A Smart Fault-Tackling Strategy Based on PFTE for AC Three-Phase-to-Ground Faults in the Multi-Terminal HVDC Wind Power Integration System: Further Foundings

Abstract

This paper describes a smart fault tackling strategy based on power flow transfer entropy (PFTE) for AC three-phase-to-ground (TPG) faults in the multi-terminal HVDC (MTDC) wind power integration system. The fault characteristics and transient energy transfer of different positions and properties are analyzed. Then, a double integral discrimination method based on PFTE is proposed to further distinguish the fault property. Considering the power flow balance, an adaptive coordination strategy of wind farms and energy dissipation resistors is proposed to deal with different AC faults. Finally, a smart fault-tackling strategy based on PFTE for AC three-phase-to-ground (TPG) faults in the MTDC wind power integration system is proposed. Under the proposed smart fault-tackling strategy, the MTDC wind power integration system achieves uninterrupted operation during any AC TPG fault at the receiving end. The experiment results confirm the applicability of the proposed fault-tackling strategy.

1. Introduction

With the transformation from traditional fossil energy to clean energy in the 21st century, wind power has broad application prospects due to its pollution-free and easy transformation characteristics [1,2]. In 2020, China’s installed wind power capacity showed explosive growth, reaching 71 GW, which ranked it first in the world [3]. Modular multilevel converter (MMC)-based high voltage direct current (HVDC) technology can directly connect to weak AC systems and has no transmission distance limit and no risk of commutation failure [4,5]. To achieve further rational integration, distribution, and consumption of onshore wind power, the MMC-based, multi-terminal HVDC (MTDC) wind power integration system has received notable interest [6,7,8].

The MTDC wind power integration system has the advantages of having no transmission distance limitation, no commutation failure, and enriched system efficiency [9,10]. However, it is undeniable that faults are prone to spreading in the DC grid due to its reticulate structure [11,12]. When an AC three-phase short circuit fault occurs on the valve side of the receiving MMC station, the transient excess power generated by the maximum power tracking control operation of the wind farms cannot transmit into the AC grid [13]. The surplus wind power causes an overvoltage in the DC line, and the converter station sub-module capacitance voltages also rise consistently [14,15]. If the MMCs remain unblocked, the whole HVDC power transmission system may break down.

To protect the safety of the HVDC power transmission system during AC faults, [16] proposed a voltage droop control in the wind farm to quickly reduce power when the DC voltage deviation exceeds the threshold value of 0.1 pu. However, the approach is only effective in point-to-point HVDC wind power integration systems. In [17], power reduction control in the wind farm MMC (WFMMC) of an MTDC wind power integration grid by detecting the DC voltage variation was proposed. However, this method cannot locate on which receiving grid an AC fault has occurred, and power transmission is entirely interrupted during the fault. Furthermore, even though traditional AC relay protection has fault location capability, the measurement signals are enormous, and the protection setting calculation in the MTDC wind power integration system is complex [18,19]. Currently, effective detection and location methods for receiving AC faults in MTDC grids, especially those integrated with wind farms, are lacking.

In [20], fault detection criteria based on power flow transfer entropy (PFTE) is proposed to locate the AC fault’s position. This method uses the power flow trend of the DC line to locate AC fault quickly, with less measured signal and no complex setting calculation. More importantly, this method realizes the remote fault detection and location of more than 100 km so that the converter station can perceive the fault location independently of the communication system.

However, the literature [20] lacks further analysis and application of PFTE, and the criteria have difficulty with permanent AC faults. To enable the system to cope with permanent AC faults, property identification of the faults is an essential middle link. Unlike DC faults, to ensure that the non-fault terminals are not severely affected, DC circuit breakers usually remain closed when AC faults occur. Hence, it is impossible to determine permanent faults by reclosing the circuit breaker. Additionally, methods based on AC circuit breakers are invalid because the operation delay time of AC circuit breakers is too slow to meet the requirements of the HVDC grid [21]. Therefore, it is important to determine how to identify the fault property without relying on the DC and AC circuit breakers.

Additionally, most existing methods can only solve one particular fault but are invalid for use with other faults. In [22], a reactive power compensation and AC voltage margin control method is proposed to suppress the rise in DC voltage during transient AC faults. However, when permanent faults occur, unbalanced power is continuously injected into the sub-modules. Then, the voltage equalization module and the inner-loop current control break down, leading to a severe DC overvoltage hazard. In [23], a DC braking resistor was proposed to achieve transient AC fault ride-through, but this causes the DC braking resistors to operate for long, causing them to become overheated when permanent faults occur. In [24], power coordinative control with braking resistors was proposed to solve permanent AC faults, but this method reduces power transmission and delays the system’s recovery time in transient AC faults. Further, the above fault ride-through methods have little fault location capability in the MTDC grid.

