betaFeedback

WorksEngineering

Enhancing Transient Stability of VSG-Controlled Converters Under Grid Frequency Disturbances

Abstract

Strong damping feedback can substantially improve the transient synchronization stability of renewable energy converters employing virtual synchronous generator (VSG) control. However, conventional damping feedback uses the nominal frequency as its reference. When the frequency of a weak grid deviates from nominal, strong damping tends to keep the VSG frequency near nominal. This can sustain a frequency difference between the VSG and the grid and compromise transient synchronization. To address this problem, this paper proposes a complex-frequency-based method to estimate the grid reference frequency from local terminal measurements. The estimated frequency replaces the nominal reference in the VSG synchronization loop, so that damping acts on the frequency difference between the VSG and the real grid. Controller hardware-in-the-loop tests validate the effectiveness of the proposed method under grid frequency disturbances and its robustness to grid parameter uncertainty and adverse grid conditions.

Introduction

The large-scale integration of renewable generation through grid-following (GFL) converters has reduced system inertia and short-circuit capacity, making modern power systems more vulnerable to wideband oscillations and voltage or frequency violations [1]. Against this backdrop, grid-forming (GFM) control has attracted increasing attention for renewable energy converters. GFM converters establish an autonomous voltage source at the grid connection point with self-regulated voltage and frequency [2]. Among various GFM strategies, virtual synchronous generator (VSG) control emulates synchronous generator (SG) dynamics, enabling renewable energy converters to provide virtual inertia, damping, and voltage support [2], [3].

Like synchronous generators, VSG-controlled converters may lose synchronism following large disturbances [4]. Considerable efforts have therefore been devoted to improving their transient synchronization stability. Some studies reshape the post-fault power angle relationship by adjusting the converter voltage, active power response, or virtual impedance, thereby modifying the acceleration and deceleration areas and enlarging the operating region with a stable post-fault equilibrium [5]–[7]. Other studies improve transient synchronization stability by adaptively tuning control parameters, such as power feedback gains, virtual inertia, damping coefficients, and current-saturation angles, thereby reshaping the transient power imbalance and energy dissipation [4], [8], [9]. For multi-VSG systems, control logic design based on transient energy functions has also been developed to prevent excessive frequency excursions and promote transient energy dissipation [10]. These studies have provided valuable guidance for designing VSG controls to improve transient stability.

Overall, although VSGs inherit instability mechanisms similar to those of SGs through swing equation emulation, their flexible control allows substantial improvements in transient stability. Among these methods, frequency deviation feedback is one of the simplest and most widely used approaches, providing fast primary frequency response or strong damping [4], [8], [11]. However, most of these studies represent the external grid as an infinite bus operating at the nominal frequency. This assumption is restrictive for weak power systems, where severe disturbances can cause substantial grid frequency deviations.

Following a sudden load increase or generation loss, the resulting power imbalance can drive the grid frequency away from nominal [12], [13]. Primary frequency control arrests the frequency decline, while secondary control restores the nominal frequency over a longer timescale [14]. The grid frequency may therefore remain off-nominal throughout the transient and early recovery stages [15].

An off-nominal grid frequency does not in itself prevent synchronization in conventional power systems. SGs synchronize through electromagnetic interactions with the connected grid and may regain synchronism at an off-nominal frequency [16], [17]. Frequency control subsequently restores the interconnected system to the nominal frequency. However, VSG-controlled converters commonly use the nominal frequency as the reference for damping or frequency regulation [2], [4]. When the grid frequency deviates from nominal, strong feedback toward this fixed reference can sustain a frequency difference between the converter and the grid, causing the power-angle deviation to accumulate and potentially leading to loss of synchronism.

Accurately estimating the actual grid frequency and using it as the VSG synchronization reference is key to addressing this problem. Wide-area monitoring systems (WAMS) can provide remote grid frequency measurements, but require communication infrastructure and introduce communication delays [18]. Local phase-locked loop (PLL) measurements offer an alternative, but face two limitations. First, frequency divider theory [19] indicates that the terminal frequency can be dominated by the local VSG frequency in weak grids, so that it may not accurately represent the remote grid frequency. Second, when PLL measurements are used for feedback control, inappropriate PLL tuning can compromise system stability [20]. Further challenges arise under practical operating conditions. VSGs may switch between constant voltage control (CVC) and current limiting control (CLC) during transients, while GFM converters often operate alongside GFL-controlled renewable energy converters at the same point of common coupling [21]. Estimating the grid frequency from local measurements while accounting for these operating conditions therefore remains an important challenge.

This paper proposes a method based on complex frequency theory to enhance VSG synchronization stability by adapting its frequency reference to the actual grid frequency. The main contributions are as follows.

  1. Transient energy analysis reveals how time-varying grid frequency can introduce effective negative damping into the relative dynamics between the VSG and the grid.

  1. A method based on complex frequency theory is developed to estimate the equivalent Thevenin grid frequency without remote communication. It accommodates transitions between normal and current-limiting control of GFM converters, as well as parallel operation with GFL converters.

  1. The estimated grid frequency is used as an adaptive reference for VSG frequency feedback. This allows strong damping to improve stability during grid frequency dips while retaining its benefits during voltage sags. The proposed method is also robust to grid parameter uncertainty and variations in filter settings.

Impact of Grid Frequency Deviations on VSG Transient Synchronization Stability

System Overview

Consider the grid-connected system shown in Figure 1. The VSG has an internal voltage Eˉ=Eejθe\bar{E} = E e^{j\theta_e} and a virtual frequency ωvsg=θ˙e\omega_{\mathrm{vsg}} = \dot{\theta}_e. The outputs of the VSG and parallel GFL sources are combined at the point of common coupling (PCC). The connection from the PCC to the Thévenin equivalent grid includes step-up transformers and transmission lines, collectively modeled as the equivalent impedance Zˉ=R+jX\bar{Z} = R + jX. The positive-sequence PCC voltage and total current flowing from the PCC toward the grid are denoted by Vˉ=Vejθv\bar{V} = V e^{j\theta_v} and Iˉ=Iejθi\bar{I} = I e^{j\theta_i}, respectively. The Thevenin equivalent grid voltage is Vˉg=Vgejθg\bar{V}_g = V_g e^{j\theta_g}, with instantaneous frequency ωg=θ˙g\omega_g = \dot{\theta}_g. Following a large disturbance, ωg\omega_g may deviate substantially from the nominal frequency ωN\omega_N.

