Overview of Frequency-Control Technologies for a VSC-HVDC-Integrated Wind Farm

In modern power systems, the increasing penetration of renewable energy resources has reduced the overall system inertia. However, the intermittent nature of wind power generation reduces frequency stability, which is a crucial issue. Consequently, modern power systems require that renewable energy sources, such as offshore wind farms, support the frequency regulation. High voltage direct current (HVDC) systems based on voltage source converters (VSCs) are currently being used to connect wind farms with the main grid. By adjusting the DC-link voltage of the VSC-HVDC, a capacitor can absorb or release energy to provide frequency support. In addition, the VSC-HVDC system can coordinate wind farms to support frequency regulation and thereby achieve better performance. Thus, this paper reviews and compares various frequency-control strategies for a VSC-HVDC-connected wind farm. This work implements in PSCAD/EMTDC a simulation of typical frequency-control techniques to verify their effectiveness and robustness. Furthermore, this paper shows the advantages and drawbacks of each frequency regulation method. Other auxiliary services for supporting frequency regulation, such as by energy storage systems, are also discussed. Finally, this paper provides complete recommendations for frequency regulation techniques for VSC-HVDC-integrated wind farms.

Equaling to |V o | V gen /X k g The gain related to the number of poles and the moment of inertia k d The gain related to the droop constant k o The output gain of the GEC controller P 0 , Q 0 Initial active power and initial reactive power E 0 No load electromotive force D p Damping coefficient D q Reactive power coefficient K U Voltage coefficient for VSG

I. INTRODUCTION
Modern power systems are facing a succession of challenges of frequency stability, owing to the increasing penetration of wind energy [1]. Unlike conventional synchronous generators (SGs), wind turbines (WTs) cannot naturally provide frequency support to a power system. This is because the output power of wind farms (WFs) is decoupled from the system frequency, owing to the AC/DC/AC converters [2]. If load shedding or an SG outage occurs in a power system, then the power system frequency will deviate from its nominal value because of the power imbalance between the supply and demand. A large overall system inertia can help to stabilize the frequency. Therefore, there have been many studies to increase inertia from renewable energy. The frequency support techniques for WTs have been addressed in [3]- [7], which include inertial control, static/dynamic droop control, stepwise inertial control, fast power reserve and so on. The authors in [3] analyzed the energy stored in the rotating mass of WTs. They compared two different control methods based on changing the reference value of WT's torque for providing inertia response. In [4], the control of the electrical torque and the output power of a WT were studied and compared. The controller was based on the time derivative of the frequency, so as to provide an inertia response.
Voltage source converter based High Voltage Direct Current (VSC-HVDC) transmission systems are seen as a solution to connect offshore wind farms (OWFs) with onshore grids. Although the costs of the converters are high, a longer transmission distance will lead to a lower overall investment cost [8]. As a result, the number of cases for VSC-HVDCconnected OWFs has been increasing in recent years. However, similar to other conventional converters, the converters of the VSC-HVDC systems also decouple the available wind generation from the system frequency. The decoupled characteristics of the grid frequency f , the DC-link voltage V DC and the WF's side frequency f WF are depicted in Fig. 1 where V DC and f WF remain constant irrespective of the f value. Energy storage systems (ESSs) play a prominent role in the power systems. The numerous benefits of ESSs for wind power applications were summarized in [9]. For instance, fault ride through, power fluctuation mitigation, and spinning reserves are common services provided by ESSs. An ESS can participate in frequency regulation (FR) control and can collaborate with the WF to improve the overall system inertia [10]- [17]. Furthermore, for a further specific application in FR, the ESS can coordinate with the WF to prevent a second frequency drop (SFD) during the recovery of generator rotor speed of WTs [17].
Some review papers have organized several frequency control techniques for WTs [17]- [22]. The authors in [23] discussed the limitations of frequency support techniques for WTs and advised that the maximum power point tracking (MPPT) algorithm, minimum operating rotor speed, and nonlinearities and optimization of the frequency support be further investigated. Reference [24] reviewed the design of multiple controllers for primary and secondary frequency control for WTs.
A DC-based wind farm (WF) is viewed as a prominent and cost-effective topology for the commercial wind industry. As wind turbines (WTs) are connected to a single DC-bus, the number of DC/AC inverters and transformers can be greatly reduced. The grid connection of a DC-based WF should be worthy of attention. The influences of DC-based WFs on the AC grid during a dynamic situation and fault ride-through characteristic were investigated in [25]. Furthermore, a resonance circuit between HVDC and DC-based WFs may cause a severe oscillation when AC voltage variations or power imbalances occur in AC grids. The authors in [26] analyzed the impacts of different DC line models on the resonance phenomenon in a DC-based WF.
Several studies discussed the frequency control strategies for all-DC WFs. The authors in [27] investigated active and reactive power controls for a medium voltage DC (MVDC)-connected WF. The proposed topology contains ultra-capacitors and battery storage systems for supporting AC voltage control and frequency droop control. Reference [28] proposed a communication-free approach for all-DC OWFs for providing frequency support. The communication-free approach can regulate the voltage at onshore or offshore converters according to frequency deviation. Therefore, the OWF can participate in the frequency regulation scheme by monitoring the DC voltage at the offshore converter. Furthermore, the time delay due to the communication channel can be avoided by using ancillary voltage controllers. The authors in [29] proposed a coordinated frequency control strategy called inertia synchronizing control for all-DC WFs. The all-DC WF is connected to the onshore converter through MVDC and HVDC lines. The converter at the WF side can automatically synchronize with AC grid without phase-locked-loop (PLL). The DC-bus voltage is firstly controlled by the grid-side converter. Afterwards, the DC capacitors and the all-DC WF can provide frequency support according to the voltage variation at the DC bus. Particularly, the control method can be applied to weak grids. In [30], a DC collection was adopted for collecting the output power of WTs, and these WTs can provide voltage and frequency supports based on the DC-bus voltage. The grid-side converter was designed as a virtual synchronous generator to support inertia. By contrast, the WF-side converters deploy the power-voltage droop characteristics to distribute power flows. In summary, these aforementioned frequency control strategies for all-DC WFs rely mainly on the DC-bus voltage as a medium for transmitting the disturbance of grid frequency to a WF, which enables WTs to participate in frequency regulation.
Among articles giving overviews of FR techniques for the VSC-HVDC, reference [31] introduced some coordinated FR strategies by means of discussion, but does not include the control diagrams of these methods. Reference [32] summarized the inertial control, droop control, proportional integral (PI)-based control, and their modified controllers for FR in a VSC-HVDC system. The authors in [33] proposed a classification of inertia control strategies for the VSC-HVDC based on the type of synchronization with the grid and the coordination between other generation sources: synchronized with and without phase-locked-loop (PLL), centralized control, and distributed control. Reference [34] reviewed the numerous existing control strategies, e.g., SG-based, swingequation-based, and droop-based topologies for the inverter to provide virtual inertia support. Nonetheless, these studies did not consider the dynamic behaviors of the WF in the FR structure.
Among these studies, none gave a comprehensive overview of control strategies for coordinating the VSC-HVDC and the WF for participating in an FR scheme. However, how to integrate VSC-HVDC in a WF is a current research trend and cooperation between the VSC-HVDC and WF should be further studied. Hence, this study summarizes existing FR strategies for VSC-HVDC-connected WFs. The contributions of this work are as follows: 1) Analyzes of the pros and cons of each technique and comprehensive comparisons. 2) Detailed designs of the FR controllers for VSC-HVDCintegrated WFs. 3) In particular, several time-domain simulation cases for selected control strategies. 4) Studies of coordinated FR techniques for WFs and ESSs 5) Illustration of the challenges and difficulties of existing FR techniques, providing future solutions and recommendations. The rest of this paper is organized as follows. Section II introduces the basic operation theorems of the WTs and the VSC-HVDC system. Section III describes the various frequency-control strategies for VSC-HVDC-integrated WFs and the design of the controllers. Simulation results and a comprehensive comparison of different FR strategies for a VSC-HVDC-connected WF are presented in Section IV. Next, the FR strategies for WTs and ESSs are introduced in Section V. The future research trends are presented in Section VI. Finally, the conclusions are drawn in Section VII.

