betaFeedback

WorksEngineering

Grid-forming Control of Offshore Wind through a Diode-Rectifier HVDC System: a 2 GW Case Study

Abstract

Diode-rectifier-unit (DRU)-based high-voltage direct current (HVDC) transmission considerably reduces the volume and weight of the offshore converter platform for remote wind farms and, hence, its cost, but shifts the task of forming and regulating the offshore AC network to the wind-turbine converters. This paper examines which grid-forming control strategy is the most suitable for DRU power transfer. The DRU commutation relationship shows that the delivered active power depends on the offshore voltage magnitude (but not on the converter voltage angle), so the angle-based active-power loop of virtual synchronous machine control is not closed through the rectifier. A P−V/Q−ωP-V/Q-ω grid-forming architecture, which regulates active power via the voltage magnitude and derives frequency from the reactive power error, is therefore adopted. Electromagnetic transient simulations of a 1GWaggregated grid-forming turbine on a 66 kV collector network with a ±\pm525 kV bipolar DRU link show that the commanded power is delivered over 800 to 1000MW with the offshore voltage held near nominal. The paper also examines how grid-forming operation changes the allocation of control duties within the Type-4 turbine: once the grid-side converter forms the network, DC link regulation must move to the machine-side converter, which limits the achievable bandwidth of the DC-link control. Extending the system to 2 GW with two 500 MW grid-following units achieves full power transfer in simulation, but with the grid-forming converter loaded beyond its continuous rating and a steady-state offshore frequency deviation of approximately 1.8 Hz.

Introduction

As offshore wind farms move farther from shore and grow in rating, high-voltage direct-current (HVDC) transmission becomes the preferred option for bringing the power ashore. The charging current of a submarine AC cable grows with its length and occupies part of its current rating, which limits the length of a high-voltage AC (HVAC) connection but not of an HVDC one [1]. In the established arrangement, an offshore voltage-source converter forms the offshore AC network and is also the heaviest and most expensive item used in this topology. Diode-rectifier-unit (DRU)-HVDC replaces it with uncontrolled diode bridges, removing the offshore switching devices, their cooling, and much of the accompanying equipment [2, 3].

The price of this simplification is a control problem. An uncontrolled rectifier can neither establish offshore voltage and frequency nor supply the reactive power that its own commutation consumes. These functions must come from the grid-side converters (GSCs) of the wind turbines, so grid-forming (GFM) operation is a physical requirement rather than a preferred mode [2, 3, 4]. This raises the question of how the turbines should do this: which GFM control strategy can both form the offshore network and regulate the power delivered through an uncontrolled rectifier, and which strategy cannot?

Several strategies have been proposed in answer: distributed voltage and frequency droop [2], modified virtual synchronous machine (VSM) schemes [5], power-synchronisation control [6], remote active-power regulation through the onshore terminal [7], adaptive reactive-power allocation [8], and hybrid plants mixing GFM and grid-following (GFL) turbines [9]. They differ mainly in the variable through which active power is controlled, either the converter angle, as in VSM and power-synchronisation control, or the voltage magnitude.

Existing results already point to the voltage magnitude, rather than the converter angle, as the variable that sets the power delivered through a DRU. Power transfer through a diode rectifier is set by the difference between the AC and DC voltage magnitudes [2, 7]. From the dynamics of the offshore network, [10] concluded that active power couples to voltage and reactive power to frequency, the reverse of a conventional grid, an assignment adopted by several subsequent controllers [11, 12]. Even where angle-based control has been used, as in the droop and VSM schemes that [13] operated together in one DRU-connected plant, the total power was set by a separate plant-level controller acting on the voltage magnitude, and the angle-based loops only shared that power among the turbines. What is missing is the other half of the argument. In [12], researchers derive the sensitivities of the DRU equations that motivate the P-V and Q-ω\omega assignment, but neither their work nor the other works show why the angle-based alternative fails.

Existing DRU GFM studies are also typically at or below 1 GW and include AC filters or reactive compensation at the rectifier, for example the 1 GW, ±320\pm320 kV system of [4], the 400 MW monopolar system of [5] and the 200 MW system of [12], and they commonly represent each turbine by its GSC fed from a constant DC source [4, 12], while [10] uses grid-following turbines with an external frequency controller. Such models hide how the GFM duty interacts with the rest of the turbine and how much reactive power the converters must supply in the absence of filters, so whether a suitable controller scales to a full 2 GW plant remains open.

This paper addresses these gaps at industrial scale. Its contributions are: (i) the DRU commutation equation is used to show why VSM angle control fails on a DRU, since the delivered power has no direct control path through the converter angle, and hence why the P-V/Q-ω\omega architecture of [12] suits it; (ii) the architecture is implemented in PSCAD/EMTDC on a 1 GW aggregated Type-4 turbine, tuned by pole placement for the inner loops and electromagnetic transient (EMT) simulation for the outer loops, and validated across operating setpoints; (iii) the GFM turbine is modelled with both its machine-side converter (MSC) and GSC in a 2 GW, ±525\pm525 kV bipolar system with two 24-pulse DRUs and no AC filters, which exposes the conflict between the MSC and GSC over DC-link regulation and places the full DRU reactive demand on the GFM converter; and (iv) the design is extended to a 2 GW hybrid GFM-GFL configuration, showing that the active power injected by the grid-following units raises the reactive power the GFM converter must supply, which with a fixed reactive reference shifts the offshore frequency.