To solve the above problems, based on the literature [20], the characteristics of PFTE are further analyzed and utilized in this paper, proposing the fault property criteria based on PFTE to distinguish permanent and transient faults. Then the adaptive coordination strategy of wind farms and energy dissipation resistors is proposed to deal with different AC faults. Finally, combined with the fault location strategy proposed in the literature [20], a smart fault-tackling strategy based on PFTE for AC three-phase-to-ground (TPG) faults in the MTDC wind power integration system is proposed in this paper. The contributions of this study are as follows: A novel AC fault location and property identification approach based on the PFTE is proposed. Dissipation resistors (DRs) and power reduction control of the wind farms are conducted to achieve power rebalance in the system. Under the proposed smart fault-tackling strategy, the MTDC wind power integration system achieves uninterrupted operation during any AC TPG fault at the receiving end.

This paper is organized as follows: Section 2 introduces the configuration and control strategy for the MTDC wind power integration system. The fault characteristics and transient energy transfer of different positions and properties are analyzed in Section 3. Section 4 proposes a smart fault-tackling strategy based on PFTE for AC TPG faults in the MTDC wind power integration system, which includes fault location, transient energy dissipation, fault property identification, and coordinative control of the wind farm and DR. In Section 5, the validation results demonstrate the validity of the proposed methods. Conclusions are drawn in Section 6.

2. Configuration and Control Strategy for the MTDC Wind Power Integration System

2.1. Configuration of the MTDC Wind Power Integration System

The configuration of the MTDC-integrated wind power system is shown in Figure 1a. Wind farms 1 and 2 use permanent magnet synchronous generators (PMSG) with an integrated wind rating of 1500 and 3000 mw. Each PMSG is connected to a full-power converter composed of generator side VSC (GS VSC) and integrated side VSC (IS VSC). In the MTDC power grid, MMC stations are composed of hundreds of half-bridge sub-modules (HBSMs), and power transmission depends on overhead lines (OL1~OL4). To absorb the excess energy, dissipation resistors R₁ and R₂ consisting of multiple resistors in parallel are installed on the AC side of WFMMC2, as shown in Figure 1a.

2.2. Control of the MTDC Wind Power Integration System

(1) System level control

The control system of the MTDC wind power integration system includes wind farm control, WFMMC control, and grid side MMC (GSMMC) control. The wind farms adopt maximum power point tracking (MPPT)-based active power control to generate the rated power. Since wind farms rely on a stable AC voltage supply for integration, WFMMCs are set at a constant AC voltage. GSMMC1 adopts active power control to balance the power flow among the rectifier and inverter side MMC stations, and GSMMC2 adopts DC voltage control to stabilize the voltage in the MTDC grid.

(2) Converter level control

Each wind farm uses a direct-drive, permanent-magnet-type wind turbine [25], and the power is controlled by the cooperation of GS VSC and IS VSC, whose simplified control system is shown in Figure 1b. The pitch and torque controller, which carries out pitch angle control, torque control, and pitch compensation control, prompts the angular frequency ω_m to coincide with the reference value ω_ref. The outer controller decouples the output power $P_{\omega }^{\ast }$ and $Q_{\omega }^{\ast }$ to generate $I_{gdq}^{\ast }$. Then, the dq current controller produces a reference voltage $V_{gdq}^{\ast }$ for the SVPWM. The d-axis of the IS VSC uses constant DC voltage ($V_{dc}^{\ast }$) control to balance the power flow, and the reference value of the q-axis reactive power ($Q_i^{\ast }$) is set to zero.

As shown in Figure 1c, the DC power ($P_{dc}^{\ast }$), DC voltage ($V_{dc}^{\ast }$), or AC voltage($V_{ac}^{\ast }$) is adopted as the controlled variable of the outer controller to meet the needs of different application scenarios. The dq current controller produces a modulation ratio $m_{dq}^{\ast }$ based on the external controller’s input signal $I_{dq}^{\ast }$. Then, the inner MMC controller sends the dynamic sub-module switching number to the arm controller to generate each arm voltage (V_arm).