Topology of the grid-connected VSG system

Figure 1. Topology of the grid-connected VSG system.

Mechanism of Synchronization Instability Under Grid Frequency Decline

To explain VSG transient synchronization stability during grid frequency dips, the theoretical analysis uses a reduced-order model. The controller hardware-in-the-loop (CHIL) tests, however, use the full nonlinear switching model to validate the proposed method.

Let δ=θe−θg\delta= \theta_e - \theta_g denote the VSG power angle relative to the grid, with relative frequency Δω=δ˙=ωvsg−ωg\Delta\omega= \dot{\delta} = \omega_{\mathrm{vsg}} - \omega_g. The frequency deviation used for feedback is eω=ωvsg−ωrefe_\omega= \omega_{\mathrm{vsg}} - \omega_{\mathrm{ref}}, where ωref\omega_{\mathrm{ref}} is the feedback reference.

Frequency feedback can be implemented through the active power reference, the damping loop, or both. When these paths use the same frequency reference, their proportional and integral contributions can be combined. The dynamic equations of the VSG synchronization loop are

Me˙ω=Pm−Pe(δ)−Deω−Kiz,(1)M\dot{e}_\omega= P_m - P_e(\delta) - D e_\omega- K_i z, \tag*{(1)}
z˙=eω,(2)\dot{z} = e_\omega, \tag*{(2)}
ωvsg=eω+ωref.(3)\omega_{\mathrm{vsg}} = e_\omega+ \omega_{\mathrm{ref}}. \tag*{(3)}

where MM is the virtual inertia, PmP_m is the active power reference before frequency-feedback correction, and Pe(δ)P_e(\delta) is the electrical output power. The coefficients DD and KiK_i represent the combined proportional and integral gains, respectively. Conventional control uses ωref=ωN\omega_{\mathrm{ref}} = \omega_N, so that eω=Δω+(ωg−ωN)e_\omega= \Delta\omega+ (\omega_g - \omega_N). Since ωN\omega_N is constant, e˙ω=Δω˙+ω˙g\dot{e}_\omega= \Delta\dot{\omega} + \dot{\omega}_g, which gives

MΔω˙=Pm−Pe(δ)−DΔω−Kiz−D(ωg−ωN)−Mω˙g,(4)M\Delta\dot{\omega} = P_m - P_e(\delta) - D\Delta\omega- K_i z - D(\omega_g - \omega_N) - M\dot{\omega}_g, \tag*{(4)}
δ˙=Δω,(5)\dot{\delta} = \Delta\omega, \tag*{(5)}
z˙=Δω+(ωg−ωN).(6)\dot{z} = \Delta\omega+ (\omega_g - \omega_N). \tag*{(6)}

Let δs\delta_s denote the stable equilibrium point angle. The transient energy of the relative motion is defined as

W(δ,Δω)=12MΔω2+∫δsδ[Pe(ξ)−Pm] dξ.(7)W(\delta, \Delta\omega) = \frac{1}{2}M\Delta\omega^2 + \int_{\delta_s}^{\delta} [P_e(\xi) - P_m]\,\mathrm{d}\xi. \tag*{(7)}

Differentiating (7) and substituting (4) and (5) gives

W˙=Δω[MΔω˙+Pe(δ)−Pm]=−DΔω2−KizΔω−D(ωg−ωN)Δω−Mω˙gΔω.(8)\dot{W} = \Delta\omega[M\Delta\dot{\omega} + P_e(\delta) - P_m] = -D\Delta\omega^2 - K_i z\Delta\omega - D(\omega_g - \omega_N)\Delta\omega- M\dot{\omega}_g\Delta\omega. \tag*{(8)}

The proportional feedback produces both the dissipative term −DΔω2-D\Delta\omega^2 and the reference mismatch term −D(ωg−ωN)Δω-D(\omega_g - \omega_N)\Delta\omega. The reference mismatch also enters the integral feedback through z˙=Δω+(ωg−ωN)\dot{z} = \Delta\omega+ (\omega_g - \omega_N).

During a grid frequency decline below nominal, −D(ωg−ωN)Δω-D(\omega_g - \omega_N)\Delta\omega and −Mω˙gΔω-M\dot{\omega}_g\Delta\omega both inject energy when Δω>0\Delta\omega> 0, opposing the damping dissipation. As the VSG approaches ωg\omega_g during the frequency dip, zz decreases once ωvsg<ωN\omega_{\mathrm{vsg}} < \omega_N. If zz becomes negative while Δω>0\Delta\omega> 0, −KizΔω-K_i z\Delta\omega also injects energy. Together, these energy inputs can exceed damping dissipation, causing transient energy to grow and threatening synchronization stability.

If ωref=ωg\omega_{\mathrm{ref}} = \omega_g, then eω=Δωe_\omega= \Delta\omega. Damping therefore acts on the actual frequency difference between the VSG and the grid. This choice can improve transient synchronization stability during a grid frequency dip.

Grid Frequency Estimation and Adaptive Frequency Feedback

This section develops a local estimator of the Thévenin grid frequency based on complex frequency theory. The estimated frequency then replaces the nominal reference in VSG frequency reconstruction and feedback.

Independent Reference Frame and Network Model

To avoid introducing PLL tracking errors into the estimator, the PCC voltage and current are transformed into an independent dqdq reference frame with angle

θ0(t)=θ0(t0)+ωN(t−t0),(9)\theta_0(t) = \theta_0(t_0) + \omega_N(t-t_0), \tag*{(9)}

where θ0(t)\theta_0(t) is the angle of the independent dqdq reference frame at time tt and ωN\omega_N is the nominal angular frequency. The initial time and angle are denoted by t0t_0 and θ0(t0)\theta_0(t_0), respectively.