System description and main circuit design

The study starts from a 2 GW PSCAD/EMTDC model provided by TenneT, in which each pole of the ±525\pm525 kV bipole is fed by a pole of an offshore modular multilevel converter (MMC) and four aggregated grid-following Type-4 turbines of 500 MW (560 MVA, 112 units of 5 MVA) are connected, two per pole, Figure 1. The positive and negative poles of the offshore MMC were disconnected and replaced with two 24-pulse DRUs, and all turbines were connected to a common 66 kV collector bus that feeds both DRUs.

Original TenneT model showing four aggregated 500 MW grid-following turbines, offshore MMCs, and onshore MMC bipole

Figure 1. Original TenneT model: four aggregated 500 MW grid-following turbines, two feeding the positive-pole and two the negative-pole offshore MMC, with the ±525\pm525 kV bipolar link and metallic return to the onshore MMC. In this study the offshore MMCs are replaced by two 24-pulse DRUs on a common collector bus, Figure 7.

The final studied system, shown in Figure 7 of § 4.2, is a 2 GW offshore wind farm connected through a bipolar ±525\pm525 kV DRU-HVDC link with metallic return to an onshore bipolar MMC link. Aggregated Type-4 turbines with permanent-magnet synchronous generators (PMSGs) feed a common 66 kV collector bus: WT1 (1000 MW, 200 ×\times 5 MVA, a rating justified in § 4.2) operates GFM, while WT2 and WT3 (500 MW, 560 MVA each) operate GFL. The point of common coupling (PCC) is the terminal of WT1 at the 66 kV collector bus, where VpccV_{\mathrm{pcc}} is measured. Each turbine has an MSC, a DC link, and a GSC with an LC output filter,

Figure 7. Topology of the 2 GW DRU-HVDC offshore wind system: one 1000 MW GFM and two 500 MW GFL aggregated turbines, 66 kV collector bus, two 24-pulse DRUs and a bipolar ±525\pm525 kV link.

as shown in Figure 2. Except where stated otherwise, each turbine is represented by its GSC, with the DC link modelled as an ideal source, thereby isolating the GFM control from the machine-side dynamics.

Figure 2. Aggregated Type-4 wind turbine: PMSG, MSC, DC link, GSC with LC filter (LfL_f, CfC_f) and turbine transformer to the 66 kV PCC.

Each pole is fed by a 24-pulse DRU of four series-connected 6-pulse bridges supplied through two phase-shift transformers (+22.5∘+22.5^\circ, +7.5∘+7.5^\circ) and Y/Y-Y/Δ\Delta bridge transformers, Figure 3, cancelling characteristic harmonics below the 23rd order [14]. The averaged DC pole voltage follows the commutation equation:

One 24-pulse DRU pole

Figure 3. One 24-pulse DRU pole: two phase-shift transformers, four bridge transformers and four series 6-pulse diode bridges.

Udru=4×32πU0S6−12πXTIdc,(1)U_{\mathrm{dru}} = \frac{4 \times3\sqrt{2}}{\pi} U_0 \frac{S}{6} - \frac{12}{\pi} X_T I_{\mathrm{dc}}, \tag*{(1)}

where Udru=525U_{\mathrm{dru}} = 525 kV, U0U_0 is the collector line-to-line voltage in kV, SS the bridge-transformer secondary voltage in kV, IdcI_{\mathrm{dc}} the DC current, and XTX_T the commutation reactance per bridge, which aggregates all series reactances in the AC current path between the 66 kV collector bus and the diode valves; the ±525\pm525 kV DC cables rice DC current only and plays no role in commutation. Two properties of (1) shape this work: it contains the voltage magnitude U0U_0 but no angle, and its second term reflects commutation overlap. The leakage reactance XTX_T prevents the current from transferring instantly from one diode to the next, so two diodes conduct together for a short interval, known as the overlap angle. As a result, the AC current lags the voltage, and the DRU consumes reactive power that the diode bridge cannot supply and that must come from the turbine converters or from compensation on the platform.

The bridge-transformer ratio is 66/102 kV, chosen so that the DRU transmits the 1 GW design power of the first study stage, that is 500 MW per pole (Idc≈952I_{\mathrm{dc}} \approx952 A), at rated collector voltage. With this ratio, the ideal no-load DC voltage of the pole, given by the first term of (1) at U0=66U_0 = 66 kV, is Ud0=551U_{\mathrm{d0}} = 551 kV; in operation the commutation drop brings the DRU output down to the 525 kV held by the onshore converter. The reactive power consumed by the commutation overlap is [12]

QDRU=P(Ud0Udru)2−1=0.319P,(2)Q_{\mathrm{DRU}} = P\sqrt{\left(\frac{U_{\mathrm{d0}}}{U_{\mathrm{dru}}}\right)^2 - 1} = 0.319P, \tag*{(2)}

about 319 MVAr at 1 GW.