II. BASIC THEOREMS OF WIND TURBINES AND VSC-HVDC A. AERODYNAMICS OF WIND TURBINES
The doubly-fed induction generator (DFIG) is selected as the type of WT to be studied. The structure of the DFIG is shown in Fig. 2. The mechanical energy captured from wind, P wind , can be expressed [35] as: where ρ is the air density; R is the radius of the rotor blade; V w is the wind speed and C p is the power coefficient, which is a function of tip speed ratio (TSR), λ, and the pitch angle, β. The definition of λ is: where ω t is the rotational speed of the blade. The active power generated by the DFIG is controlled by the MPPT algorithm. The reference value of active power is set according to wind speed and rotor speed. By using the MPPT algorithm, the DFIG can operate on an optimal power curve for various wind speeds. The relationship between the mechanical power and the electrical power, P WT , of the DFIG is expressed as: where H D is the inertia constant of the DFIG; ω t is the rotational speed of the DFIG and the upper bar represents the per-unit value. Equation (3) is the well-known swing equation.

B. DELOADING METHODS OF WIND TURBINES
By using the MPPT algorithm, the WTs can capture maximum power from the wind to maintain a high level of wind energy harvesting. However, operating at the optimal point indicates that there is no additional power reserve for the WTs to provide frequency support. If the WTs need to release the kinetic energy stored in the rotor back to the grid, then the rotor speed must be reduced and the extra energy must be absorbed by the WTs during rotor speed recovery, which requires another control strategy. To provide reserve power for frequency support, there are two common deloading methods [36]: pitch angle control and the overspeeding technique.
In general, the pitch angle is zero in low wind-speed conditions. When the wind speed exceeds the rated wind speed, the WT will operate at full load. Meanwhile, the rotational speed should be limited to the rated speed to protect the rotor and the equipment; this is done by activating the pitch angle controller. The control diagram of the pitch angle controller is shown in Fig. 3. The mainstream PI controller is utilized for pitch angle control. The reference power, P * WT , is the deloaded signal according to the desired power margin and P WT is the measured WT power. The difference between them is passed through a PI controller, a first-order transfer function and the rate limiter, in order to obtain the pitch angle. The time constant T d is often related to the time delay. The upper limit β max and the lower limit β min are set to be 27 • and −27 • , respectively. The rate of change of the pitch angle is limited within ±10 • /s to prevent a sudden variation. Fig. 4 shows the output power of a WT under different wind speeds. The red curve represents the MPPT curve, e.g., point A is the optimal point at a wind speed of 9 m/s. The operating point of the WT can be shifted to the right or the left of the optimal power point to change the TSR, and thus the deloading operation based on speed control can be achieved, as presented in Fig. 4. The overspeeding technique is preferred because the kinetic energy stored in the rotor mass can be released to the grid, when the WT increases its output power from point B towards point A. It is noted that, in a low wind-speed range, only the overspeeding technique can be applied. In a medium wind-speed range, the pitch angle control and overspeeding technique can be activated simultaneously. Finally, in a high wind-speed range, the pitch angle controller plays a crucial role in limiting the maximum output power of the WT since the overspeeding technique causes the WT to exceed its rating.
The desired power margin, x, decides the deloaded reference power as: where P opt is the optimal power reference. Hence, the target operating points lie on the sub-optimal power curve instead of the MPPT curve. Once the sub-optimal point is obtained, the corresponding power coefficient, C p,del , can be determined as: Hence, for a given pitch angle and wind speed, the rotor speed corresponding to the sub-optimal curve can be obtained. Then the deloaded power can be calculated according to (4) and (5).

C. FREQUENCY REGULATION CAPABILITY OF WIND TURBINES
In general, the process of FR after a sudden imbalance of the power flow can be divided into four different successive steps: inertial response (IR), primary frequency response (PFR), secondary frequency response (SFR) and tertiary frequency response (TFR) [37]- [39]. SFR is usually called as automatic generation control (AGC) [7]. Fig. 5 (a) shows the frequency response for a low-frequency event. The IR is immediately activated when detecting a generator trip-off or a load increase, and it lasts for a few seconds. The purpose of IR is to reduce the rate of change of frequency (ROCOF) and to increase the frequency nadir. The PFR related to droop control owing to the SG governor should start within a few seconds and be fully deployed by 30 s. It is noted that the PFR often contributes to a temporary steady state where the grid system is stabilized at a specific value [40]. Subsequently, the SFR is deployed from 30 s to 15 min to bring the frequency back to the nominal value and to free the PFR. Finally, the system operators manually use the TFR to free the previous reserves [41], [42]. A schematic diagram of the successive power reserves is depicted in Fig. 5 Renewable energy sources such as solar and wind do not possess inertia. However, by utilizing synthetic or virtual inertia techniques, WTs are capable of similar inertia response behaviors as conventional SGs. This study summarizes two common control schemes [43], [44].
According to the method proposed in [3], the aim of inertial control is to mimic the inertial response of a conventional SG. When the system frequency drops, the kinetic energy stored in the rotating mass of the WT is released to the grid. The  inertial control of the WT can be expressed as: where P in is the additional output power of the WT; K in is the gain related to the inertia constant H D , and f is the measured frequency. Based on (6), it can be observed that the output power of the WT is proportional to the derivative of the frequency. The inertial controller is shown in Fig. 6. A low-pass filter is generally used to filter out noise signals and reduce the overshoot. To imitate the response of the speed governor of a conventional SG, droop control is applied [45]. For droop control, the additional output power P f is proportional to the frequency deviation of the measured frequency from the nominal frequency, which is expressed as: where f 0 is the nominal frequency and R is the droop constant. The droop controller is shown in Fig. 7. The typical droop characteristics are depicted in Fig. 8.  As shown in Fig. 8, the slant line with a negative slope represents the frequency versus the output power of a WT. Assuming the original operating point of the WT is (P 0 , f 0 ), when a low-frequency event occurs, the operating point moves toward (P 1 , f 1 ), indicating that the WT should increase its output power to assist the frequency recovery process.
If both inertial control and droop control are applied, then the total additional output power provided by a WT during the frequency event, P WT , can be expressed by combining (6) and (7), as follows:

D. BASIC PRINCIPLES OF THE VSC-HVDC SYSTEM
In brief, the VSC-HVDC system can be viewed as a zeroinertia generator that is capable of controlling the active and reactive powers instantaneously [46]. The structure and control of a typical VSC-HVDC-connected WF system are presented in Fig. 9. The wind-farm side voltage source converter (WFVSC) collects the active power from the WF. It sets the reference frequency, voltage amplitude, and phase angle. Moreover, the power flow between the grid side voltage source converter (GSVSC) and WFVSC can be stabilized by controlling the DC-link voltage. In particular, a constant DC-link voltage indicates a balanced power flow between the grid and WF.
For the purpose of stabilizing the power flow and voltage in the VSC-HVDC system, different control methods at both sides of the converters should be considered. The detailed outer and inner current control loops for the VSC-HVDC system are introduced as follows.
The outer current loop determines the reference currents, which are referred to the synchronous d-q reference frame.  The design of the power controllers is based on the power flowing through the converter [2], as shown by: where v d and v q are the d-axis and q-axis voltages, respectively, and i d and i q are the d-axis and q-axis currents, respectively.
To enable the WFVSC to control the captured active power from the WF, the active power controller can be designed according to (9). It is shown in Fig. 10, where P * WF is the reference active power from the WF and i * d is the reference d-axis current. Generally, the reference values of the d-axis and q-axis currents can be determined by two types of inputs, which depend on the system operators. The common outer current loop is shown in Fig. 11, where V DC is the DC-link voltage, v rms is the AC voltage at the point of common coupling (PCC), and i * q is the reference value of the q-axis current. To derive the inner current loop, the electrical circuit of the VSC-HVDC is analyzed. The equivalent circuit of the VSC-HVDC is shown in Fig. 12, where the left part of the  diagram is the AC side and the right part is the DC side. L and R represent the equivalent inductance and resistance of the reactor and the transformer, respectively. V s is the AC side VOLUME 9, 2021 voltage and V C is the output voltage of the converter [47]. Coupling between the AC and the DC sides is achieved through a controlled voltage source and a controlled current source.
Subsequently, the abc voltage vectors are transformed to dq0 voltage vectors through Park's transformation to control the VSC-HVDC. The d-axis and q-axis equivalent circuit models are depicted in Fig. 13. The dynamic equations describing the equivalent circuits are expressed as: where v cd and v cq are the d-axis and q-axis output voltages of the converter, respectively; v sd and v sq are the d-axis and q-axis grid AC-side voltages, respectively, and ω is the synchronous angular velocity. The grid AC voltage can be further derived from (10): where v * sd and v * sq are the reference values of the d-axis and q-axis grid AC voltages, respectively, and i * sd and i * sq are generated from the outer current loop. Accordingly, by substituting (11) into (10), the inner current loop is derived as: The control diagram of the inner current loop is shown in Fig. 14. The d-axis and q-axis components are decoupled to control the active and reactive powers or the DC and AC voltages separately. Next, the reference values of the output voltages of the converter, i.e., v * cd and v * cq , are transformed into abc voltage vectors through an inverse Park's transformation to generate the reference signals for pulse width modulation (PWM), which produces the signals for controlling the switches. A constant DC-link voltage ensures the power balance between the GSVSC and WFVSC. By contrast, the power flow can be regulated by adjusting the DC-link voltage, V DC . Obviously, since the capacitors in the DC-link are the storage elements, they can be used to release the stored energy or absorb the excess energy by reducing or increasing V DC , respectively. The energy change W C via the equivalent capacitance C and the power change P C in the period of t 1 − t 2 are expressed as: Consequently, when the system frequency falls, the DC capacitors can release the stored energy to the grid. By contrast, when the system frequency increases, the DC capacitors can absorb the extra energy. By regulating the DC voltage, the VSC-HVDC will be able to contribute to FR.

III. FREQUENCY REGULATION STRATEGIES FOR A VSC-HVDC-CONNECTED WF
In fact, the VSC-HVDC and the WTs cannot provide frequency support if there is no additional controller. In order to let the VSC-HVDC and the WTs participate in FR, the synthetic inertia techniques turn out to be an essential component. In recent years, several FR strategies for VSC-HVDC-connected OWFs have been proposed. This study summarizes the main strategies and makes a detailed comparison of them. The classification of these FR strategies is presented in Fig. 15. Two aspects are considered for the classification: 1) whether the grid frequency is communicated to the WF side and 2) the algorithm for the WF to perceive the grid frequency disturbance. A detailed introduction to these strategies is presented in the following sections.

A. CONVENTIONAL SCHEME (COMMUNICATION-BASED APPROACH)
The idea of the conventional scheme is to use a communication link to transmit the grid frequency signal to the WF [48]. This type of method can be categorized as a communicationbased approach. In general, the conventional scheme relies on a supervisory control and data acquisition system or other fast communication links. To enable a VSC-HVDC-connected WF to provide frequency support, the WFVSC should be operated under an AC voltage and active power control, while the GSVSC should keep the DC-link voltage constant. A schematic diagram of the conventional scheme is depicted in Fig. 16. With a reliable and accurate communication channel, the conventional scheme allows WFs to participate in FR. However, the communication of the grid frequency may result in the delay of frequency support from the WF. Hence, the transmission time delay should be reduced as much as possible.

B. SIMULTANEOUS CONTROL
When using the communication-based method, there is a concern regarding the reaction time and the reliability of the dynamics of the WF. For example, the ROCOF and the frequency nadir can be more significant if there is a delay in the WF providing frequency support. Since information transmission involves measurement, processing, communication, and reception, the typical delay values are in the range of 50 ms -1 s [49] [50]. Some researchers proposed an alternative based on DC-link voltage regulation to address the problem of transmission delay [48], [51]- [53].
The principle of simultaneous control is that a system frequency disturbance first changes the DC-link voltage of the VSC-HVDC in case of power imbalance in the main grid. When the DC voltage is changed, the DC capacitors immediately release or absorb the energy. Next, the deviation of DC voltage contributes to a new WFVSC frequency. Thus, the WF will provide frequency support according to the new WFVSC frequency.
The control scheme for simultaneous control is shown in Fig. 17, where V * DC is the reference value of the DC voltage; V DC0 is the initial DC voltage; f is the nominal grid frequency; f * WF is the reference value of the WFVSC frequency and f WF0 is the nominal WFVSC frequency. The equations that describe the simultaneous control are expressed as: where f WF is the deviation of the WFVSC frequency from the nominal value and K A and K B are the multipliers. The larger the values of K A and K B , the larger the deviations of the DC voltage and the WFVSC frequency will be. Thus, the DC capacitors and the WTs will be able to provide more active power to the grid. The WFVSC frequency and the grid frequency are strongly correlated through (17). If the product of K A and K B is equal to one, then the WFVSC frequency will exhibit a similar dynamic response as the grid frequency.
Notably, simultaneous control does not need an additional communication channel to transmit the grid frequency to the WFVSC. In addition, reference [54] analyzed the effects of different time delays and controller parameters for the conventional scheme and simultaneous control and showed that simultaneous control yields more stable dynamics in P, V DC and f during the FR period.