The positive-sequence PCC voltage and current in this frame are expressed as

V‾dq=V‾e−jθ0=vd+jvq,\overline{V}^{dq} = \overline{V}e^{-j\theta_0} = v_d + jv_q,
I‾dq=I‾e−jθ0=id+jiq,(10)\overline{I}^{dq} = \overline{I}e^{-j\theta_0} = i_d + ji_q, \tag*{(10)}

where vdv_d and vqv_q are the dd- and qq-axis voltage components, and idi_d and iqi_q are the corresponding current components.

The Thévenin grid voltage in the same frame is

V‾gdq=V‾ge−jθ0=Vgej(θg−θ0),(11)\overline{V}^{dq}_g = \overline{V}_g e^{-j\theta_0} = V_g e^{j(\theta_g-\theta_0)}, \tag*{(11)}

where VgV_g and θg\theta_g are the magnitude and phase angle of the Thévenin grid voltage, respectively.

Retaining the inductor dynamics gives

V‾−V‾g=RI‾+LI‾˙,(12)\overline{V} - \overline{V}_g = R\overline{I} + L\dot{\overline{I}}, \tag*{(12)}

where RR is the equivalent resistance and L=X/ωNL = X/\omega_N is the equivalent inductance, with XX denoting the reactance at the nominal frequency.

Substituting I‾˙=ejθ0(I‾˙dq+jωNI‾dq)\dot{\overline{I}} = e^{j\theta_0}(\dot{\overline{I}}^{dq} + j\omega_N\overline{I}^{dq}) into (12) gives

V‾gdq=V‾dq−(R+jX)I‾dq−LI‾˙dq.(13)\overline{V}^{dq}_g = \overline{V}^{dq} - (R+jX)\overline{I}^{dq} - L\dot{\overline{I}}^{dq}. \tag*{(13)}

Grid Frequency Estimation Based on Complex Frequency

The complex frequency of the Thévenin grid voltage in the stationary reference frame is defined as [22]

η‾g=V‾˙gV‾g=ρg+jωg,(14)\overline{\eta}_g = \frac{\dot{\overline{V}}_g}{\overline{V}_g} = \rho_g + j\omega_g, \tag*{(14)}

where the real part ρg=V˙g/Vg\rho_g = \dot{V}_g/V_g describes the relative rate of change of the voltage magnitude, and the imaginary part ωg=θ˙g\omega_g = \dot{\theta}_g is the grid angular frequency.

Combining (11) and (14) gives the complex frequency relative to the independent dqdq reference frame as

V‾˙gdqV‾gdq=η‾g−jωN=ρg+j(ωg−ωN).(15)\frac{\dot{\overline{V}}^{dq}_g}{\overline{V}^{dq}_g} = \overline{\eta}_g - j\omega_N = \rho_g + j(\omega_g-\omega_N). \tag*{(15)}

Taking the imaginary part of (15) gives the grid angular frequency as

ωg=ωN+Im⁡(V‾˙gdqV‾gdq).(16)\omega_g = \omega_N + \operatorname{Im}\left(\frac{\dot{\overline{V}}^{dq}_g}{\overline{V}^{dq}_g}\right). \tag*{(16)}

From (13), the dd- and qq-axis components of the reconstructed grid voltage are

vgd=vd−Rid+Xiq−Li˙d,v_{gd} = v_d - Ri_d + Xi_q - L\dot{i}_d,
vgq=vq−Riq−Xid−Li˙q.(17)v_{gq} = v_q - Ri_q - Xi_d - L\dot{i}_q. \tag*{(17)}

Substituting V‾gdq=vgd+jvgq\overline{V}^{dq}_g = v_{gd} + jv_{gq} into (16) gives the grid angular frequency estimate as

ω^g=ωN+vgdv˙gq−vgqv˙gdvgd2+vgq2.(18)\hat{\omega}_g = \omega_N + \frac{v_{gd}\dot{v}_{gq} - v_{gq}\dot{v}_{gd}}{v_{gd}^{2}+v_{gq}^{2}}. \tag*{(18)}

Since the derivation does not depend on the VSG control law, the same estimation formula applies to both CVC and CLC.

In principle, the reconstructed Thévenin voltage V‾gdq\overline{V}^{dq}_g, rather than the local PCC voltage, could also be supplied to a PLL or a frequency-locked loop (FLL) to obtain ω^g\hat{\omega}_g. However, a PLL or FLL is itself a closed-loop tracker whose convergence depends on the voltage magnitude and on its own gain tuning. When the VSG switches to CLC, the reconstructed voltage may change abruptly, and the tracking dynamics would be coupled into the VSG synchronization loop through the frequency feedback, introducing an additional stability concern. The proposed estimator avoids this issue because it is an algebraic computation followed by low-pass filtering, so its dynamics are determined only by the filter bandwidth, which is examined in Section IV.

Adaptive Frequency Feedback and Implementation

As shown in Figure 2, the estimated grid frequency is used as the reference for frequency reconstruction and feedback, giving

VSG active power control with adaptive frequency feedback

Figure 2. VSG active power control with adaptive frequency feedback.

ωref=ω^g,eω=ωvsg−ω^g.(19)\omega_{\mathrm{ref}} = \hat{\omega}_g,\qquad e_\omega= \omega_{\mathrm{vsg}}-\hat{\omega}_g. \tag*{(19)}

The feedback error is therefore the difference between the VSG frequency and the estimated grid frequency. The first integrator produces eωe_\omega according to (1). The controller then adds ω^g\hat{\omega}_g to reconstruct ωvsg\omega_{\mathrm{vsg}} through (3). The reconstructed absolute frequency drives the angle integrator.