3. Analysis of Fault Characteristics and Transient Energy Transfer

To analyze the mechanism associated with AC TPG faults on the MTDC wind power integration system, the simplified equivalent circuit of the MTDC grid is displayed in Figure 2. As shown in Figure 2, P_wind1 and P_wind2 represent the output power of wind farms 1 and 2; P_{dc_WF1} and P_{dc_WF2} represent the DC transmission power of WFMMC1 and WFMMC2; P_{dc_GS1} and P_{dc_GS2} are the received DC power of GSMMC1 and GSMMC2; P_{ac_GS1} and P_{ac_GS2} are the AC transmission power of GSMMC1 and GSMMC2; and P_{MMC_WF1}, P_{MMC_WF2}, P_{MMC_GS1}, and P_{MMC_GS2} represent the instantaneous power of each converter station.

Since MMC relies on sub-module capacitors and arm inductors to store energy, the energy storage of each MMC is calculated as

$$E_{MMC}=6\cdot \frac{1}{2}NC{V}_{cavg}^2+{\displaystyle \sum \frac{1}{2}L{I}_{jk}^2}$$

(1)

where V_cavg is the average capacitor voltage of all SMs in the MMC, and I_jk (j = a, b, c; k = up, dn) is the instantaneous current of each leg.

Take the derivative of E_MMC, there is

$$\{\begin{array}{l}{P}_{MMC\_WF1}=(6N{C}_1{V}_{cavg}^{WF1}d{V}_{cavg}^{WF1}+{\displaystyle \sum L_1^{}}{I}_{jk}^{WF1}d{I}_{jk}^{WF1})/dt\\ P_{MMC\_WF2}=(6N{C}_2{V}_{cavg}^{WF2}d{V}_{cavg}^{WF2}+{\displaystyle \sum L_2^{}}{I}_{jk}^{WF2}d{I}_{jk}^{WF2})/dt\\ P_{MMC\_GS1}=(6N{C}_1{V}_{cavg}^{GS1}d{V}_{cavg}^{GS1}+{\displaystyle \sum L_1^{}}{I}_{jk}^{GS1}d{I}_{jk}^{GS1})/dt\\ P_{MMC\_GS2}=(6N{C}_2{V}_{cavg}^{GS2}d{V}_{cavg}^{GS2}+{\displaystyle \sum L_2^{}}{I}_{jk}^{GS2}d{I}_{jk}^{GS2})/dt\end{array}$$

(2)

where N represents the SM arm input number; C₁ and L₁ represent the SM capacitance and arm inductance of WFMMC1 and GSMMC1, respectively; and C₂ and L₂ represent the SM capacitance and arm inductance of WFMMC2 and GSMMC2, respectively.

Under normal operating conditions, the MTDC grid power balance equation is

$$\{\begin{array}{l}{P}_{dc\_WF1}=P_{OL2}-P_{OL1}+2L_{dc}{I}_{dc}d{I}_{dc}/dt\\ P_{dc\_WF2}=P_{OL1}+P_{OL4}+4L_{dc}{I}_{dc}d{I}_{dc}/dt\\ P_{dc\_GS1}=P_{OL2}-P_{OL3}-2L_{dc}{I}_{dc}d{I}_{dc}/dt\\ P_{dc\_GS2}=P_{OL3}+P_{OL4}\\ P_{dc\_WF1}+P_{dc\_WF2}=P_{dc\_GS1}+P_{dc\_GS2}+8L_{dc}{I}_{dc}d{I}_{dc}/dt\end{array}$$

(3)

According to Figure 2, the AC and DC balance equation of MMC is

$$\{\begin{array}{l}{P}_{wind1}=6N{C}_1{V}_{cavg}^{WF1}d{V}_{cavg}^{WF1}/dt+P_{dc\_WF1}\\ P_{wind2}=6N{C}_2{V}_{cavg}^{WF2}d{V}_{cavg}^{WF2}/dt+P_{dc\_WF2}\\ P_{dc\_GS1}=6N{C}_1{V}_{cavg}^{GS1}d{V}_{cavg}^{GS1}/dt+P_{ac\_GS1}\\ P_{dc\_GS2}=6N{C}_1{V}_{cavg}^{GS1}d{V}_{cavg}^{GS2}/dt+P_{ac\_GS2}\end{array}$$

(4)

where C₁ and L₁ are the capacitance of sub-module and arm inductance of WFMMC1 and GSMMC1, respectively and C₂ and L₂ indicate the capacitance of sub-module and arm inductance of WFMMC2 and GSMMC2, respectively.