Methodology

The study followed an isolate-and-solve approach. Replacing the offshore MMC with a DRU changes several parts of the system at once: the turbine GSCs must form the offshore voltage and frequency, the active-power control path moves from the converter angle to the voltage magnitude, the DC-link duty inside the turbine moves to the MSC, and the full commutation reactive demand falls on the turbine converters. Because these changes interact, the study was developed in stages: each stage isolates one problem and solves it before the next is added, until the whole system works (Figure 4). After the offshore MMCs of the reference model were replaced by DRUs, the GFM strategy was selected by testing on the GSC the VSM and the P-V/Q-ω\omega control described in § 3. The controller was then applied to all four turbines, after which a single GFM turbine, represented by its GSC with an ideal DC source and enlarged from 500 MW to 1 GW (§ 4.2), was tuned and tested from 800 to 1000 MW. The MSC was added next to take over the DC-link duty, followed by the grid-following units, first one (1.5 GW) and then two (2 GW), and finally the 2 GW case was run with the MSC active. Each stage, and the choice it led to, is reported in § 4.

Staged methodology flowchart

Figure 4. Staged methodology: each stage isolates one problem before the next is added.

500 MW to 1 GW (§ 4.2), was tuned and tested from 800 to 1000 MW. The MSC was added next to take over the DC-link duty, followed by the grid-following units, first one (1.5 GW) and then two (2 GW), and finally the 2 GW case was run with the MSC active. Each stage, and the choice it led to, is reported in § 4.

Grid-forming control design

Since the wind turbines must operate grid-forming, the first question is which GFM control can regulate the power delivered through the DRU, and which cannot.

VSM angle control and the DRU power path

GFM control is most commonly realised by emulating a synchronous machine, and the VSM is one of its most established forms; it has also been applied to DRU-connected plants [5, 13]. It is therefore the natural first candidate and is examined first.

A VSM forms voltage and frequency through the swing equation:

2Hdωdt=P∗−Pe−D(ω−ω0),dθdt=ω,(3)2H\frac{\mathrm{d}\omega}{\mathrm{d}t}=P^{*}-P_{\mathrm{e}}-D(\omega-\omega_0), \qquad\frac{\mathrm{d}\theta}{\mathrm{d}t}=\omega, \tag*{(3)}

so the active-power error acts on the converter angle. Since the delivered power, Pe=UdruIdcP_{\mathrm{e}}=U_{\mathrm{dru}}I_{\mathrm{dc}} with IdcI_{\mathrm{dc}} from (1), contains no angle, linearising it gives

ΔPe=∂Pe∂U0⏟κDRU>0ΔU0+∂Pe∂θ⏟=0Δθ.(4)\Delta P_{\mathrm{e}}=\underbrace{\frac{\partial P_{\mathrm{e}}}{\partial U_0}}_{\kappa_{\mathrm{DRU}}>0}\Delta U_0+\underbrace{\frac{\partial P_{\mathrm{e}}}{\partial\theta}}_{=0}\Delta\theta. \tag*{(4)}

(4) shows that the delivered power is not controllable through the angle: the gain from the VSM output θ\theta to the controlled output PeP_{\mathrm{e}} is zero, so the active-power loop of (3) remains open; see Figure 5. Angle-based loops may still share power among parallel converters, but the total power must be controlled through the voltage magnitude.

Figure 5. VSM synchronising loop: the DRU provides no angle-to-total-power gain, so the active-power loop remains open.

Adopted P-V/Q-ω\omega architecture

With UdruU_{\mathrm{dru}} held by the onshore converter, (1) shows that the DC current, and hence the power delivered per pole, P=UdruIdcP=U_{\mathrm{dru}}I_{\mathrm{dc}}, rises with the collector voltage magnitude U0U_0 [12]. Active power must therefore be regulated by the voltage magnitude, thereby forming a P-V loop. The angle then carries no active-power information, so the frequency needs a different reference, and the reactive power provides it. For the offshore capacitance CC, formed by the GSC filter and cable capacitance, the q-axis equation in a frame rotating at ω\omega is C dvq/dt=iq−iq,dru−ωCvdC\,\mathrm{d}v_q/\mathrm{d}t=i_q-i_{q,\mathrm{dru}}-\omega C v_d. In a frame aligned with the PCC voltage, vq=0v_q=0 and dvq/dt=0\mathrm{d}v_q/\mathrm{d}t=0, so this equation becomes algebraic [10],

ωCVd=iq−iq,dru,(5)\omega C V_d=i_q-i_{q,\mathrm{dru}}, \tag*{(5)}

where iqi_q is the q-axis current injected by the turbines and iq,drui_{q,\mathrm{dru}} is the q-axis current drawn towards the DRU. The frequency is therefore fixed by the reactive-current balance and the voltage by the active current [10]. Researchers in [12] reach the same assignment from the sensitivities of the DRU equations, ∂P/∂U>0\partial P/\partial U > 0 and ∂Q/∂f=−(Qdru+Qfil)/f\partial Q/\partial f = -(Q_{\mathrm{dru}} + Q_{\mathrm{fil}})/f, where QfilQ_{\mathrm{fil}} is the reactive power of the shunt capacitance, which gives the P-V and Q-ω\omega loops.