The independent reference angle avoids introducing PLL tracking dynamics or VSG angle dynamics into the coordinate transformation. In a synchronous steady state, the reconstructed grid voltage rotates at ωg−ωN\omega_g - \omega_N in this frame. Near nominal frequency, its components vary slowly, which facilitates filtering and numerical differentiation. In the digital implementation, the required derivatives are computed using backward differences followed by first-order low-pass filtering. The reconstructed voltage is filtered before differentiation, and the resulting frequency estimate is further smoothed before feedback.

Under ideal estimation, ω^g=ωg\hat{\omega}_g = \omega_g and eω=Δωe_\omega= \Delta\omega. Substituting this relation into (1) gives

MΔω˙=Pm−Pe(δ)−DΔω−Kiz,(20)M\Delta\dot{\omega} = P_m - P_e(\delta) - D\Delta\omega- K_i z, \tag*{(20)}

with z˙=Δω\dot{z} = \Delta\omega. Both the reference mismatch and the explicit grid frequency derivative term disappear from the relative dynamics. The energy rate therefore becomes

W˙=−DΔω2−KizΔω.(21)\dot{W} = -D\Delta\omega^2 - K_i z\Delta\omega. \tag*{(21)}

The proportional term is dissipative for D>0D > 0. Over an interval with Δω>0\Delta\omega> 0, both the power angle and the integral state increase. For Ki>0K_i > 0, the increase in zz relative to its value at the start of the interval provides an additional negative contribution to the energy rate.

Practical Considerations

Next, we examine how errors introduced by grid impedance uncertainty and filtering affect frequency estimation and feedback control.

In a synchronous steady state, the positive-sequence voltage and current rotate at the same frequency. With constant line parameter errors, the reconstructed grid voltage can therefore be expressed as

V^gdq=cVˉgdq,(22)\hat{V}_g^{dq} = c\bar{V}_g^{dq}, \tag*{(22)}

where V^gdq\hat{V}_g^{dq} denotes the reconstructed voltage and cc is a constant complex factor representing magnitude and phase errors. For c≠0c \ne0, ideal differentiation gives

ω^g=ωN+Im⁡(cV^˙gdqcV^gdq)=ωg.(23)\hat{\omega}_g = \omega_N + \operatorname{Im}\left(\frac{c\dot{\hat{V}}_g^{dq}}{c\hat{V}_g^{dq}}\right) = \omega_g. \tag*{(23)}

Thus, constant line parameter errors do not bias the ideal frequency estimate in steady state, provided the reconstructed grid voltage remains nonzero.

During transients, reconstruction errors and filtering can affect the estimated frequency. Let εω=ω^g−ωg\varepsilon_\omega= \hat{\omega}_g - \omega_g denote the frequency estimation error, so that eω=Δω−εωe_\omega= \Delta\omega- \varepsilon_\omega. Since e˙ω=Δω˙−ε˙ω\dot{e}_\omega= \Delta\dot{\omega} - \dot{\varepsilon}_\omega, the relative dynamics become

MΔω˙=Pm−Pe(δ)−DΔω−Kiz+Dεω+Mε˙ω,(24)M\Delta\dot{\omega} = P_m - P_e(\delta) - D\Delta\omega- K_i z + D\varepsilon_\omega+ M\dot{\varepsilon}_\omega, \tag*{(24)}

with z˙=Δω−εω\dot{z} = \Delta\omega- \varepsilon_\omega. The corresponding energy rate is

W˙=−DΔω2−KizΔω+(Dεω+Mε˙ω)Δω.(25)\dot{W} = -D\Delta\omega^2 - K_i z\Delta\omega + (D\varepsilon_\omega+ M\dot{\varepsilon}_\omega)\Delta\omega. \tag*{(25)}

The above expression shows that the estimation error affects the energy rate through its value εω\varepsilon_\omega and its rate of change ε˙ω\dot{\varepsilon}_\omega. For disturbances that do not lead to system collapse, the grid frequency deviation and its rate of change remain bounded. We further assume that the estimation error and its derivative are bounded:

∣εω∣≤εω,max⁡,∣ε˙ω∣≤ε˙ω,max⁡,(26)\lvert\varepsilon_\omega\rvert\le\varepsilon_{\omega,\max}, \qquad \lvert\dot{\varepsilon}_\omega\rvert\le\dot{\varepsilon}_{\omega,\max}, \tag*{(26)}

where the right-hand sides are constant bounds on the error magnitude and its derivative magnitude, respectively. Then

−DΔω2+(Dεω+Mε˙ω)Δω≤−DΔω2+(Dεω,max⁡+Mε˙ω,max⁡)∣Δω∣.(27)-D\Delta\omega^2 + (D\varepsilon_\omega+ M\dot{\varepsilon}_\omega)\Delta\omega \le-D\Delta\omega^2 + (D\varepsilon_{\omega,\max} + M\dot{\varepsilon}_{\omega,\max})\lvert\Delta\omega\rvert. \tag*{(27)}

For D>0D > 0, this combined contribution is negative when

∣Δω∣>εω,max⁡+MDε˙ω,max⁡.(28)|\Delta\omega| > \varepsilon_{\omega,\max} + \frac{M}{D}\dot{\varepsilon}_{\omega,\max}. \tag*{(28)}

Increasing DD lowers this threshold.

The estimation error affects the integral state through z˙=eω=Δω−εω\dot{z}=e_{\omega}=\Delta\omega-\varepsilon_{\omega}. If Δω>0\Delta\omega> 0 and ∣εω∣<Δω|\varepsilon_{\omega}| < \Delta\omega hold throughout an interval, then z˙>0\dot{z} > 0. For Ki>0K_i > 0, the resulting increase in zz above its value at the start of the interval provides an additional negative contribution to the energy rate.

The proposed feedback therefore provides a degree of robustness to frequency estimation errors.

Controller Hardware in the Loop Test

Controller hardware in the loop (CHIL) tests are conducted to validate the proposed method under grid frequency disturbances. The CHIL platform is shown in Figure 3. The power circuit is simulated on a Typhoon HIL404, with two converters connected in parallel at the PCC. The converter controllers are implemented on TMS320F28377S DSPs. The default grid and controller parameters are listed in Table II in the Appendix.