C. INERTIA EMULATION CONTROL FOR A VSC-HVDC SYSTEM
Although the VSC-HVDC and WF can participate in FR by using the two former strategies, the performance is questionable because of the communication latency or the V DC variation process. That is, the derivation of (15), (16), and (17) should be closely analyzed. If the VSC-HVDC can provide frequency support, then its inertia constant should be found. To verify the synthetic inertia of the VSC-HVDC and WF, several researchers have addressed the issue of inertia emulation control (IEC) strategies for the VSC-HVDC [55]- [58]. The concept of IEC is to let the non-synchronous units emulate the behaviors of SG. The authors of [55] proposed a coordinated control strategy for a WF with line-commutatedconverter-based HVDC for providing frequency support. The method introduces a virtual droop constant for the HVDC and thus enhances the overall FR performance. It is noted that the WF does not require the additional frequency controller, and the electrical power of the WF is determined by the delivered power of the HVDC. Hence, the mechanical power and the  pitch angle can be changed to match the electrical power. Reference [56] demonstrates a coordinated control strategy by using the DC capacitor energy and the KE of the WF to improve the overall system inertia. The strategy can provide more inertia support than the single VSC-HVDC or single WF control by assuming the virtual inertial time constants for the VSC-HVDC and WF.
In general, the step-by-step IEC algorithm for the VSC-HVDC is as follows: 1) Capacitor DC voltage versus system frequency.
where E K is the stored kinetic energy at synchronous speed ω 0 and S is the machine rating in MVA [59]. For steps 1) and 2), by mapping the KE of the SG into the electromagnetic energy of the DC-link capacitors, the emulated inertia constant of VSC-HVDC H VSC can be obtained as: S VSC (19) where E e is the electromagnetic energy of the DC-link capacitors, N is the total number of capacitors installed in the DC-link, C is the capacitance of the DC-link and S VSC is the capacity of the VSC converter. The electromagnetic energy stored in the DC-link capacitors is very similar to the KE of the SG because the angular speed of the SG and the nominal DC voltage are squared. The swing equation describing the rotor motion of the SG can be expressed as: where P m is the mechanical power inputted to the machine and P e is the electrical power. Equation (20) can be simplified by approximating f ≈ f 0 and using the per-unit expression as: The target is to equate the exchange power of VSC-HVDC to that of an SG. Hence, equation (21) can be modified by substituting H VSC for H , V DC for f and V DC0 for f 0 , as follows: Equation (22) reveals that the variation of V DC also changes the active power through the VSC-HVDC, as well as the stored energy in the DC-link capacitors.
Next, to investigate the relationship between C, S VSC , H VSC , V DC and f , equating (21) and (22) yields By integrating the left side of (23) from f 0 to f and the right side of (23) from V DC0 to V DC , the corresponding integration and its solution can be expressed as, respectively: Rearranging (25) and Equation (27) Fig. 18. It can be seen from Fig. 18 that a larger H VSC leads to a larger V DC for a given f as a larger H VSC means a larger energy exchanged with the grid when the grid frequency changed. Furthermore, the effect of different values of C with fixed H VSC is investigated in Fig. 19. It is shown that the DC-link voltage changes within a smaller range with a larger capacitance compared to the case with a smaller capacitance. In conclusion, it is a trade-off between a larger H VSC and the cost of larger C for V DC within the acceptable range of ±10%.
As a result, the reference DC-link voltage V * DC can be determined by the grid frequency f . The corresponding schematic diagram of the IEC with the DC-link voltage controller is shown in Fig. 20. In addition, the WF can participate in FR by using (16) to perceive the grid frequency VOLUME 9, 2021 disturbance. In this way, step 4) is completed via the artificial coupling between the grid frequency and the WF frequency.

D. BLENDING CONTROL OF VSC-HVDC AND WF FOR FAST FREQUENCY RESPONSE
To enhance the inertial response and improve the overall FR capability, the authors of [57] and [60] proposed to utilize the energy stored in the DC-link capacitors during the DC voltage variation. However, using the stored energy in the DC-link capacitors to provide a fast inertial response has the serious disadvantage of requiring an extremely large capacitance. Rather than solely relying on the frequency support from the VSC-HVDC or WF, blending the energy stored in the VSC-HVDC and the kinetic energy of the WTs to provide frequency support can yield better performance.
The author of [60] proposed another IEC strategy that can provide a fast frequency response while not requiring a very large capacitance nor excessive active power from the WF. That is, the high-performance WT is not required. The inertia emulation control is developed for the VSC-HVDC. The more detailed process of derivation can be found in [60]; this paper merely shows the key concepts. The step-by-step derivation of the controller is described as follows.
The aim is to obtain an inertia constant for the VSC-HVDC. That is, the VSC-HVDC can provide inertial response when the grid frequency changes. Therefore, the relationship between the active power flowing through the GSVSC and the grid frequency can be expressed in a swing equation as: where H is the emulated inertia constant of the converter, S is the rated capacity of the VSC converter, P in is the input active power of the GSVSC, and P out is the output active power toward the grid. Assuming the grid frequency drops, the reference active power resulting from inertia response P * H can be found from For example, if the emulated inertia constant of the VSC-HVDC is 4 s, which is the same as the typical value of a conventional SG [61], [62], and the rated capacity of the converter is 300 MVA in a 60 Hz system, then the active power P * H is 20 MW when the ROCOF is −0.5 Hz/s. The reference value of the variation of DC-link voltage can be obtained from the reference active power [60] where τ is a designed parameter as well as a time constant relating to the variation of DC-link voltage and the dynamic of the WF and it is defined as a third of the reference settling time of the WF. Considering that the WF needs 0.3 s to adjust its output power to the reference value, for example, i.e., 3τ = 0.3 s, the resulting time constant τ can be obtained as 0.1 s. The reference active power of the WF is determined by a simple DC voltage-active power droop controller where K WF is the gain affecting the amount of output power of the WF. Overall, the active power flowing through the  GSVSC can be expressed as where the former term denotes the active power from the VSC-HVDC and the latter term denotes the contribution from the WF. By introducing (33) into the control scheme and deploying a larger value of τ , the WF is not required to ramp up its output power as quickly as the method introduced in Section III B. However, the larger τ also leads to a larger deviation of the DC-link voltage. It is a trade-off between the requirement of a high-performance WF and the large deviation of the DC-link voltage. By combining (30) - (33), the corresponding control diagram for this inertia emulation control is shown in Fig. 21.

E. MODIFIED VSC CONTROL WITH ANCILLARY FREQUENCY CONTROL
In [46], the GSVSC and WFVSC are equipped with the ancillary frequency controller to adjust the modulation indices for PWM. To detect the frequency disturbance at the grid side, the reference frequency at the WFVSC, f * WF , is made timevarying according to the grid frequency f , by applying an ancillary frequency controller. This is shown in Fig. 22 (a), where V WF is the AC voltage at the PCC of the WFVSC; m a is the amplitude modulation index for PWM and f WF is the frequency at the WFVSC. The grid frequency f is transmitted to the WFVSC through a transmission channel. As a consequence, the OWF can provide frequency support by detecting the frequency at the WFVSC. As for the GSVSC, an ancillary DC voltage controller can be added to regulate the DC-link voltage according to the grid frequency, as shown in Fig. 22 (b), where V * DCf is VOLUME 9, 2021 the additional reference value of DC voltage based on the deviation between the measured frequency f and the nominal frequency f 0 . The amount of V * DCf is decided by the gain, K DC . The time constant T DC of the first-order transfer function represents the speed of the variation of stored energy of the DC capacitors. The upper limit, V DC,max , is set to +0.1 p.u., while the lower limit, V DC,min , is set to -0.1 p.u. to ensure the safe operation of DC voltage. Thus, the DC capacitors can be utilized to release or absorb the energy based on the frequency deviation.
However, one of the disadvantages of the above strategy is the latency issue due to the frequency signaling. Thus, the WTs need to monitor the WFVSC frequency, which is affected by the frequency signal transmitted through a communication interface from the grid. Nevertheless, this process seems to be superfluous because the grid frequency can be directly transmitted to the OWF rather than the WFVSC. In addition, the variation of the WFVSC frequency is regulated by a PI controller, which increases the operation time and slows down the response of the WTs.

F. CASCADING CONTROL
Although all the methods in Section III A -E can achieve the function of FR, they ignore the dynamics of the generator rotor speed of the WTs. In particular, the variation of rotor speed can affect the operation of WTs and the wind energy harvesting, and even can cause a worse frequency disturbance, i.e., the SFD. Consequently, the authors of [63] proposed a method called cascading control that can ensure a small variation of rotor speed for WTs and a high wind energy harvest. The difference between simultaneous control and cascading control is that the latter sets a higher priority for the VSC-HVDC than the OWF for providing frequency support. That is, the OWF can provide inertia support only when the DC voltage of the VSC-HVDC reaches the limit. Therefore, the OWF is the last resort for providing inertia support. In the cascading control, a communication channel is established to transmit the grid frequency to the WFVSC. Furthermore, the WTs can also provide frequency support according to the grid frequency.
The overall control scheme for cascading control is shown in Fig. 23. To achieve sequential utilization of the stored energy in the VSC-HVDC and the inertia support from the OWF, an appropriate deadband for the deviation of WFVSC frequency from the nominal value f WF is designed as: From (34), the WFVSC frequency does not change when the DC voltage is within the normal range, and the WT operates in its MPPT mode. Thus, the wind energy harvesting and economic profit are guaranteed.