The P-V/Q-ω\omega control of [12] has exactly this structure and is adopted because it was designed for Type-4 GSCs on a DRU-HVDC link and provides the complete cascade. Here it is applied without AC filters, so the converter supplies the full DRU reactive demand. Its start-up and fault-ride-through functions are not used. Active power acts on the voltage magnitude,

Eref=E0+(Kp,PV+Ki,PVs)(P∗−P),(6)E_{\mathrm{ref}} = E_0 + \left(K_{\mathrm{p,PV}} + \frac{K_{\mathrm{i,PV}}}{s}\right)(P^* - P), \tag*{(6)}

where ErefE_{\mathrm{ref}} is the d-axis voltage reference passed to the GSC voltage loop (the q-axis reference is zero), E0=1 puE_0 = 1\ \mathrm{pu} its nominal value, P∗P^* and PP the reference and measured active power of WT1, and Kp,PVK_{\mathrm{p,PV}}, Ki,PVK_{\mathrm{i,PV}} the PI gains. This is the closed path identified in (4): raising the GSC voltage raises the PCC voltage, the DC current and the transmitted power. The P-V loop settles at whatever voltage the DRU needs to transmit P∗P^*; the 66/102 kV ratio of § 1 places this voltage at 1.0 pu for the 1 GW design point, so that the converter keeps its voltage headroom. The frequency is generated from the reactive-power error,

ω=ω0−KG1+sKT(Q−Qref),θ=∫ω dt,(7)\omega= \omega_0 - \frac{K_G}{1 + sK_T}(Q - Q_{\mathrm{ref}}), \qquad\theta= \int\omega\,\mathrm{d}t, \tag*{(7)}

so the GSC is its own frequency source, without a phase-locked loop, and the Q-ω\omega loop also aligns the dqdq frame. Here ω0\omega_0 is the nominal angular frequency, QQ the reactive power delivered by WT1, and the frequency falls when it exceeds QrefQ_{\mathrm{ref}}; this sign differs from the form written in [12]. The reference is set slightly below the commutation demand of (2), at Qref=Q∗=0.30PQ_{\mathrm{ref}} = Q^* = 0.30P.

The complete cascade, Figure 6, comprises a dqdq current loop with cross-coupling feedforward, a voltage loop with virtual impedance for LC-filter damping [15], the P-V loop, and the Q-ω\omega loop. Zhang et al. [12] adapt the virtual impedance during low-voltage ride-through to limit fault current; since fault ride-through is outside the scope of this work, a fixed virtual impedance is used.

GSC GFM control structure

Figure 6. GSC GFM control structure, reproduced from [12]. The P-V loop corresponds to (6); the ΔU\Delta U and Δω\Delta\omega inputs pertain to start-up and fault-ride-through functions that are not used here, and the virtual impedance is fixed.

Tuning procedure

The cascade of Figure 6 must be tuned before it can be implemented; it is tuned for the 1 GW GFM unit used in the rest of the paper. In this work tuning was the most time-consuming step, and after sequential loop-by-loop tuning proved unsatisfactory, the gains were set by bandwidth separation between the nested loops and pole placement of the inner loops, in three steps (see Table 1). First, the current-loop bandwidth is set to ω0/5\omega_0/5 and the voltage loop is placed a factor of five lower [16, 17]. Second, the current and voltage loops have integrating plants G(s)=1/(sτ)G(s) = 1/(s\tau), with τ\tau the per-unit filter inductance or capacitance, and the gains of the PI controller Kp(1+1/(sTi))K_{\mathrm{p}}(1 + 1/(sT_{\mathrm{i}})), with proportional gain KpK_{\mathrm{p}} and integral time constant TiT_{\mathrm{i}}, are chosen by pole placement [16],

LoopBandwidth or gainsTuning
Currentω0/5=62.8 rad/s\omega_0/5 = 62.8\ \mathrm{rad/s}pole placement, (8)
VoltageωCC/5=12.6 rad/s\omega_{\mathrm{CC}}/5 = 12.6\ \mathrm{rad/s}pole placement, (8)
P-Vset in EMTEMT
Q-ω\omegaKG=0.114K_G = 0.114, KT=0.4 sK_T = 0.4\ \mathrm{s}from [12], EMT