According to the power balance equation of the DC grid, this paper analyses the AC faults at different locations and with different properties, including the following four types of faults:

• F₁ fault: instantaneous AC TPG fault on the converter at GSMMC1;
• F₂ fault: permanent AC TPG fault on the converter at GSMMC1;
• F₃ fault: instantaneous AC TPG fault on the converter at GSMMC2;
• F₄ fault: permanent AC TPG fault on the converter at GSMMC2.

3.1. Analysis of AC Faults at Different Positions

Compared with SM capacitors, the power fluctuation in the arm inductance hardly varies, which can be ignored to simplify the analysis.

In the case of an F₁ (or F₂) fault, P_{ac_GS1} immediately decreases to zero. Since the MMC isolates its AC and DC sides, the value of P_{dc_GS1} will not instantly drop, so $V_{cavg}^{GS1}$ rises causing an overvoltage according to the Equation(4). Then, the active power control of GSMMC1 becomes invalid, and the surplus wind power also feeds into the SM capacitors of GSMMC2, causing an overvoltage in $V_{cavg}^{GS2}$.

In the MTDC grid, GSMMC2 adopts constant DC voltage control, and its DC side voltage is

$$V_{dc\_GS2}=N\cdot V_{cavg}^{GS2}\cdot V_{C_{2}N}$$

(5)

where V_C2N represents the rated voltage of each sub-module capacitor in the GSMMC2. Thus, the DC voltage of GSMMC2 increases with $V_{cavg}^{GS2}$ during the AC fault procedure. Since it is located at the end of the OL4, according to the capacitance-boosting effect, the increase in $V_{cavg}^{GS2}$ further causes the transmission power to drop, resulting in an insufficient load supply. According to Figure 3, the DC voltage and sub-module voltage of GSMMC1 exceed the safety limit of 1.3p.u at 900 ms after the fault occurs.

When an F₃ (or F₄) fault occurs, P_{ac_GS2} immediately drops down to zero, and $P_{dc}^{GS2}$ decreases slightly, so the voltage of the sub-module capacitor $V_{cavg}^{GS2}$ rises. According to Equation (4), as wind farms operate in the rated power state, P_{dc_WF1} and P_{dc_WF2} remain constant. Thus, P_{dc_GS1} and $V_{cavg}^{GS1}$ increase as the fault spreads. Since the fault occurs at the DC voltage-controlled station, the transient DC voltage and sub-module capacitor voltage rise more swiftly. As shown in Figure 4, the overvoltage of the sub-module capacitor exceeds the safety limit of 1.3 p.u at 60 ms after the fault has occurred.

Comparing the most severe permanent faults, AC faults at different positions all cause the DC voltage to rise. Assuming the conventional detection method based on the change amplitude of DC voltage is applied, when an F₁ fault occurs, the DC overvoltage is still around the normal operation threshold (1.05 pu) within 70 ms of the fault occurrence, which may lead to protection-refusing actions.

The fault location must be determined within 60 ms of the fault occurrence or the DC overvoltage will exceed the safety limitation (1.3 pu). However, most AC fault protection methods take about 100 ms, which is not suitable for the MTDC wind power integration system. Therefore, other more apparent signals and more effective methods need to be adopted for fault location.

To analyze the relationship between DC transmission power and the AC fault position, the MTDC grid was disassembled, as shown in Figure 5.

According to the “torque method” of the ring network, power flow can be calculated as follows:

$$\{\begin{array}{l}{P}_{OL1}=\frac{-(Z_{OL2}+Z_{OL3}+Z_{OL4})P_{dc\_WF1}+(Z_{OL3}+Z_{OL4})P_{dc\_GS1}+Z_{OL4}{P}_{dc\_GS2}}{Z_{OL1}+Z_{OL2}+Z_{OL3}+Z_{OL4}}\\ P_{OL4}=\frac{(Z_{OL1}+Z_{OL2}+Z_{OL3})P_{dc\_GS2}+(Z_{OL1}+Z_{OL2})P_{dc\_GS1}-Z_{OL1}{P}_{dc\_WF1}}{Z_{OL1}+Z_{OL2}+Z_{OL3}+Z_{OL4}}\end{array}$$

(6)

In the DC network, the per unit length parameters of all lines are approximately equal, so Equation (6) can be calculated as

$$\{\begin{array}{l}{P}_{OL1}=\frac{-(l_{OL2}+l_{OL3}+l_{OL4})P_{dc\_WF1}+(l_{OL3}+l_{OL4})P_{dc\_GS1}+l_{OL4}{P}_{dc\_GS2}}{l_{OL1}+l_{OL2}+l_{OL3}+l_{OL4}}\\ P_{OL4}=\frac{(l_{OL1}+l_{OL2}+l_{OL3})P_{dc\_GS2}+(l_{OL1}+l_{OL2})P_{dc\_GS1}-l_{OL1}{P}_{dc\_WF1}}{l_{OL1}+l_{OL2}+l_{OL3}+l_{OL4}}\end{array}$$

(7)

According to Equations (4) and (7), the change in power flow in P_OL1 and P_OL4 is only related to the changes in P_{ac_GS1} and P_{ac_GS2}, so the position of the AC fault can be determined via the change in the DC line power flow.