G. GENERATOR EMULATION CONTROL FOR VSC-HVDC-CONNECTED OWF
Using the fast communication channel to transmit the grid frequency signal to the WF may have a high cost. Moreover, if the VSC-HVDC is designed to provide a large inertia support, then the DC-link capacitors will have large volumes and high cost. Thus, another method called generator emulation control (GEC) was proposed to fix these problems. The GEC was first applied to the PV inverter to provide synthetic inertia [64]. Consequently, this technique can also be applied to VSC-HVDC [65]. By using GEC, the VSC-HVDC can emulate the behavior of a conventional SG to provide synthetic inertia.
Under a three-phase balance and ignoring the loss caused by the internal resistance of the SG, the equivalent model of an SG is constructed and is shown in Fig. 24, where E is the internal EMF of the SG; X is the equivalent reactance; i s is the current flowing out of the SG; V o is the terminal voltage of the SG and δ is the difference in phase angle between E and V o . If V o is constant and δ is marginal, then the output active power of the SG can be expressed as: After the SG is connected to the power system, the grid frequency disturbance will affect the power output from the SG. Thus, Equation (35) is rewritten as: where k s is equal to |V o | |E| /X , ω gen is the electrical angular frequency of the EMF, and ω grid is the electrical angular frequency of the grid voltage.
If the rotor dynamics of the SG are considered, then the inertial behaviors of the SG can be expressed in a swing equation. Therefore, ω gen is expressed as: where k g is the gain related to the number of poles and the moment of inertia. The speed governor of the SG possesses the frequency-active power droop characteristic for primary frequency support. Thus, the additional input power P i owing to the droop characteristics is expressed as: where k d is the gain related to the droop constant and ω ref is the reference value of rotor speed, i.e., nominal speed. By combining (36), (37) and (38), the dynamic model of the SG with the frequency-active power droop control is found and is shown in Fig. 25. Given the dynamic model (Fig. 25), the transfer function is derived from the signal flow graph and Mason's gain formula [66]. Fig. 26 depicts the signal flow graph based on Fig. 25. Based on the signal flow graph, the step-by-step derivation of the transfer function is presented as follows.
where is the determinant of the graph; M 1 is the gain of the forward path from ω grid to P o , and 1 is the part of that is nontouching with the forward path from ω grid to P o . As a consequence, the transfer function between the output power and the grid frequency in a short-term or high frequency range, can be expressed as Equation (42). By contrast, the transient response in a long-term or low frequency range is expressed as Equation (42).
Equations (42) and (42) represent the inertial response and the droop characteristic of an SG, respectively. According    to [64], Fig. 25 is transformed into a simplified PLL, as shown in Fig. 27. The transfer function of the PLL is expressed as: where k o is the output gain of the GEC controller. Equation (44) can be approximated as −k o /s in a high frequency range, which is the same as (42). Equation (44) can be approximated as −k o /k p in a low frequency range, which is similar to the droop characteristic in (42). Consequently, the PLL can be used to represent the dynamics of the SG. The PLL enables the VSC-HVDC to emulate the behavior of the SG. To adjust the value of DC voltage of the VSC-HVDC when a frequency disturbance occurs, the modified GEC controller was proposed and shown in Fig. 28 [65]. The corresponding equation of the controller for adjusting the DC-link voltage is: Hence, the DC-link voltage can be regulated according to the frequency deviation. As the frequency drops, the DC-link voltage decreases, and the DC capacitors release the stored energy to the grid.
The system scheme in [65] is shown in Fig. 29. It is noted that the inertia support provided by the DC capacitors is limited. Thus, the OWF also needs to participate in FR. The DC voltage is used to transmit the grid frequency information to the WFVSC. The controller for adjusting the WFVSC frequency [65] is expressed as: When selecting the gain K B , the DC voltage should be limited within ±10% of the nominal value. To regulate the output power of the OWF, the frequency-power droop control is applied by: where P WF0 is the nominal active power from the WF. Therefore, the GEC controller (45) with (46) and (47) can be used to coordinate the OWF and the VSC-HVDC to provide frequency support.

H. ADAPTIVE FREQUENCY REGULATION METHOD
Although a VSC-HVDC-based OWF can provide inertia support to improve frequency stability, WTs need to reduce their output power and absorb the required power to restore the rotor speed after a grid disturbance, which causes a power imbalance throughout the rotor speed recovery. Additional active power must be injected to compensate for the power shortage. Otherwise, a significant SFD may occur. To resolve the potential risk, [67] proposed a strategy that uses the permanent power reserve of the deloaded WTs to offset the power shortage during the rotor speed recovery of the WTs. The proposed strategy is called adaptive frequency regulation (AFR) [67], which is based on real operating conditions. The WTs can adaptively provide the releasable kinetic energy according to the rotor speed. In addition, the VSC-HVDC can also release the stored electrostatic energy according to the DC voltage to provide frequency support. The control scheme proposed by [67] is shown in Fig. 30. Owing to the decoupling of the VSC-HVDC, the OWF cannot detect frequency disturbances. However, the sequential droop control strategy can be employed to achieve coupling between the OWF and the grid. It is expressed by: where K G , K A , K B and K WF are the droop coefficients of the onshore SGs, GSVSC, WFVSC and OWF, respectively; P G0 and P WF0 are the nominal values of the output power of onshore SGs and OWFs, respectively, and f 0 and f WF0 are the nominal frequencies of the onshore grid and the OWF, respectively. In previous studies, the parameters for the FR are almost constant. However, the releasable kinetic energy stored in the WT's rotor and the electrostatic energy stored in the DC capacitors are time-varying according to their operating statuses. The releasable kinetic energy, µ, and the electrostatic energy, ν, are defined as [68]:   where J WT is the moment of inertia of the WT; ω t is the rotor speed; ω t,min is the minimum rotor speed; ω t0 is the nominal rotor speed; C is the DC capacitor value; V DC,min is the minimum DC voltage, and V DC0 is the nominal DC voltage.
For the frequency controllers of the WTs, the inertial gain k in and the droop gain k f are calculated according to the releasable kinetic energy µ as expressed in (54) and (55), and the additional output power of inertial control P in and droop control P f are expressed by (56) and (57), respectively.
where k in0 and k f 0 are the gains of inertial control and droop control at the rated speed, respectively. The rotor speed recovery can be achieved by a PI controller, as expressed by (58), where ω t0 is the initial rotor speed and k p ω and k i ω are the proportional gain and the integral gain of the PI controller, respectively. Hence, the reference of the output power from the WTs with continuous rotor speed recovery,  P WT , is expressed as (59) and shown in Fig. 31 [67].
The deloaded WTs provide a persistent power reserve for offsetting the power shortage during rotor speed recovery to provide temporary frequency support and electrostatic energy recovery for the VSC-HVDC. As shown in Fig. 32, the area S 1 represents the released energy during the temporary frequency support and the area S 2 represents the absorbed energy during the rotor speed recovery, defined, respectively, as: where P WT and P m are the electrical power and the mechanical power of the WTs, respectively; t 0 is the initial time of the frequency event; t W _re0 is the initial time of the rotor speed recovery, and t de represents the duration of the recovery stage. According to the law of conservation of energy, S 1 equals S 2 . The persistent energy reserve provided from the deloaded WTs must exceed the energy absorbed by other WTs and the VSC-HVDC, which is expressed by: Notably, only a few WTs in commercial OWFs operate in the deloaded mode. Most WTs operate in MPPT mode. Thus, the use of deloaded WTs is a disadvantage of the adaptive strategy.
To illustrate the differences between the FR techniques introduced in this section, a comprehensive comparison of these strategies is shown in TABLE 1.
It is hard to tell which FR technique is the best. Every technique has its own advantages and drawbacks. Take the conventional scheme and the cascading control for example, cascading control can achieve a higher wind harvest, but it requires a larger deviation of the DC-link voltage, which may require a larger capacitance or may violate the DC-link voltage threshold when providing the inertia support. In contrast, the conventional scheme does not guarantee the high wind harvest nor the large deviation of the DC-link voltage. Therefore, it is a trade-off between the economic profits from the wind energy and the normal range of the DC-link voltage when using the cascading control.