Table 1. Cascade bandwidths and tuning basis

Kp=τωbw,Ti=1/ωbw,(8)K_{\mathrm{p}} = \tau\omega_{\mathrm{bw}}, \qquad T_{\mathrm{i}} = 1/\omega_{\mathrm{bw}}, \tag*{(8)}

which places the closed-loop poles at the natural frequency ωbw\omega_{\mathrm{bw}} with a damping ratio of 0.5. With the filter values of Figure 2, Lf=0.106 mHL_{\mathrm{f}} = 0.106\ \mathrm{mH} and Cf=2.65 mFC_{\mathrm{f}} = 2.65\ \mathrm{mF} per 5 MVA, 600 V unit, τ=0.463\tau= 0.463 and 0.060 pu on the turbine base, giving Kp=29.09K_{\mathrm{p}} = 29.09 and 0.754 with Ti=0.016T_{\mathrm{i}} = 0.016 and 0.0796 s for the current and voltage loops. Third, because the outer loops act on the nonlinear DRU relation, the P-V gains are set in EMT simulation. For the Q-ω\omega loop, the gain KG=0.114K_G = 0.114 was initialised from the reactive-power coefficient of [12] and, like the time constant KT=0.4K_T = 0.4 s, retained after trial in EMT simulation. All outer-loop gains are verified over the 800 to 1000 MW operating range.

Results and discussion

Control strategy and configuration

Implemented on the GSC, the VSM could not regulate the delivered power, as predicted by (4), so the P-V/Q-ω\omega control was used in all further stages. With the P-V/Q-ω\omega controller applied to all four aggregated 500 MW turbines, each converter integrated its own angle from its own Q-ω\omega loop, and the competing integrators produced persistent oscillations in power and frequency. Parallel GFM operation in DRU-connected plants requires additional coordination, such as secondary control, consensus or adaptive reactive allocation [2, 8, 13, 18]. Here, the problem was isolated instead: a single turbine, WT1, was made GFM and represented by its GSC with an ideal DC source, while the other turbines were disconnected and later added as GFL units (§ 4.5), consistent with [9].

Rating of the grid-forming unit

WT1 is rated 1 GW (200 units of 5 MVA), replacing two of the original 500 MW turbines. Until the grid-following units are connected, the GFM turbine carries the DRU alone, so its rating sets the margin above the conduction threshold of the DRU. The onshore converter holds the HVDC cable at Udru=525U_{\mathrm{dru}} = 525 kV regardless of the offshore power, so the DRU diodes conduct only when the rectified collector voltage exceeds the cable voltage [10, 12]. Since Ud0=551U0U_{\mathrm{d0}} = 551U_0 kV with U0U_0 in pu (Section 2), this conduction threshold is U0=525/551≈0.953U_0 = 525/551 \approx0.953 pu. Above it, rearranging (1), the collector voltage U0(P)U_0(P), in pu, needed to transmit a total power PP over both poles is

U0(P)=Udru+ΔUc(P)551 kV,ΔUc(P)=ΔUc,1GWP1 GW,(9)U_0(P) = \frac{U_{\mathrm{dru}} + \Delta U_{\mathrm{c}}(P)}{551\ \mathrm{kV}}, \qquad\Delta U_{\mathrm{c}}(P) = \Delta U_{\mathrm{c},1\mathrm{GW}}\frac{P}{1\ \mathrm{GW}}, \tag*{(9)}

where ΔUc(P)=12XTIdc/π\Delta U_{\mathrm{c}}(P) = 12X_T I_{\mathrm{dc}}/\pi is the commutation drop. Since it is proportional to IdcI_{\mathrm{dc}} and hence to PP, it can be scaled from its value ΔUc,1GW\Delta U_{\mathrm{c},1\mathrm{GW}} at 1 GW. At 1 GW the PCC voltage is about 1.0 pu (Section 5.3), so ΔUc,1GW≈551−525=26\Delta U_{\mathrm{c},1\mathrm{GW}} \approx551 - 525 = 26 kV. The same relation predicts U0≈(525+52)/551=1.047U_0 \approx(525 + 52)/551 = 1.047 pu at 2 GW, close to the 1.04 pu observed in § 4.5. The delivered power therefore changes by about KDRU=551/26≈21K_{\mathrm{DRU}} = 551/26 \approx21 pu per pu of collector voltage.

From (9), a 500 MW unit would operate at U0=(525+13)/551≈0.976U_0 = (525 + 13)/551 \approx0.976 pu, only 0.024 pu above the threshold, against 0.047 pu at 1 GW. With a plant this steep, a transient voltage dip of that size stops conduction altogether. Consistent with this, a 500 MW case, with the gains of § 3.3 tuned around 1 GW, showed a sustained power oscillation of about 1.4 Hz in which the PCC voltage reached the threshold in every cycle, whereas at 800 to 1000 MW the corresponding transient decays. In addition, as the only GFM unit, a 500 MW WT1 would give a GFM share of 25% at 2 GW, below the critical share of about 37% reported in [9], against 50% at 1 GW. The minimum rating of the GFM unit was not determined because the aim of the study is to establish whether the DRU can carry the full 2 GW; the aggregated unit was instead enlarged to 1 GW, which doubles the margin above the threshold. The resulting configuration is shown in Figure 7.

Single grid-forming turbine validation

WT1 alone forms the offshore network at 800, 900 and 1000 MW; the steady states are listed in Table 2.