3.2. Analysis of AC Faults with Different Properties

Permanent faults last longer, have more negative significance and present more difficulty for system recovery than transient faults. In [26], a method involving using an energy dissipation resistor to achieve AC fault ride-through was proposed. However, the energy dissipation resistors (DRs) cannot operate continuously due to overheating and the high cooling cost. Once the DRs are out of operation, the DC voltage will rise again to attack the system, as shown in Figure 6. Therefore, additional power coordination control is necessary to allow the system to identify permanent faults accurately.

Generally, the determination of a permanent fault depends on the reclosing of AC or DC circuit breakers. However, when AC faults occur, the DC circuit breakers remain closed at all times, so this method is unable to be applied for fault property identification. Additionally, to ensure that the DC overvoltage does not exceed 1.3 pu when an F₄ fault occurs, AC permanent fault identification should be completed within 60 ms of DR withdrawal, as shown in Figure 6b.

4. Smart Fault-Tackling Strategy Based on PFTE for AC TPG Faults

4.1. Power Flow Transient Entropy

According to Equations (4) and (7), the fault location can be determined by the two opposite DC transient power changes (P_OL1 and P_OL4). However, the power change of the MTDC grid is relatively slow, so the evolution of transient energy transfer cannot be predicted in advance. (The detailed proof is shown in Appendix A.)

In the MTDC power grid, the real-time dynamic balance of power generation and consumption is the premise of system stability. When the system occurs faults, the power flow is bound to change, so the power flow entropy can be used to reflect the state change trend of the system [27]. Reference [20] defined the power flow transfer entropy (PFTE) for the first time to reflect the DC line power flow change trend, which is expressed as follows.

$$H(t)=\{\begin{array}{l}K{\displaystyle \sum _{i=1}^m(1+P_i^{\ast }(t))\ln (1-P_i^{\ast }(t))}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}-1<P<1\\ K{\displaystyle \sum _{i=1}^m(1+P_i^{\ast }(t))\ln (P_i^{\ast }(t)-1)}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}P>1\end{array}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}$$

(8)

where m represents the number of DC lines between two converter stations. According to Equation (8), when $P_i^{\ast }$(t) =1, PFTE is prone to negative infinity. To avoid the calculation of dead zone in the experimental system, the boundary of H(t) is set as [−1000, 1000]. $P_i^{\ast }$(t) represents the load ratio of overhead line i at time t.

$$P_i^{\ast }(t)=P_i(t)/{\overline{P}}_{iN},\hspace{0.17em}\hspace{0.17em}(i=1,\hspace{0.17em}2,\hspace{0.17em}\dots ,\hspace{0.17em}{N}_l)$$

(9)

where P_i(t) is the instantaneous power of the DC line i, ${\overline{P}}_{iN}$ is the average transmission power of lines between two converters, N_l is the number of overhead lines in the MTDC grid. In the MTDC grid shown in Figure 1, under the rated operation, ${\overline{P}}_{iN}$ is

$$\{\begin{array}{l}{\overline{P}}_{1N}\approx {\overline{P}}_{3N}=\frac{P_{WF2}-P_{WF1}}{N_{l(WF1-WF2,GS1-GS2)}}\\ {\overline{P}}_{2N}\approx {\overline{P}}_{4N}=\frac{P_{WF2}+P_{WF1}}{N_{l(WF1-GS1,WF2-GS2)}}\end{array}$$

(10)

In the four-terminal MTDC wind power integration system, when the line is in a steady state, the value of PFTE is much lower than zero; otherwise, the PFTE value will be around zero. Based on the sensitivity of PFTE, the PFTEs of DC lines represent the trend of the transmission line from stable state to fault. The smaller its value is, the more stable the line state will be; otherwise, the line will be prone to fault occurrence.