Controller hardware-in-the-loop experimental setup

Figure 3. Controller hardware-in-the-loop experimental setup.

ParameterSymbolValue
Power circuit
Power baseSbaseS_{\mathrm{base}}15 kVA
Nominal phase voltage (rms)VNV_N220 V
Nominal frequencyfNf_N50 Hz
DC voltageVdcV_{\mathrm{dc}}750 V
Filter inductanceLfL_f1 mH
Filter resistanceRfR_f0.01 Ω\Omega
Filter capacitanceCfC_f20 μ\muF
Capacitor damping resistanceRdR_d1 Ω\Omega
Converter branch resistancesRc1,Rc2R_{c1}, R_{c2}0.1, 0.1 mΩ\Omega
Converter branch inductancesLc1,Lc2L_{c1}, L_{c2}0.1, 0.5 mH
Grid resistanceRR0.01 Ω\Omega
Grid inductanceLL20 mH
5th voltage harmonicV5/V1V_5/V_15%
7th voltage harmonicV7/V1V_7/V_13%
Phase A voltage (unbalanced)Va/VNV_a/V_N1.00 p.u.
Phase B voltage (unbalanced)Vb/VNV_b/V_N0.95 p.u.
Phase C voltage (unbalanced)Vc/VNV_c/V_N1.05 p.u.
VSG control
Equivalent inertia coefficientM0M_01.9635 W s2^2/rad
Base damping coefficientD0D_04000 W s/rad
Voltage feedback gainKvK_v1500 var/V
Reactive power integral gainKqK_q0.01 V/(var s)
Virtual resistanceRvR_v2.2 Ω\Omega
Virtual inductanceLvL_v6 mH
Current PI gainskpiVSG,kiiVSGk_{pi}^{\mathrm{VSG}}, k_{ii}^{\mathrm{VSG}}3 V/A, 330 V/(A s)
GFL control
Positive-sequence current PI gainskpi+,kii+k_{pi}^{+}, k_{ii}^{+}3 V/A, 330 V/(A s)
Negative-sequence current PI gainskpi−,kii−k_{pi}^{-}, k_{ii}^{-}2 V/A, 51 V/(A s)
PLL PI gainskpPLL,kiPLLk_{p\mathrm{PLL}}, k_{i\mathrm{PLL}}0.7 rad/(V s), 12.8 rad/(V s2^2)
Frequency estimation
Reconstructed voltage filter cutofffvf_v50 Hz
Derivative filter cutofffdf_d80 Hz
Estimated frequency filter cutofffωf_\omega5 Hz
Digital implementation
Real-time simulation stepTsimT_{\mathrm{sim}}1 μ\mus
Controller sampling periodTsT_s78.125 μ\mus
Switching frequencyfswf_{\mathrm{sw}}12.8 kHz

Table II. Default Parameters of the CHIL Platform

All tests include fifth- and seventh-order voltage harmonics of 5% and 3% of the fundamental voltage, respectively. The default phase A, B, and C rms voltages are 1.00, 0.95, and 1.05 p.u., respectively. Cases C1 to C11 use one VSG and one GFL converter. Case C12 uses two VSGs with different control parameters. The tests cover frequency dips, combined frequency and voltage disturbances, and changes in grid conditions, power output, and controller settings.

The twelve cases are summarized in Table I. Each case compares two frequency references under the same operating conditions and disturbance. Panel (a) shows the estimation performance and system response with ωref=ωN\omega_{\mathrm{ref}}=\omega_N. Panel (b) shows the response with ωref=ω^g\omega_{\mathrm{ref}}=\hat{\omega}_g. The grid frequency falls from 50 to 48.3 Hz, with a maximum decline rate of approximately 1.8 Hz/s. Cases C9 and C10 also include a voltage sag.

CaseOperating conditionsDisturbance
C1PVSG=5 kWP_{\mathrm{VSG}} = 5\ \mathrm{kW}, PGFL=4.67 kWP_{\mathrm{GFL}} = 4.67\ \mathrm{kW}. CVC operation (Imax⁡=3INI_{\max} = 3I_N). R/X≈0.0016R/X \approx0.0016, SCR = 2.39, fω=5 Hzf_\omega= 5\ \mathrm{Hz}.Frequency dip from 50 to 48.3 Hz. Maximum ∣dfg/dt∣≈1.8 Hz/s\lvert df_g/dt\rvert\approx1.8\ \mathrm{Hz/s}.
C2Power settings as in C1. Transitions between CVC and CLC (Imax⁡=1.5INI_{\max} = 1.5I_N). R/X≈0.0016R/X \approx0.0016, SCR = 2.39, fω=5 Hzf_\omega= 5\ \mathrm{Hz}.Frequency dip as in C1.
C3As in C2, with R/X=0.6R/X = 0.6 and updated impedance parameters in the estimator.Frequency dip as in C1.
C4As in C2, with R/X=0.6R/X = 0.6 and the original impedance parameters retained in the estimator.Frequency dip as in C1.
C5As in C2, with SCR = 1.5, unchanged R/XR/X, and updated impedance parameters in the estimator.Frequency dip as in C1.
C6As in C2, with SCR = 1.5, unchanged R/XR/X, and the original impedance parameters retained in the estimator.Frequency dip as in C1.
C7As in C2, with fω=20 Hzf_\omega= 20\ \mathrm{Hz}.Frequency dip as in C1.
C8As in C2, with fω=0.5 Hzf_\omega= 0.5\ \mathrm{Hz}.Frequency dip as in C1.
C9As in C2, with PVSG=8 kWP_{\mathrm{VSG}} = 8\ \mathrm{kW}.Frequency dip as in C1 and balanced voltage sag (Va=Vb=Vc=0.1 p.u.V_a = V_b = V_c = 0.1\ \mathrm{p.u.}, hold 2 s).
C10As in C2, with PGFL=7.47 kWP_{\mathrm{GFL}} = 7.47\ \mathrm{kW}.Frequency dip as in C1 and unbalanced voltage sag (Va=0.30V_a = 0.30, Vb=0.285V_b = 0.285, Vc=0.315 p.u.V_c = 0.315\ \mathrm{p.u.}, hold 2 s).
C11As in C2, with additional frequency PI feedback: kpω=50k_{p\omega} = 50 and kiω=500k_{i\omega} = 500.Frequency dip as in C1.
C12Two VSGs with P1=5 kWP_1 = 5\ \mathrm{kW} and P2=3 kWP_2 = 3\ \mathrm{kW}. No GFL converter. M1=M0M_1 = M_0, M2=1.25M0M_2 = 1.25M_0, D1=4000D_1 = 4000, D2=3000D_2 = 3000, Imax⁡,1=1.2INI_{\max,1} = 1.2I_N, Imax⁡,2=1.5INI_{\max,2} = 1.5I_N. Other grid and filter settings follow C2.Frequency dip as in C1.