IV. SIMULATION RESULTS
To validate the control methods described in Section III, this paper implements these methods by PSCAD/EMTDC. The parameters of the VSC-HVDC-connected WF model are shown in TABLE 2. An aggregated generic model of WTs is built to represent a WF in the simulation. An SG is tripped off at t = 10s to simulate an under-frequency event.
A total of eight methods were discussed in Section III. However, only the typical methods were selected for performing the simulation. The study cases are described in TABLE 2. Throughout the simulation, a deloaded WF is connected to the same bus where a WF provides frequency support based on the AFR strategy.
Case 1 is assumed to be the worst case, owing to the lack of frequency support from both VSC-HVDC and WF. In Fig. 33 (a), the frequency in Case 2 has a higher frequency nadir than Case 1, because the WF provides frequency support in which the grid frequency is transmitted through a communication channel to the WF. In Fig. 33 (b) and Fig. 33 (c), the frequency at the WF side and the DC-link voltage in Case 2 are kept constant throughout the simulation because the conventional scheme does not activate the inertia support from VSC-HVDC. By contrast, in Fig. 33 (a), the frequency with simultaneous control has an SFD caused by the reduction of output power from the WF. The output power from the WF using simultaneous control is the maximum among all cases. Nevertheless, in Fig. 34 (a) and Fig. 34 (b), the output power from the WF using simultaneous control is so large that the rotor speed of the WTs drops quickly. The output power of WTs for providing frequency support mainly comes from the stored kinetic energy in the rotor. Thus, the higher the rotor speed, the more the additional active power the WTs can provide. Although the rotor speed does not violate the minimum limit in Case 3, the output power is still slowly reduced because of the decrease of kinetic energy stored in the rotor.
The DC-link voltage is depicted in Fig. 33 (c). As described in Section III F, the frequency support from WF is activated only when the DC-link voltage reaches the limits. Therefore, the DC-link voltage in Case 4 drops the most among all cases. Furthermore, as the DC-link voltage recovers, the frequency at the WF side in Case 4 also starts to recover. Once the frequency returns to the nominal value, the frequency support from the WF stops and the output of the WF is lower than the pre-fault value to restore the rotor speed. In Fig. 34  (b), the rotor speed of WTs with cascading control has the smallest reduction and the fastest recovery. As a result, the cascading control ensures less wind energy loss compared to other methods.
The frequency response in Case 5 outperforms other cases. The difference between Case 3 and Case 5 is that the SFD is eliminated in Case 5. The frequency in Case 5 can be stabilized by an additional power from the deloaded WTs, as shown in Fig. 34 (c). Unlike Case 3, the SFD is avoided in Case 5, because the deloaded WTs managed to offset the power shortage during rotor speed recovery. The predefined rotor speed profile of the deloaded WTs is shown in Fig. 34 (d). As discussed in Section II-B, the power margin x of the deloaded WTs is set to 10% by using the overspeeding technique. The output power of the deloaded WTs cannot be increased immediately when the fault occurs VOLUME 9, 2021 because the output power is decreased with the reduction of rotor speed, which contradicts the goal of providing an energy reserve. As a result, the output power of the deloaded WTs drops slightly during the transient period.
To further investigate the validity of the above methods, various important metrics are summarized in TABLE 3. It shows that simultaneous control yields the largest active power delivered by the WF. However, this large active power causes significant rotor speed deviation. Therefore, the SFD is more severe than the first frequency drop, and the corresponding time to reach the frequency nadir, t nadir , is much longer than in other cases. In addition, the frequency metrics between Case 2 and Case 4 are similar, but Case 4 demonstrates a better rotor speed response for the WTs, as shown in Fig. 34 (b). With the assistance of the deloaded WTs, the output power of the WTs for providing frequency support lasts for an adequate amount of time and remains at a high  level in Case 5. It can be seen that the frequency nadir and the steady-state frequency in Case 5 are the highest. Consequently, the frequency metrics in Case 5 are the best among all cases.
Nonetheless, each method has its benefits and drawbacks. In fact, it is difficult to obtain a single optimal strategy. For example, the inertial gain and droop gain of WTs should be selected carefully to prevent excessive output power from the WTs when using simultaneous control. In terms of the cascading control, the rotor speed recovery of WTs is the fastest, but the DC-link voltage drops to the minimum limit. In terms of the deloading strategy, it is not an economic choice for commercial WFs, although the frequency response with AFR is the most efficient.

