the reported property that the new modes introduced by the adaptive inertia law are damped with $H$9 after linearization (Fritzsch et al., 2023).

4. System-level optimization, placement, and coordination

Beyond local controllers, a substantial literature treats inertia and damping as allocatable system resources. One formulation introduces a dynamic optimization problem over virtual inertia $k$0, damping $k$1, and trajectories $k$2, minimizing a weighted combination of cost, energy dissipation, and slack variables subject to system dynamics, RoCoF limits, nadir limits, local and global resource bounds, and feasibility constraints. The reported implementation uses discretized time-domain dynamics via orthogonal collocation and explicitly accounts for disturbance magnitude, disturbance location, and disturbance probabilities through weighted averaging of scenario-dependent optima (Krajacic et al., 9 Jun 2026). On a standard three-area 12-bus system with converter buses 4, 8, and 12, the paper reports that small disturbances favor increased damping, whereas larger local disturbances induce higher inertia allocation, especially at directly affected buses; computation time is stated as approximately 3 seconds per scenario (Krajacic et al., 9 Jun 2026).

A related but different approach formulates optimal virtual synchronous machine tuning as a constrained and regularized $k$3-norm minimization,

$k$4

with Lyapunov equations for the controllability and observability Gramians and a projected gradient descent update

$k$5

Here $k$6 explicitly tunes the trade-off between faster settling and lower nadir/RoCoF, and the paper reports that when disturbance location is known, more inertia and damping are allocated near the vulnerable bus (Ademola-Idowu et al., 2018).

Analytical placement results have also been derived through matrix perturbation theory. Using susceptibilities of an $k$7-type performance measure with respect to heterogeneous inertia and primary-control perturbations, two simple sorting-based algorithms are constructed. The reported conclusion is that optimal inertia distribution is geographically homogeneous, whereas primary control should be mainly located on the slow modes of the network, where intrinsic grid dynamics takes more time to damp frequency disturbances (Pagnier et al., 2019). This differs from the objective of dynamic or RoCoF-responsive schemes, and suggests that static placement and adaptive device behavior need not produce the same spatial pattern.

A further extension couples damping and inertia procurement with market design. For converter-interfaced generation, a convex parametric formulation imposes sufficient small-signal constraints

$k$8

together with RoCoF, steady-state frequency, and empirically linearized nadir constraints. The service cost is minimized over quadratic bids $k$9, and compensation is determined using a Vickrey–Clarke–Groves mechanism. The paper reports scalability to 500- and 2000-bus systems, market clearing in $$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$0 seconds for the 2000-bus case, and a total cost for damping/inertia services of approximately $$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$1 of the total energy market size (Feng et al., 2023).

5. Data-informed adaptation and operating-point dependence

Adaptive damping and inertia control is increasingly tied to measurement-based and data-driven surrogates, particularly when accurate device models are unavailable. In damping-controller coordination, a deep neural network is trained to estimate the Total Action metric

$$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$2

as a function of operating-condition features, disturbance features, and a binary controller-status vector $$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$3. The optimal switching combination is then obtained from

$$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$4

On the Western North America Power System, the reported Total Action values for a short-circuit at bus 157 are 1.241 with no damping controllers, 0.998 for a fixed combination, 0.500 for model-based coordination, and 0.328 for the data-informed coordination; for bus 69, the corresponding values are 0.840, 0.998, 0.473, and 0.295 (Zelaya-Arrazabal et al., 2023). The same study reports real-time determination of the optimal action within approximately 78 ms and retained performance under communication and computation delays up to 2 s (Zelaya-Arrazabal et al., 2023).

For black-box inverters under varying operating points, a self-adaptive active damping framework combines on-line grid-impedance estimation, ANN-based inverter-admittance identification, and ANN-based mapping from operating conditions to active-damper parameters $$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$5 and $$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$6. The reported impedance-estimation method uses only reactive-power-reference perturbation together with a frequency-integral-based dq-axis aligning method; simulation and experimental results state estimation errors below $$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$7 and completion within tens of milliseconds (Li et al., 2024). Stability enhancement is interpreted through the shift of nodal-admittance eigenvalues by the active-damper admittance, with the tuning goal

$$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$8

at the crossing frequency (Li et al., 2024).