PP (MW)QQ (MVAr)VpccV_{\mathrm{pcc}} (pu)ff (Hz)Q/PQ/P
800220.0≈1.0\approx1.050.0200.275
900259.5≈1.0\approx1.049.8960.288
1000300.6≈1.0\approx1.049.7550.301

Table 2. Single GFM turbine steady state (Q∗=0.30PQ^{*} = 0.30P).

The commanded power is delivered at all setpoints with the PCC voltage within 1 % of 1.0 pu, confirming that the 66/102 kV ratio places the natural operating point at the design target, and the frequency remains within 0.25 Hz of 50 Hz. The measured 300.6 MVAr at 1000 MW is within 6 % of the commutation estimate (2), and the rise of Q/PQ/P from 0.275 to 0.301 is consistent with the overlap angle growing with DC current. The reactive surplus Q−QrefQ - Q_{\mathrm{ref}} grows from −20 MVAr at 800 MW to about zero at 1000 MW, while the frequency falls from 50.02 to 49.76 Hz. The deviation does not follow the mismatch in proportion, so in the single-turbine cases the frequency is not set by the Q-ω droop alone, and the reference is an operating-point choice rather than a universal Q/PQ/P constant.

Machine-side converter integration

Replacing the ideal DC source by the MSC of WT1 raises the question of which converter regulates the DC link. Its voltage obeys Cdc dVdc/dt=PMSC−PGSCC_{\mathrm{dc}}\,dV_{\mathrm{dc}}/dt = P_{\mathrm{MSC}} - P_{\mathrm{GSC}}, so one of the two converters must keep the two powers balanced. In a grid-following turbine the roles follow from the hardware. The MSC is connected directly to the generator and is the only converter that can set its torque, so it controls the rotor speed for maximum-power-point tracking, while the GSC regulates VdcV_{\mathrm{dc}} by exporting whatever power the MSC delivers into a grid formed elsewhere. In GFM operation the GSC outer loops are occupied by the P-V/Q-ω control, which itself sets the exported power, so the GSC can no longer balance the DC link and the duty must pass to the MSC, the reverse of the conventional assignment. This is a conflict rather than a simple swap. Because the MSC acts through the generator torque, the rotor and drivetrain now sit inside the DC-link loop: every change in the power drawn by the GSC must be met by a change in generator torque, and the MSC can no longer track the optimum rotor speed independently, so maximum-power-point tracking would have to enter through the power reference P∗P^{*} of the GSC instead.

With the DC-link duty assigned to the MSC, the GSC is deblocked first while an auxiliary DC source holds the link at 1.6 kV. At about 2.3 s the source is disconnected and the MSC is deblocked, so its PI loop takes over the link in a single step. The active power then drops almost to zero, Figure 8, and the DC-link voltage to about 0.5 kV. With the MSC DC-link loop deliberately made slow, the system recovers only gradually: the power returns to its setpoint after about 8 s, with an overshoot to about 1100 MW near 10 s, and the DC-link voltage settles back at 1.6 kV only after about 20 s. A faster loop, with an integral time of the order of 0.02 s, excited a growing oscillation in VdcV_{\mathrm{dc}} within a few seconds of deblocking, ending in integrator windup and collapse, consistent with DC-link/drivetrain interaction reported for GFM PMSG turbines [19].

Active power of WT1 at the PCC during start-up with the MSC active

Figure 8. Active power of WT1 at the PCC during start-up with the MSC active. The auxiliary DC source is disconnected and the MSC deblocked at about 2.3 s.

Hybrid GFM-GFL extension to 2 GW

The grid-following units are then added. All hybrid cases use the transformer ratio and controller gains of the 1 GW design, and in each WT1 first forms the network and settles alone. With one grid-following unit (1.5 GW), WT2 is then deblocked at 500 MW and 150 MVAr; with two (1.6 to 2 GW), WT2 and WT3 are deblocked together at the same setpoints. Table 3 summarises the configurations. The first three reached and held their target power in the EMT runs, as confirmed by measurements through the DRU, with the frequency settling to a bounded deviation; Figure 9 shows the full 2 GW case. With WT1’s MSC active (§ 4.4) and WT1 settled before WT2 and WT3 are deblocked, the 2 GW case again reaches 1000/500/500 MW, a PCC voltage of about 1.04 pu and a frequency of about 48.2 Hz, the same as with the ideal source. The 37.5% GFM-share case exhibited a sustained oscillation as seen in Figure 10. This is consistent with the reduced-GFM-share limitation reported in [9], but the present study does not establish a minimum GFM share, since reduced GFM capacity and operating-point-dependent tuning were not separated.

WT1 (GFM)TotalGFM (%)ff (Hz)Outcome
1000 MW1.5 GW67≈\approx49.0stable
1000 MW2.0 GW50≈\approx48.2stable
800 MW1.8 GW44≈\approx48.7stable
600 MW1.6 GW37.5n/aunsettled

Table 3. Hybrid configurations (WT2, WT3: GFL, 500 MW).

EMT response plots for the 2 GW hybrid configuration

Figure 9. EMT response of the 2 GW hybrid configuration: WT1 GFM at 1000 MW, WT2 and WT3 GFL at 500 MW each (WT2 and WT3 curves coincide).