V. FREQUENCY REGULATION STRATEGIES FOR WTs AND ESSs
With the increasing penetration of renewable energy, grid frequency stability is facing serious challenges. Many grid codes have requested renewable energy resources to provide a fault ride through capability [69]. If ESSs can contribute to the ancillary services in power systems, then frequency stability can be further improved. Thus, the ESS has become an important role. This study conducted a comprehensive comparison of different FR strategies by WTs and ESSs.
Traditionally, the control methods of ESSs used to provide FR are generally based on inertial control and droop control. In [70], the authors validated the virtual inertia of the BESS by deploying an inertial controller in its energy management system. The inertia in a high wind penetration system is increased if the BESS has a significantly large capacity. The authors of [71] applied a lead-lag controller to control the BESS for stabilizing the system frequency. The lead-lag controller is used to control the output power by adjusting the d-axis and q-axis currents of the inverter of the BESS. The BESS with lead-lag controller outperforms that with a traditional PI controller for minimizing the frequency disturbance with increased wind penetration.
For advanced control, [72] utilized the fuzzy logic controller to control the output power of the ESS for providing frequency support. The reference power of the ESS is determined by fuzzy logic. The inputs of the fuzzy controller are the frequency deviation and the ROCOF, and the output is the reference power of the BESS. For example, if the frequency deviation is negative and large, then the reference power should be large and last for a long period to supply the required energy. If the frequency deviation is negative and small, and the ROCOF is positive and small, then the frequency would slowly increase to the nominal value and the reference power of the BESS would be not too large.
The power reserve can be quickly injected to the network if the WTs are operating in deloaded mode. The hybrid operation of deloaded WTs and the BESS ensures the power balance between supply and demand during a load variation [73]. The fast response of the BESS combined with the power reserve of the WTs can significantly improve the FR capability. The coordination between the level of the power margin of the deloaded WTs and the reference power of the FESS is discussed in [74]. In this coordinated strategy, the deloading operation of WTs can be avoided if the power margin can be handled by the FESS, which means the FESS is prioritized for providing frequency support. This method enables WTs to maintain sufficient power reserve while alleviating the need for the deloading operation.
The ESS can also help to maintain the constant output power of WTs. In [75] and [76], a supercapacitor is installed in the DC-link of a WT for alleviating the power fluctuation. The WT with a supercapacitor can be seen as a combined wind source that is always operating in MPPT mode. When there are wind speed variations and load steps, the supercapacitor can absorb or release active power to stabilize the system frequency, while keeping the WTs in MPPT operation. The installment of an ESS in the DC-link can also be applied to the VSC-HVDC system for mitigating power fluctuation and limiting the fault current [77]. Considering the transient state when the fault occurs, the superconducting coil and the flywheel are suitable for provide fault ride through support, owing to their fast response. The aim is to minimize the DC-link voltage deviation and fault current, as presented in the optimization problem in [78].
The frequency-control algorithm of the ESS and WF can be designed as a step-by-step strategy. A two-stage FR strategy was developed in [79]. In the first stage, wind speed is predicted to determine the reference power for the ESS. The ESS is only activated when the capacity of SGs is insufficient for mitigating the frequency disturbance. Then, the actual frequency deviation is measured to modify the reference power of the ESS in the second stage. In this two-stage strategy, the life of the ESS can be prolonged because the frequent charging and discharging cycles are prevented. Additionally, the duration of the FR process can be shortened, as the ESS is prepared in advance by predicting the wind power disturbance. In contrast, a three-stage coordinated control strategy among the WF, BESS and conventional SGs was proposed in [80]. The first stage is activated when a serious frequency drop is detected, and the WTs will perform the torque limit control to provide frequency support. In the second stage, the PMSGs switch to the deloading operation to prepare for the rotor speed recovery and the BESS will replace the WTs for providing frequency support. Finally, if the system frequency decreases to a specified value, then conventional SGs with power reserves can boost their output power to the grid. This strategy effectively improves the frequency response and avoids an SFD during the rotor speed recovery of the WTs.
In [81], a strategy based on superconducting magnetic energy storage (SMES) was proposed to compensate the required power for the generator rotor speed recovery of WTs. In principle, a VSC-HVDC-connected OWF with ESSs was capable of exhibiting good performance for providing frequency support. Reference [82] proposed an FR strategy by using the power reserve of deloaded WTs and the fast response of SMES to alleviate the burden of the diesel generator.
Typically, the types of ESSs can be classified into two categories according to their power and energy characteristics. The first category is high power density (e.g. supercapacitors, flywheels, SMES, etc.), and the other is high energy density, e.g., batteries. An ESS with high power density can absorb or inject a large amount of power within a very short period of time, especially for providing virtual inertia in a few seconds. By contrast, an ESS with high energy density can provide primary or secondary FR by sustaining its output power for a longer period, up to tens of minutes.
Utilizing hybrid types of ESSs for providing frequency support has been investigated in some studies [83]- [86]. Commonly, there are at least one high-power-density ESS and one high-energy-density ESS in hybrid operation. For example, the job of a supercapacitor is to deal with the highfrequency components or rapidly changing signals in the grid, while leaving low-frequency components or smooth signals for the BESS. The supercapacitor can absorb the ripple in power systems and help prevent the BESS from operating with a high depth-of-discharge (DOD) rate; thus, the life of the battery is extended.
When adding the ESS into the network for providing frequency support, the state of charge (SOC) of the ESS should be carefully considered. A SOC value of 50% was reported to be the optimal working point of the BESS in [87]. The SOC can be maintained at approximately 50% by using the SOC feedback control proposed by [87]. The pitch angle control of WTs and the droop control of SGs can be activated to mitigate the frequency disturbance when the SOC of the BESS exceeds the limitations. Improved droop control based on SOC was investigated in [88] and [89]. The droop gain can be adaptively changed according to the current value of SOC. That is, the higher the SOC, the more power the ESS can generate and vice versa. The level of SOC has a tremendous impact on battery lifetime, when it is used to provide primary frequency support. Consequently, the revenues for the Li-ion BESS were analyzed in [90]. The simulation results show that the primary frequency support strategy from ESSs should be carefully designed to avoid rapid deterioration in order to achieve a good economic profit.
The size of the ESS plays a crucial role, especially when considering the installation of the ESS to increase the system inertia. The size of the ESS can be determined according to the targeted frequency dynamics [91]. The proposed method in [91] estimates the size of the ESS in terms of the power shortage, so as to achieve the target inertia constant and the power/frequency characteristics. Additionally, the sizing is calculated for inertial response and primary frequency response.

VI. FUTURE TRENDS A. OPTIMIZATION OF COORDINATED CONTROL OF FREQUENCY REGULATION TECHNIQUES FOR VSC-HVDC-INTEGRATED WFs AND ESSs
To improve the robustness and reliability of the coordinated frequency control strategy under different frequency events, cooperation among the conventional SGs, renewable energy sources (RESs), and ESSs is required. The allocations of the amounts of generation among the conventional SGs, RESs, and ESSs for providing FR can be made based on the installation capital operation costs, lifespan, and system requirements. Furthermore, the precedence, duration, and ramp-up rate should be taken into consideration when implementing the IR, PFR, SFR, and TFR for different power reserves. Therefore, the need for research into optimization is urgent and essential. Additionally, different wind penetrations, variable wind-speed scenarios, and realistic cases should be taken into account when performing such analysis.
An optimal control strategy called model predictive control (MPC), which uses state-space predictions, has been popularly investigated in recent years. The adaptive MPC was utilized for load frequency control (LFC) in [92] and [93]. The adaptive MPC can minimize the frequency deviation in the presence of load disturbance. The weights of the cost functions of traditional MPC are chosen by trial-anderror in several simulation cases. By contrast, the adaptive MPC can tune the dynamic parameters based on the real conditions automatically. Consequently, the adaptive MPC displays greater robustness and reliability than the traditional MPC and PID controller in terms of the FR capability.
For advanced control, the MPC can be utilized for the LFC of a multi-area power system that suffers from malicious cyber attacks [94]. It reveals that more severe cyber-attacks such as denial of service (DoS) or deception attacks are worthy of research in detail, especially in high-tech power systems.
References [95], [96], [97] presented control strategies for an MPC providing LFC in a system containing an ESS. The authors of [97] demonstrated that the MPC can help reduce the capacity and energy of an ESS compared to a traditional PI controller. Another example of MPC for FR considering the power constraints of an ESS was proposed in [98], in which the optimal power allocation of an ESS was achieved in a tested real Taiwan power system. The applications of MPC for AC voltage and FR in the case of RES power variation and load disturbance in microgrids can be found in [99], [100], [101].
MPC can also be employed for coordinating the AGC for FR and DC power regulation in multi-terminal HVDC (MTDC) grids [102], [103]. Moreover, the communication delays and DC voltage constraints were considered when implementing the AGC in [104] and [105]. It is recommended that the nonlinear/adaptive MPC be further applied for the realistic model of a large-scale power system, e.g., the European super grid.
Some other optimization methods have been proposed in [106], [78]. The authors of [106] proposed a low-order response model for presenting the WF-MTDC participating in FR. This model helps establish the appropriate variation of DC-link voltage and frequency deviation for MTDC by setting the droop gain. In [78], the author proposed an optimization strategy for installing SMES in the DC-line of HVDC for alleviating wind power fluctuation and providing a fault-current-limiting capability. The optimal problem aims to minimize the DC power deviation, DC-link voltage fluctuations, the initial inductance and the initial current of the SMES. The optimal SMES needs less power and energy than non-optimal SMES when suppressing the fault current. Thus, the economic benefit is achieved.
Aside from the MPC algorithms, there are some advanced optimization frequency control strategies based on artificial neutral networks (ANNs), machine learning, generic algorithm and so on. Reference [107] calculated the optimal electrical frequency for an offshore WF to generate the maximum power during a normal operation. By contrast, the output power of a WF is reduced when a voltage disturbance occurs in the HVDC link, which corresponds to a fault at the onshore AC grid. The authors in [108] proposed an iterative procedure to compute the optimum parameters for supporting the fault ride-through capability of a VSC-HVDC-connected WF by using the mean variance mapping optimization (MVMO) algorithm. The developed meta-heuristic algorithm can provide the best solution to minimize the DC over-voltage. In addition, the frequency of WF can be regulated based on the DC-bus voltage, and the dynamic response of the VSC-HVDC-connected offshore WF can be improved by using the proposed evolutionary optimization algorithm. An adaptive droop control was proposed for frequency regulation in a VSC-multi-terminal DC (VSC-MTDC) system [109]. Additionally, a voltage-current-frequency characteristic was also proposed to autonomously redistribute the power sharing among all the grid-side VSCs (GSVSCs). Significantly, the optimal parameters for the adaptive droop control were obtained by carrying out the modal analysis and nonlinear simulations; thus, the frequency deviation can be minimized. The authors in [110] studied the integration of a WF into a VSC-MTDC system for load frequency control. A traditional objective function called integral time absolute error (ITAE) cannot decrease the overshoot of the system response. To resolve this problem, a modified objective function was proposed, in which the control parameters for all generation units and the modified objective function are determined by particle swarm optimization (PSO) algorithm. Moreover, the multiple constraints were considered in [110] to investigate a more realistic power system. The performance of the load frequency control by using the proposed advanced control is greatly improved. In [111], the authors investigated the virtual inertia control for the wind energy conversion system (WECS). A novel optimization control called artificial bee colony (ABC) algorithm was applied to calculate the optimal gain for the virtual inertia control of WTs. The proposed control strategy was tested in a multi-area system containing a VSC-HVDC, and the simulation results demonstrated that a high frequency stability and a low power deviation are achieved. Reference [112] proposed a coordinated frequency control strategy based on situation awareness (SA) for achieving the dynamic adjustment of the parameters of the frequency controller. The SA can automatically change the controller settings by monitoring the on-line operations, and it is divided into perception, analysis and prediction stages. By applying the SA in frequency regulation, the optimal performance can be obtained.
For a VSC-HVDC-integrated-WF system, traditional advanced optimization algorithms rarely consider energy storage systems (ESSs) for providing frequency support. The ESSs have numerous benefits for providing ancillary services. Hence, the interaction between them can be further studied in the future work.