High-fidelity EMT-scale studies reinforce the importance of operating-point adaptation. In a modified IEEE 9-bus system implemented on RTDS NovaCor 1.0 with a two-level universal inverter model simulated at 1–3 microseconds time step, dynamic virtual inertia and damping are reported to improve survivability during three-phase faults, generator tripping, step load increases, and islanding. Under some faults, reactive-power surges are reported as 250 MVAr with the adaptive VSG approach versus 800 MVAr for synchronous-generation cases, and the adaptive scheme remains operational during long 300 ms faults and islanding scenarios (Khamisov et al., 2024). The same study states that virtual inertia was set up to 4 times larger and damping up to 3 times larger than typical synchronous-generator counterparts (Khamisov et al., 2024).

6. Broader formulations, neighboring domains, and unresolved issues

Although the most developed application is low-inertia electric power systems, adaptive inertia and damping control also appears in broader control-theoretic settings. In adaptive civil structures, a port-Hamiltonian model

$$\mathrm{Var}[\Delta\omega]\simeq \sigma^2=\frac{b^2}{2\alpha D_L(D_L+2H\alpha)},$$9

is paired with the cost

$D_L=0$0

so that the control action is interpreted through the physical energy balance

$D_L=0$1

For a 152-degree-of-freedom reduced high-rise model with 28 hydraulic cylinders, the reported withdrawn energy is 9019 J for the port-Hamiltonian optimal controller, compared with 8221 J for LQR with $D_L=0$2, and the analysis establishes a turnpike property and an explicit singular-arc characterization (Schaller et al., 2023). This is not a power-system inertia formulation, but it places damping control within the same energy-based design vocabulary.

In uncertain port-controlled Hamiltonian systems, adaptive interconnection-and-damping assignment passivity-based control is combined with an immersion-and-invariance estimator. The estimator update law uses a free design function $D_L=0$3, and the reported stability theory establishes Lyapunov asymptotic stability of the estimator dynamics and local asymptotic stability of the closed-loop system in a sufficiently large set, provided suitable monotonicity and Lipschitz conditions hold (Alkrunz et al., 2024). In inertial optimization, adaptive damping is introduced through

$D_L=0$4

with Lyapunov analysis showing global asymptotic stability of the continuous-time system and numerical results on Rosenbrock’s function indicating improved performance over fixed-schedule momentum methods (Bhattacharjee et al., 2021).

Complex-network synchronization offers a further explicit adaptive-inertia law. For a Kuramoto model with inertia, Laplacian eigenvector decomposition yields modal dynamics

$D_L=0$5

and variational optimization produces

$D_L=0$6

Across regular, random, small-world, scale-free, and spider-web networks, the reported reductions in the fragility performance function $D_L=0$7 are $D_L=0$8–$D_L=0$9, relaxation-time reductions are $$P_{\text{total,ref}}=P_{\text{ref}}+K_{dr}\Delta\omega+K_{ie}\Delta\dot{\omega},$$0–$$P_{\text{total,ref}}=P_{\text{ref}}+K_{dr}\Delta\omega+K_{ie}\Delta\dot{\omega},$$1, and the real parts of all system eigenvalues are less than $$P_{\text{total,ref}}=P_{\text{ref}}+K_{dr}\Delta\omega+K_{ie}\Delta\dot{\omega},$$2 (Zhou et al., 20 Jan 2026).

Across these domains, several unresolved directions are stated explicitly in the virtual synchronous generator review: simultaneous adaptive multi-parameter control rather than single-parameter tuning, optimal inertia allocation in hybrid sources and configurations, and cooperative or distributed strategies for multiple VSGs operating with coordinated inertia and damping (Li et al., 2021). Taken together with the stochastic-frequency analysis showing that inertia restoration alone is insufficient when damping is weak (Giudice et al., 2020), these works indicate that the central research problem is not the maximization of inertia per se. It is the coordinated shaping of transient arrest, oscillation damping, recovery speed, and resource usage under time-varying physical and network constraints.