Table I. Operating Conditions of the CHIL Tests

In Case C1, the VSG and GFL converter deliver 5 and 4.67 kW, respectively. The VSG current limit is set to 3IN3I_N so that it remains in CVC during the disturbance. As shown in Figure 4(a), the VSG frequency fvsgf_{\mathrm{vsg}} does not follow the actual grid frequency decline and exhibits large oscillations. The accumulated frequency difference leads to loss of synchronism. The estimated frequency f^g\hat{f}_g follows the grid frequency dip and recovery, with transient deviations during the VSG oscillations. With ωref=ω^g\omega_{\mathrm{ref}}=\hat{\omega}_g, the VSG follows the grid frequency and its relative angle remains bounded, as shown in Figure 4(b).

CHIL results for Case C1 with the VSG operating in CVC

Figure 4. CHIL results for Case C1 with the VSG operating in CVC.

Case C2 reduces the current limit to 1.5IN1.5I_N. With the nominal frequency reference, the VSG repeatedly switches between CVC and CLC, as indicated in Figure 5(a). The estimate still captures the grid frequency trajectory, although ripple increases during the mode transitions. With the estimated frequency reference, the large frequency oscillations are suppressed. The VSG remains in CVC throughout the disturbance in Figure 5(b).

CHIL results for Case C2 with transitions between CVC and CLC

Figure 5. CHIL results for Case C2 with transitions between CVC and CLC.

Cases C3 and C4 increase R/XR/X to 0.6. The estimator uses updated impedance parameters in C3 and retains the original parameters in C4. With matched parameters, the estimate follows the grid frequency during the extended CLC interval in Figure 6(a). Parameter mismatch produces larger transient estimation errors in Figure 7(a). Nevertheless, the estimated frequency feedback maintains stable frequency tracking in both cases, as shown in Figure 6(b) and Figure 7(b).

CHIL results for Case C3 at $R/X = 0.6$ with matched impedance parameters

Figure 6. CHIL results for Case C3 at R/X=0.6R/X = 0.6 with matched impedance parameters.

CHIL results for Case C4 at $R/X = 0.6$ with mismatched impedance parameters

Figure 7. CHIL results for Case C4 at R/X=0.6R/X = 0.6 with mismatched impedance parameters.

Cases C5 and C6 reduce the SCR from 2.39 to 1.5 while keeping R/XR/X unchanged. The impedance parameters used by the estimator are updated in C5 and left unchanged in C6. With the nominal reference, the VSG loses synchronism in both cases. Its frequency oscillates, and it repeatedly switches between CVC and CLC. The estimation errors are larger with mismatched parameters, as shown in Figure 9(a). With the estimated frequency reference, the VSG follows the grid frequency without entering CLC in either case, as shown in Figure 8(b) and Figure 9(b). These results demonstrate effective control under weaker grid conditions and impedance uncertainty.

CHIL results for Case C5 at SCR $= 1.5$ with matched impedance parameters

Figure 8. CHIL results for Case C5 at SCR =1.5= 1.5 with matched impedance parameters.

CHIL results for Case C6 at SCR = 1.5 with mismatched impedance parameters

Figure 9. CHIL results for Case C6 at SCR = 1.5 with mismatched impedance parameters.

Cases C7 and C8 change the cutoff frequency of the final frequency estimation filter from 5 Hz to 20 and 0.5 Hz,

respectively. The higher cutoff allows more ripple in the estimate, as shown in Figure 10(a). The lower cutoff produces a smoother estimate but introduces a visible delay during the frequency decline in Figure 11(a). Despite these differences, the estimated frequency feedback suppresses the large VSG oscillations in both cases. The responses in Figure 10(b) and Figure 11(b) remain stable.

CHIL results for Case C7 with an estimated frequency filter cutoff of 20 Hz

Figure 10. CHIL results for Case C7 with an estimated frequency filter cutoff of 20 Hz.

CHIL results for Case C8 with an estimated frequency filter cutoff of 0.5 Hz

Figure 11. CHIL results for Case C8 with an estimated frequency filter cutoff of 0.5 Hz.

Case C9 increases the VSG output to 8 kW and combines the frequency dip with a balanced voltage sag. All three phase voltages fall to 0.1 p.u. and remain at this level for 2 s. With the nominal reference, the VSG loses synchronism, as shown in Figure 12(a). With the estimated frequency reference, the voltage sag still causes current saturation, as shown in Figure 12(b). However, the VSG subsequently returns to CVC

CHIL results for Case C9 with increased $P_{\mathrm{VSG}}$ under a frequency dip and a severe balanced voltage sag

Figure 12. CHIL results for Case C9 with increased PVSGP_{\mathrm{VSG}} under a frequency dip and a severe balanced voltage sag.

and follows the recovering grid frequency.