B. VIRTUAL SYNCHRONOUS MACHINE TECHNIQUE FOR A WEAK AC SYSTEM
As the penetration of non-synchronous generation increases, the system inertia and strength are correspondingly reduced. The ability of a PLL to synchronize with the main grid loses accuracy when the short-circuit ratio (SCR) is less than 1.3, which is a possible value for a microgrid or a weak AC grid [113]. A virtual synchronous machine (VSM) was proposed as an alternative to PLL when the VSC is connected to the grid. The VSM is viewed as a potential solution for the next generation of grid-connected converters, especially for an islanded microgrid or an extremely weak grid [114], [115], [116]. The main idea of VSM is to emulate the electromagnetic equations and the rotor motion of conventional synchronous machines. Typically, there are two forms of VSM: a virtual synchronous generator (VSG) with a synchronous inverter [117]- [119] and a virtual synchronous motor -with a synchronous rectifier [120], [121]. A schematic diagram of a VSM for a WF and local loads connected to the main grid via VSC-HVDC is depicted in Fig. 35. Most literature has focused on the VSG for providing frequency support.
The basic control structure of the VSG is shown in Fig. 36. The active power adjustment of VSM relies on the P−f droop characteristic and the swing equation as shown in Fig. 37, while the output reactive power depends on the Q − V droop characteristic, as shown in Fig. 38 [122], [123]. The VSG emulates the rotor motion equation and the excitation controller of a traditional SG as follows: where P m is the mechanical input power, P 0 is the initial active power, K f is the frequency coefficient, f 0 is the nominal frequency, J is the moment of inertia, P is the electrical power, D p is the damping coefficient, ω 0 is the nominal angular frequency, E is the electromotive force, K U is the voltage coefficient, Q 0 is the initial reactive power, Q is the electrical reactive power, D q is the reactive power coefficient, U is the rms grid side voltage, U 0 is the nominal grid side voltage, and E 0 is the no-load electromotive force. An example of VSG applied at a VSC-HVDC stations can be found in [124], where primary and secondary frequency controls were achieved using a VSG. In [125], the inverter of the VSC-HVDC acts like the VSG providing a reference angle for the weak AC grid. It is advised that the VSG be applied in a zero-inertia or low-inertia system for providing frequency support. The stability and dynamic performance of the MTDC integrated into a very weak AC system using VSG were analyzed in detail in [126]. A VSG control that can suppress the negative sequence voltage in the passive network with unbalanced loads can be seen in [127]. The various types of impedance of the VSM-based VSC-HVDC were studied using Bode plots and the Nyquist criterion in [128]. Reference [129] proposed an adaptive strategy for the adjustable moment of inertia for a VSG Although these VSM strategies can achieve the frequency and AC voltage regulation capabilities for VSC-HVDC, none of them include the control mechanisms for both WF and ESS when providing frequency support. The importance of WF cannot be ignored when it is connected via VSC-HVDC. Therefore, it is vital to research the coordinated control strategy of VSMs for the VSC-HVDC integrated WF and ESS. Moreover, the frequency-control strategy of VSMs when a three-phase fault, voltage harmonics, and a voltage drop occur simultaneously in the grid is worthy of further study.

VII. CONCLUSION
As a result of the increasing penetration of wind power generation, the number of VSC-HVDC connected OWFs is increasing. Thus, both VSC-HVDC and WTs play very crucial roles in maintaining frequency stability. This study investigated and compared various FR techniques from WTs, VSC-HVDC, and ESSs. The contributions of this paper are summarized as follows: • Most of previous overview papers have been about the FR techniques for only WTs or only VSC-HVDC. However, this paper provided a detailed review of coordinated frequency control techniques for VSC-HVDCintegrated WFs.
• Unlike other review papers, this paper not only presented a comprehensive comparison of the FR techniques but also implemented a time-domain simulation to further analyze the effectiveness of the selected FR techniques.
• This work introduced some methodologies for using an ESS to coordinate WTs to provide frequency support.
• This paper showed current challenges and provided recommendations for the future. The FR strategies for a VSC-HVDC-connected WF can be divided into two categories: communication-based strategies and non-communication-based strategies. In communicationbased strategies, the grid frequency signal is transmitted to the WFVSC or OWF through a communication channel. However, the performance of the communication-based strategy can be affected by the latency, which delays frequency support from the OWF. By contrast, the non-communicationbased strategy can utilize an ancillary voltage controller [46], GEC controller [65] or sequential droop controller [67] at the VSC stations. The response from the OWF by the noncommunication-based strategy may be slower than those by the communication-based strategy, owing to controller operations and switch delays.
In the simulation results, it was shown that the AFR strategy yielded the best performance. The frequency nadir of the AFR was increased by 0.1241 Hz compared to the conventional scheme, and the maximum ROCOF was reduced by 25.5%. The respective frequency metrics of AFR are the best among these FR control strategies, showing that it is necessary to develop such an adaptive or optimal control approach. Notably, the minimum WT rotor speed of cascading control is increased by 12% compared to simultaneous control. Cascading control has the largest value for the minimum WT rotor speed, which is good for preventing the SFD.
The SFD problem caused by the rotor speed recovery of the WTs has rarely been studied. However, it is definitely worthy of attention because an SFD can cause a severe frequency nadir. To solve this issue, the rotor speed recovery of WTs should be properly designed using the energy stored in the DC-link capacitors of VSC-HVDC, the energy reserve of deloaded WTs, or the instant power of an ESS.
Several FR strategies, especially for the primary frequency response for WTs and ESSs, have been proposed. These strategies use supercapacitors, flywheels, and electrochemical batteries. Although secondary and tertiary frequency response strategies have seldom used WTs and ESSs, their importance cannot be ignored and the ancillary services provided by ESSs should be investigated further in the future.