Reduced GFM share response plots

Figure 10. Reduced GFM share: WT1 GFM at 600 MW with WT2 and WT3 GFL at 500 MW each (1.6 GW); the system does not settle.

The central finding of the extension is the coupling between GFL injection, the GFM reactive operating point, and the offshore frequency. WT1’s reactive output rises from about 300 MVAr alone to 425 to 430 MVAr with WT2 and about 550 MVAr with WT2 and WT3. Each GFL unit’s injection raises the power through the DRU and, with it, the commutation reactive demand. At 2 GW, (9) gives U0≈1.047U_0 \approx1.047 pu, so Ud0≈577U_{d0} \approx577 kV and (2) gives QDRU≈0.46×2000≈910Q_{\mathrm{DRU}} \approx0.46 \times2000 \approx910 MVAr, close to the 850 MVAr supplied by WT1, WT2 and WT3 together. WT1’s reference, however, remains at the 1 GW design value, and with the sign of (7) the growing surplus lowers the frequency to about 49.0 Hz with one GFL unit and 48.2 Hz with two. The DC link isolates the offshore frequency from the onshore grid, so this violates no synchronisation requirement; it is nonetheless undesirable, as it shifts the cascade bandwidths and may excite network-filter resonances.

Discussion and limitations

Unlike the current-source turbines of [10], whose injected reactive current sets the frequency through (5), a GFM GSC imposes the frequency and the network draws the reactive power it needs. Since ∂Q/∂f\partial Q/\partial f is small, about 6 MVAr per Hz at 1 GW, the QQ-ω\omega loop appears to set the frequency from the reactive mismatch rather than regulate QQ; its role is to fix the frequency and dqdq frame and, with several GFM units, to share reactive power through their angles [12].

The DRU reactive demand must be supplied offshore. At 2 GW, WT1 delivers 1000 MW and about 550 MVAr, an apparent power of about 1140 MVA against its 1000 MVA nameplate; already at 1 GW standalone it reaches about 1045 MVA. This is within the converter current limit but beyond its continuous rating. The burden can be shared with the GFL units, which here supply 150 MVAr each out of about 250 MVAr of headroom at 560 MVA, or carried by platform compensation such as AC filters or a STATCOM; [9] shows that the minimum GFM share is set by this current loading and falls when filters carry part of the demand. Because the QQ-ω\omega loop of a GFM unit cannot move its own reactive output, removing

the frequency deviation requires changing QrefQ_{\mathrm{ref}} or supplying the demand elsewhere: for example, an outer loop that adjusts QrefQ_{\mathrm{ref}} until the frequency returns to 50 Hz, as in the central PI controller of [10], of which the fixed Q∗=0.30PQ^* = 0.30P used here is only the feedforward part, or secondary frequency restoration [2, 18], adaptive reactive-power allocation [8] and GFM operating-point compensation [9].

The 2 GW result rests on a single 1000 MW aggregated GFM converter. The bridge-transformer ratio was chosen at the 1 GW design point; the commutation overlap and transformer loading at 2 GW, where the DC current doubles, the DC overvoltage towards Ud0U_{d0} after a loss of the onshore converter, the intra-group interaction of individually controlled turbines and the margin of ErefE_{\mathrm{ref}} to its limits were not examined.

Conclusion

DRU-HVDC removes the conventional source of offshore voltage and frequency regulation, so grid formation must be provided by the wind turbines. Because the DRU commutation relationship contains the offshore voltage magnitude but no converter angle, the delivered power has no control path through the angle. A VSM implemented on the GSC could not regulate the power, and the P-V/Q-ω\omega architecture of [12], which regulates active power through the voltage magnitude, was adopted instead. Implemented on a 1 GW aggregated GSC with a 66/102 kV bridge-transformer ratio chosen for the 1 GW design point, it delivered 800 to 1000 MW with the PCC voltage near 1.0 pu. With the MSC included, DC-link regulation moves from the GSC to the MSC, and raising the bandwidth of this loop led to growing oscillation, so it had to be kept slow; as a result, the turbine needs about 20 s to recover after the DC-link handover. Extending the system to 2 GW with two grid-following units achieved full power transfer in the model, but the active power of the grid-following units raised the reactive power WT1 must supply from about 300 to 550 MVAr, beyond its continuous rating, and with the fixed reactive reference the offshore frequency settled at about 48.2 Hz. These results indicate that grid-forming operation on a DRU affects the whole turbine and the plant-level reactive balance, not only the GSC control. Future work will address the MSC-GSC allocation problem, the minimum stable GFM share, multi-GFM coordination, reactive-power sharing and frequency restoration, and black start.

Acknowledgements

This work was carried out in collaboration with TenneT TSO B.V. within the framework of an MSc thesis project at Delft University of Technology. The work is also supported by Horizon Europe project PROSECCO, under grant agreement 101160687.