In Case C10, the GFL output is increased to 7.47 kW. The disturbance combines a frequency dip with an unbalanced voltage sag. The phase A, B, and C voltages fall to 0.30, 0.285, and 0.315 p.u., respectively, and remain at these levels for 2 s. With the nominal reference, the VSG loses synchronism and experiences repeated current saturation, as shown in Figure 13(a).

CHIL results for Case C10 under a frequency dip and an unbalanced voltage sag

Figure 13. CHIL results for Case C10 with increased PGFLP_{\mathrm{GFL}} under a frequency dip and an unbalanced voltage sag.

With the estimated frequency reference, the VSG briefly enters CLC and then returns to CVC. The transient oscillations decay, and stable frequency tracking is restored, as shown in Figure 13(b).

Case C11 adds frequency PI feedback with kpω=50k_{p\omega}=50 and kiω=500k_{i\omega}=500. With ωref=ωN\omega_{\mathrm{ref}}=\omega_{\mathrm{N}}, the VSG exhibits large frequency oscillations and repeated current saturation after losing synchronism, as shown in Figure 14(a). Replacing the nominal reference with ω^g\hat{\omega}_{g} maintains synchronism and keeps the VSG in CVC, as shown in Figure 14(b). The proposed reference modification therefore remains effective with additional integral feedback.

CHIL results for Case C11 with additional frequency PI feedback

Figure 14. CHIL results for Case C11 with additional frequency PI feedback.

Case C12 replaces the GFL converter with a second VSG. The two VSGs deliver 5 and 3 kW and have different inertia, damping, and current limits. With the nominal reference, both VSGs lose synchronism with the grid, as shown in Figure 15(a). The estimator still captures the grid frequency dip and recovery. With the estimated frequency reference, both VSGs closely follow the grid frequency, as shown in Figure 15(b). Stable operation is thus achieved with two parallel VSGs despite their different parameters.

CHIL results for Case C12 with two parallel VSGs of different inertia and damping

Figure 15. CHIL results for Case C12 with two parallel VSGs of different inertia and damping.

These tests verify grid frequency estimation in both CVC and CLC. Impedance mismatch and filter settings affect transient estimation accuracy, but the proposed feedback maintains or restores stable frequency tracking in all twelve cases. The results demonstrate robustness to the tested grid and controller parameter variations, voltage harmonics, and voltage unbalance. The method is also effective under combined frequency and voltage disturbances and with multiple parallel VSGs.

Electromechanical Transient Simulation of the IEEE 39-Bus System

Electromechanical transient simulations are conducted in MATLAB to evaluate the proposed grid frequency estimator and adaptive frequency feedback in a multimachine system. The modified IEEE 39-bus test system is shown in Figure 16. The network is represented by algebraic phasor equations with constant-impedance loads. The differential equations are solved using the modified Euler method with a time step of 100 μ\mus.

Topology of the modified IEEE 39-bus system

Figure 16. Topology of the modified IEEE 39-bus system.

Gen 9, originally connected to bus 38, is replaced by a VSG. A GFL converter, denoted Gen 11, is connected in parallel at bus 38. Their initial active power outputs in the default power flow are 530 MW and 400 MW, respectively. The total active power generation and load are 7270.35 MW and 7112.17 MW, respectively. The nominal frequency is 60 Hz.

The power and voltage bases are 100 MVA and 345 kV. The prefault Thévenin impedance of the external grid seen from bus 38 is Zˉ=0.02474+j0.09341\bar{Z}=0.02474+\mathrm{j}0.09341 p.u.. The VSG has an inertia time constant of 6.9 s and a damping coefficient of 1500 p.u.. All rotor angles are measured relative to Gen 2 at bus 31.

At t=10t=10 s, the load is increased by 4000 MW and 667 Mvar. Both increments are distributed among the load buses in proportion to their initial active power demands. The same disturbance and operating conditions are used to evaluate frequency estimation and adaptive frequency feedback.

Fig. 17 shows the response when the VSG frequency feedback uses the nominal reference ωref=ωN\omega_{\mathrm{ref}}=\omega_N. Following the load increase, the synchronous generator frequencies fall. The large damping coefficient keeps the VSG frequency near the nominal value. A persistent frequency difference develops between the VSG and the synchronous generators. The VSG consequently loses synchronism, as shown in Figure 17(b). The proposed estimator follows the frequency decline and subsequent oscillations of the synchronous generators, with brief deviations during the transient. It therefore captures the grid frequency response even when the VSG loses synchronism.

Grid frequency estimation performance

Figure 17. Grid frequency estimation performance.

The nominal reference ωN\omega_N in the VSG active power loop is then replaced by the estimated grid frequency ω^g\hat{\omega}_g, as defined in (19). The resulting response is shown in Fig. 18. The VSG and synchronous generators converge to a common frequency below 60 Hz. Their relative rotor angles settle to constant values, indicating that synchronism is maintained. These results demonstrate the effectiveness of the proposed estimator and adaptive frequency feedback in the multimachine system.

Conclusion

This paper investigates a mechanism of VSG synchronization instability during frequency dips in weak grids. When damping feedback or emulated fast primary frequency control uses the nominal frequency as its reference, it can sustain a frequency difference between the VSG and the grid. The accumulated frequency difference can then lead to power angle

Transient response with the estimated grid frequency as the VSG frequency reference

Figure 18. Transient response with the estimated grid frequency as the VSG frequency reference.

instability. To address this issue, a grid frequency estimator is developed based on complex frequency theory. It uses only voltage and current measurements at the PCC, without additional measurement devices or remote communication. The estimated grid frequency replaces the nominal reference in the VSG frequency feedback loop to improve transient stability. CHIL tests validate the proposed method with a

VSG operating in parallel with a GFL converter and with two parallel VSGs. The tests verify its effectiveness in both CVC and CLC and its robustness to grid and controller parameter variations and impedance uncertainty. Effective control is also demonstrated under voltage harmonics and balanced or unbalanced voltage sags. Simulations of the modified IEEE 39-bus system further confirm the effectiveness of adaptive frequency feedback in a multimachine network.