By substituting Equation (7) into Equation (8), the PFTEs of OL1 and OL4 are expressed as

$$\{\begin{array}{l}{H}_{OL1}=K(1+\frac{-(l_{OL2}+l_{OL3}+l_{OL4})P_{dc\_WF1}+(l_{OL3}+l_{OL4})P_{ac\_GS1}+l_{OL4}{P}_{ac\_GS2}}{l_{\Sigma }\cdot \alpha \cdot {\overline{P}}_{1N}})\cdot \\ \hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\ln \left|1-\frac{-(l_{OL2}+l_{OL3}+l_{OL4})P_{dc\_WF1}+(l_{OL3}+l_{OL4})P_{ac\_GS1}+l_{OL4}{P}_{ac\_GS2}}{l_{\Sigma }\cdot \alpha \cdot {\overline{P}}_{1N}}\right|\\ H_{OL4}=K(1+\frac{(l_{OL1}+l_{OL2}+l_{OL3})P_{ac\_GS2}+(l_{OL1}+l_{OL2})P_{ac\_GS1}-l_{OL1}{P}_{dc\_WF1}}{l_{\Sigma }\cdot \alpha \cdot {\overline{P}}_{4N}})\cdot \\ \hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\ln \left|1-\frac{(l_{OL1}+l_{OL2}+l_{OL3})P_{ac\_GS2}+(l_{OL1}+l_{OL2})P_{ac\_GS1}-l_{OL1}{P}_{dc\_WF1}}{l_{\Sigma }\cdot \alpha \cdot {\overline{P}}_{4N}}\right|\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\end{array}\hspace{0.17em}$$

(11)

where l_OL3 + l_OL4 > l_OL1 + l_OL2, l_OL4 < l_OL1 + l_OL2 + l_OL3.

According to Equation (11), when an F₁ fault occurs, ΔH_OL14 = H_OL1 − H_OL4 increases. When an F₃ fault occurs, ΔH_OL14 decreases. Based on this, the PFTE can be used to locate the fault’s position. Since the logic discriminant signal (H_OL1 and H_OL4) and action protection equipment (DRs) are in the same converter station, the proposed method is not dependent on an additional communication system.

4.2. The Smart Fault-Tackling Strategy for Uninterrupted Operation of the System

On the basis of PFTE, this section proposes a smart fault-tackling strategy to support the system to determine fault locations and properties accurately. As a result, the extensive data from detection signals and the uncertainty of long-distance communication are eliminated. Therefore, the system can effectively ride through different AC TPG faults with minimized impact from the fault.

The smart fault-tackling strategy depicted in Figure 7 includes four steps: AC TPG fault location, transient energy dissipation, AC TPG fault property identification, and coordinative control of wind farm, and DRs. In Figure 7, SS1 and SS2 are control signals corresponding to dissipation resistors R₁ and R₂. ΔH_OL14 represents the subtraction value of the power flow transfer entropy (PFTE) of OL1 and OL4. FNreset acts as the fault property identification signal. CDH1 and CDH2 are the two integral values of ΔH_OL14.

In the proposed AC TPG fault diagnosis and ride-through method, instantaneous faults can be detected and located within 30 ms, and the DRs are input continuously for 200 ms. Then, after the DRs exit, the system returns to normal. Permanent AC faults can be accurately identified within 50 ms of the DRs exit. After 500 ms, the wind farm output power is decreased, and the system recovers. The details of each step are as follows.

4.2.1. Step I: AC TPG Fault Location Strategy

The AC fault location strategy is shown in Figure 8, where H_OL1 and H_OL4 represent the PFTEs of OL1 and OL4, respectively. When the DC voltage exceeds the normal operation threshold V_lim (1.05pu), the system is identified as having an AC fault. Then, if ΔH_OL14 > ΔH_th, the fault is located at GSMMC1, while if ΔH_OL14 < 0, the fault is located at GSMMC2. In the proposed fault location strategy, once the fault position has been located, the result will be locked to prevent the strategy from running repetitively.

4.2.2. Step II: Transient Energy Dissipation

In prior research, the DRs have been placed on the onshore side to enable uninterrupted system operation when onshore AC faults occur [28,29]. However, the method cannot prevent surplus power from feeding into the WFMMC. Paralleling DRs on the AC side of WFMMC is a much safer method to protect all MMCs and other power electronics. Since the maximal amount of unbalanced power during the most severe AC fault is 3000 MW, installing DRs on the AC side of WFMMC2 can meet demands. To ensure a timely and bidirectional control response, the DRs are controlled by a couple of anti-parallel fast thyristors with less than 1 ms, as detailed in [30]. The layout of the dissipation resistor is shown in Figure 9.