References

  1. [1]A. Korompili, Q. Wu, and H. Zhao. Review of VSC HVDC connection for offshore wind power integration. Renewable and Sustainable Energy Reviews, 59:1405–1414, 2016.DOI
  2. [2]R. Blasco-Gimenez, S. Año-Villalba, J. Rodríguez-D’Derlée, F. Morant, and S. Bernal-Perez. Distributed voltage and frequency control of offshore wind farms connected with a diode-based HVdc link. IEEE Transactions on Power Electronics, 25(12):3095–3105, 2010.
  3. [3]L. Yu, Z. Fu, R. Li, and J. Zhu. DRU-HVDC for offshore wind power transmission: a review. IET Renewable Power Generation, 18(13):2080–2101, 2024.DOI
  4. [4]L. Yu, R. Li, and L. Xu. Distributed PLL-based control of offshore wind turbines connected with diode-rectifier-based HVDC systems. IEEE Transactions on Power Delivery, 33(3):1328–1336, 2018.DOI
  5. [5]A. Bidadfar, O. Saborío-Romano, N. A. Cutululis, and P. E. Sørensen. Control of offshore wind turbines connected to diode-rectifier-based HVDC systems. IEEE Transactions on Sustainable Energy, 12(1):514–523, 2021.DOI
  6. [6]O. Karaca, I. Tsoumas, M. Schweizer, O. Stanojev, and L. Harnefors. Robust black start of an offshore wind farm with DRU based HVDC link using power synchronization control. In 2025 Energy Conversion Congress & Expo Europe (ECCE Europe), pages 1–6, 2025.arxiv.org/abs/2506.17060
  7. [7]K. Wang, Q. Song, B. Zhao, Z. Yu, and R. Zeng. Grid-forming control of offshore wind farms connected with diode-based HVdc links based on remote active power regulation. IEEE Transactions on Sustainable Energy, 15(2):1315–1327, 2024.DOI
  8. [8]G. Zhang, W. Xiang, X. Chen, R. Tu, X. Qiao, and J. Wen. Grid-forming control based on adaptive reactive power allocation for offshore wind farms connected to diode-rectifier-based HVDC system. Journal of Modern Power Systems and Clean Energy, 13(1):154–166, 2025.DOI
  9. [9]Y. Jin, Z. Zhang, and Z. Xu. Proportion of grid-forming wind turbines in hybrid GFM-GFL offshore wind farms integrated with diode rectifier unit based HVDC system. Journal of Modern Power Systems and Clean Energy, 13(1):87–101, 2025.DOI
  10. [10]D. Herrera, E. Galván, and J. M. Carrasco. Method for controlling voltage and frequency of the local offshore grid responsible for connecting large offshore commercial wind turbines with the rectifier diode-based HVDC-link applied to an external controller. IET Electric Power Applications, 11(9):1509–1516, 2017.DOI
  11. [11]H. Xiao, X. Huang, Y. Huang, and Y. Liu. Self-synchronizing control and frequency response of offshore wind farms connected to diode rectifier based HVDC system. IEEE Transactions on Sustainable Energy, 13(3):1681–1692, 2022.DOI
  12. [12]Y. Zhang, W. Zhu, C. Tang, N. Liu, S. Li, and H. Wang. Start-up and fault-ride-through strategy for offshore wind power via DRU-HVDC transmission system. Energies, 17(19):4968, 2024.DOI
  13. [13]A. Beneit-Barajas, P. Penades-Huesca, and J. Martínez-Turégano. Analysis of non-uniform grid-forming control techniques for the HVDC connection of renewable energy. Mathematics and Computers in Simulation, 229:1–14, 2025.
  14. [14]J. Arrillaga. High Voltage Direct Current Transmission, volume 29 of IEE Power and Energy Series. The Institution of Electrical Engineers, London, 2nd edition, 1998.DOI
  15. [15]P. A. Dahono, Y. R. Bahar, Y. Sato, and T. Kataoka. Damping of transient oscillations on the output LC filter of PWM inverters by using a virtual resistor. In Proceedings of the 4th IEEE International Conference on Power Electronics and Drive Systems (PEDS), volume 1, pages 403–407, 2001.
  16. [16]A. Yazdani and R. Iravani. Voltage-Sourced Converters in Power Systems: Modeling, Control, and Applications. John Wiley & Sons, 2010.
  17. [17]R. Teodorescu, M. Liserre, and P. Rodríguez. Grid Converters for Photovoltaic and Wind Power Systems. Wiley-IEEE Press, 2011.DOI
  18. [18]Y. Sun, D. Kong, Y. Wang, Z. Zhang, Z. Pei, D. Zhu, J. Rodriguez, W. Tian, and M. Heldwein. Decoupled power and frequency control of offshore wind farms connected with DR-HVDC based on consensus algorithm. In Proceedings of the IEEE 10th International Power Electronics and Motion Control Conference (IPEMC-ECCE Asia), pages 1634–1639, 2024.DOI
  19. [19]S. Liu, H. Wu, T. Bosma, and X. Wang. Impact of DC-link voltage control on torsional vibrations in grid-forming PMSG wind turbines. IEEE Transactions on Energy Conversion, 39(4):2631–2642, 2024.DOI

Paper details

vlnc:W_ch5z9hzq
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 4

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