This control particularly suited to enhancing the transient synchronization stability of renewable energy converters connected to the main grid through weak electrical connections. Future work will further investigate the coordination between VSG synchronization stability and inertial support.

System Parameters

References

References

  1. [1]K. Strunz, K. Alnumen, C. Wulkow, M. Kuschke, M. Valdescudero, and X. Guillaud, “Enabling 100% renewable power systems through electronic grid-forming converter and control: System integration for security, stability, and application to Europe,” Proc. IEEE, vol. 111, no. 7, pp. 891–915, Jul. 2023.
  2. [2]J. Matevosyan, B. Badrzadeh, T. Prevost, E. Quitmann, D. Ramasubramanian, H. Urdual, S. Achille, J. MacDowell, S. H. Huang, V. Vital, J. O’Sullivan, and R. Quint, “Grid-forming inverters: Are they the key for high renewable penetration?” IEEE Power Energy Mag., vol. 21, no. 2, pp. 77–86, Mar./Apr. 2023.
  3. [3]E. A. León and J. M. Mauricio, “Virtual synchronous generator design to improve frequency support of converter-interfaced systems,” IEEE Trans. Energy Convers., vol. 39, no. 2, pp. 862–871, Jun. 2024.
  4. [4]J. Lei, X. Xiang, B. Liu, W. Li, and X. He, “Quantitative and intuitive VSG transient analysis with the concept of damping area approximation,” IEEE Trans. Smart Grid, vol. 14, no. 3, pp. 2477–2480, May 2023.DOI
  5. [5]D. Pan, X. Wang, F. Liu, and R. Shi, “Transient stability of voltage-source converters with grid-forming control: A design-oriented study,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 8, no. 2, pp. 1019–1033, Jun. 2020.DOI
  6. [7]K. Wang, Q. Song, W. Liu, B. Zhao, and P. Zhang, “Reactive power characteristics and FRT strategy for grid-forming converters with virtual impedance-based current limiting,” IEEE Trans. Energy Convers., vol. 40, no. 2, pp. 1165–1177, Jun. 2025.DOI
  7. [8]X. Xiong, C. Wu, P. Cheng, and F. Blaabjerg, “An optimal damping design of virtual synchronous generators for transient stability enhancement,” IEEE Trans. Power Electron., vol. 36, no. 10, pp. 11026–11030, Oct. 2021.DOI
  8. [9]C. Luo, Y. Chen, S. Liao, Z. Wang, Z. Xie, and Z. Lian, “Design-oriented analysis and transient control for VSG with current saturation unit,” IEEE Trans. Power Electron., vol. 40, no. 1, pp. 2472–2483, Jan. 2025.DOI
  9. [10]Y. Li, Y. Lu, J. Yang, J. Liu, T. Mu, H. Ye, and Z. Du, “Synchronization stability of multiple VSGs embedded power system with controller limits,” IEEE Trans. Power Syst., vol. 40, no. 1, pp. 834–849, Jan. 2025.DOI
  10. [11]J. Li, Y. Li, Z. Du, Z. Xu, and Z. Dong, “Damping turning rule of virtual synchronous generator for global stability,” IEEE Trans. Power Del., vol. 38, no. 4, pp. 2650–2660, Aug. 2023.DOI
  11. [12]F. Milano, F. Dörfler, G. Hug, D. J. Hill, and G. Verbič, “Foundations and challenges of low-inertia systems (invited paper),” in Proc. Power Syst. Comput. Conf., Dublin, Ireland, 2018, pp. 1–25.DOI
  12. [13]Y. Zhang, Z. Hao, S. Yang, and B. Zhang, “A system-view optimal additional active power control of wind turbines for grid frequency support,” IEEE Trans. Power Syst., vol. 39, no. 2, pp. 4323–4335, Mar. 2024.
  13. [14]North American Electric Reliability Corporation, “Balancing and frequency control,” NERC Resources Subcommittee, May 2021. [Online]. Available: https://www.nerc.com
  14. [15]Z. Hao, B. Li, S. Yang, S. Huang, and P. Cheng, “Adaptive frequency anchoring for transient synchronization of grid-following renewable energy converters in weak grids,” IEEE Trans. Energy Convers., early access, Nov. 2, 2026, doi: 10.1109/TEC.2026.3749668.DOI
  15. [16]P. Kundur, Power System Stability and Control. New York, NY, USA: McGraw-Hill, 1994.
  16. [17]F. P. de Mello and C. Concordia, “Concepts of synchronous machine stability as affected by excitation control,” IEEE Trans. Power App. Syst., vol. PAS-88, no. 4, pp. 316–329, Apr. 1969.
  17. [18]L. Díez-Maroto, J. Redondo, L. Rouco, and F. Fernández-Bernal, “Lyapunov stability-based wide area control systems for excitation boosters in synchronous generators,” IEEE Trans. Power Syst., vol. 34, no. 1, pp. 194–204, Jan. 2019.DOI
  18. [19]F. Milano and A. Ortega, “Frequency divider,” IEEE Trans. Power Syst., vol. 32, no. 2, pp. 1493–1501, Mar. 2017.
  19. [20]Y. Zhou, X. Xiong, X. Li, and P. Cheng, “Effects of PLL in the damping unit of a grid-connected VSG,” in Proc. IEEE Int. Power Electron. Appl. Conf. Expo. (PEAC), 2022, pp. 589–593.DOI
  20. [21]B. Li, S. Yang, P. Cheng, and Z. Hao, “Transient stability of GFL converters subjected to mode switching of GFM converters,” IEEE Trans. Energy Convers., vol. 41, no. 3, pp. 2254–2268, Sep. 2026.
  21. [22]F. Milano, “Complex frequency,” IEEE Trans. Power Syst., vol. 37, no. 2, pp. 1230–1240, Mar. 2022.

Paper details

vlnc:W_vjh7rz3x
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Engineering

Not recorded for this paper

View statement
Looking for related papers…
Contents