To absorb the appropriate amount of surplus wind power under different situations, the DRs are divided into two groups (R₁ and R₂), symmetrically arranged in three phases. The energy dissipation resistor of each phase is composed of four resistors in parallel, and the R_max can be calculated as:

$$R_{\max }=\frac{{(V_{ac}^{{}^{}}/\sqrt{3})}^2}{P_N/3/4}=\frac{4V_{ac}^{{}^2}}{P_N}$$

(12)

where P_N is the rated output of the wind farm and V_ac is the AC line voltage. Based on the fault location criterion, the switching states of R_max can be determined. When ΔH_OL14 > ΔH_th, the S_{ki(i = 1,2,3,4)} of R₁ is closed to dissipate 1500 MW of surplus wind power; when ΔH_OL14 < 0, the DRs R₁ and R₂ are connected in the circuit to dissipate 3000 MW in total.

4.2.3. Step III: AC TPG Fault Property Identification

According to the analysis presented in Chapter 3.2, fault property identification must be completed within 60 ms after the DR exits.

Take the F₂ fault as an example, before the dissipation device exits, almost the full power of the system is transferred to GSMMC2 and the energy dissipation resistor, and P_{dc_GS1} is approximately equal to 0; after the energy dissipation device exits, P_{dc_GS1} begins to increase again, while P_{ac_GS1} is still at zero.

According to Equation (3), there is

$$P_{dc}^{GS1}=6N{C}_1{V}_{cavg}^{GS1}d{V}_{cavg}^{GS1}$$

(13)

Thus, the SM capacitor voltage will rise again, leading to DC overvoltage and power transmission reduction of the system. Taking ΔH_OL14 as the detection criterion, the comparison of ΔH_OL14 in the two scenarios is shown in Figure 10a.

It can be seen from Figure 10a that when the DRs exit, the ΔH_OL14 of the F₂ fault rises sharply and generates a high pulse peak. However, the threshold value of ΔH_OL14 has already been utilized for fault location. To avoid conflict with the fault location strategy, the ΔH_OL14 integral is utilized to determine the fault property. The first integration at 100 ms after the input of DRs into the operation is done to locate the AC TPG fault occurrence position of the system and eliminate the interference of other faults; this is named as CDH1. The second integration at 50 ms after the exit of DRs is done to judge whether the fault is permanent or instantaneous; this is named CDH2, as shown in Figure 10b.

On the other hand, when an F₄ fault occurs, since the amount of unbalanced power is relatively large after the exit of the dissipation resistance, the SM capacitor will experience overvoltage within tens of milliseconds. Therefore, it is also necessary to take the $V_{cavg}^{WF2}$ into account. The logic control diagram of the FNreset signal is shown in Figure 11, where CDH1 and CDH2 are the integral values of ΔH_OL14, CDH_lim1 and CDH_lim2 are the corresponding thresholds, and V_{cavg_lim} is the threshold of the average SM capacitor voltage of the WFMMC2.

4.2.4. Step IV: Coordinative Control of the Wind Farm and DRs

To solve the permanent AC faults, the output of the wind farm needs to be reduced by its internal GS VSC. The enhanced power reduction control method is as follows:

In Figure 12, when the FNreset signal equals to 0, the P_sref = P_ref, and the wind turbine maintains rated output; when the FNreset signal equals jumps to 1, The comparator will convert the selector input to channel B, so the power reduction control is effective. The K₁ and FNreset signals are transmitted by the WFMMC2 through cables. Referring to Figure 11, if a fault occurs on GSMMC1, then CDH1 > CDH_lim1, K₁ = 0; else, if the fault occurs on GSMMC2, then CDH1 < CDH_lim1, K₁ = 1. Considering the inertia time delay of the wind farm, it is necessary to cooperate with DRs during the decrease of the wind farm output power to ensure the safety of the system.

5. Validation Results

To verify the effectiveness of the proposed AC fault diagnosis and ride-through approach, the system shown in Figure 1 was built on RTDS setup. As shown in Figure 13, RTDS is composed of a GTWIF board and a PB5 board. The GTWIF board is connected with the workstation to ensure the synchronization of multi-core operation. The PB5 board acts as the processor to read and analyze the power system components and secondary controllers through the local area network. Since the positive and negative poles of each MMC are entirely symmetrical, the parameters of the unipolar MMC are listed in

Table 1. Unipolar MMC parameters of the MTDC wind power integration system [20].

	MMC	WFMMC1	WFMMC2	GSMMC1	GSMMC2
Parameters
Converter capacity/MVA		750	1500	750	1500
Grid side AC voltage/kV		230	230	525	525
DC voltage/kV		500	500	500	500
Connection transformer	Capacity /MVA	1800	900	900	1800
	Voltage ratio	230/260	230/260	525/260	525/260
	Leakage resistance uk (%)	15	15	15	15
sub-module rated voltage/kV		2.05	2.05	2.05	2.05
sub-module number of per arm		244	244	244	244
sub-module capacitance/mF		8	15	8	15
Arm inductance/mH		50	50	50	50
Flat-wave reactor inductance /mH		150	150	150	150

. The proportionality coefficient of PFTE was set as 200. The threshold of ΔH_OL14 was 400. CDH_lim1 = CDH_lim2 = 200, and the limitation of the submodule-capacitor voltage was 1.2p.u.

5.1. Diagnosis and Ride-through of the F₁ Fault

The system first operated in the rated normal state for 1.2 s, which is the initial moment of the RTDS simulation record. Then, an F₁ fault lasting for 0.1 s occurred at 1.3 s. During the fault, AC and DC circuit breakers remained closed, and the control strategies of the system were the same as during normal rated operation.

Figure 14a,b shows that both the DC voltage of the system and the capacitor voltage of each MMC increased after the fault occurred. During the first 20 ms after the fault occurred, the receiving power of GSMMC1 gradually decreased, but the receiving power of WFMMC1 and WFMMC2 remained the same as normal, as shown in Figure 14c. Thus, ΔH_OL14 increased to the threshold ΔH_th = 400, and the DRs were triggered. Figure 14d is the power absorbed by the DRs, which lasted for 200 ms and then automatically stopped. Subsequently, the fault property identification logic started. As CDH2 was always below the threshold, CDH_lim2 = 200, the fault was determined to be instantaneous, as shown in Figure 14f. The system returned to normal rated power transmission after 1.8 s.

5.2. Diagnosis and Ride-through of the F₂ Fault

The permanent F₂ fault occurred at 1.3 s. Other conditions were consistent with those of F₁.

Figure 15a,b show the DC voltage of the system and the capacitor voltage of each MMC increased as well. Owing to the proposed AC fault countermeasure, there was no overvoltage vision when the fault occurred. During the first 20 ms after the fault occurred, the power flow changes were the same as those for the F₁ fault. Thus, the power absorbed by the DRs and ΔH_OL14 was also the same, as shown in Figure 15c,d. After the DRs automatically exited, the fault property identification logic started. As CDH2 exceeded the threshold, CDH_lim2 = 200, the fault was determined to be permanent, as shown in Figure 15e. The DRs were re-input to gain time for wind power reduction. After 500 ms, the output of the wind farm dropped to half that of the original, and the dissipation resistors were withdrawn (see Figure 15f). Then, the system resumed its normal operation after 2.0 s but was missing 1500 MW of wind power compared to the rated operation.

5.3. Diagnosis and Ride-through of the F₃ Fault

The F₃ fault occurred at 1.3 s and lasted for 0.1 s. The system response is shown in Figure 16. Figure 16a,b shows the DC voltage of the system. The capacitor voltage of each MMC increased slightly more than for the fault in GSMMC1 but was still under the safety threshold of 1.3 pu. During the first 20 ms after the fault occurred, the receiving power of GSMMC2 and WFMMC2 gradually decreased, but the receiving power of WFMMC1 remained the same as normal, as shown in Figure 16c. According to the Equation (11), ΔH_OL14 first pulsed to 600 and then immediately dropped down to 0, so all DRs were triggered. Figure 16d shows the power absorbed by the DRs, 3000 MW for 200 ms, before they automatically exited. Subsequently, the fault property identification logic started. As CDH2 was always below the threshold, CDH_lim2 = 200, the fault was classified as instantaneous, as shown in Figure 16f. The system returned to normal rated power transmission after 1.9 s.

5.4. Diagnosis and Ride-through of the F₄ Fault

Finally, the F₄ fault also occurred at 1.3 s, and the other conditions were consistent with the previous validation conditions.

Figure 17a,b show that the DC voltage of the system and the capacitor voltage of each MMC increased as well. Owing to the proposed AC fault countermeasure, there was no overvoltage vision during the fault. In the first 20 ms after the fault occurred, the power flow changes were the same as those for the F₃ fault. Thus, the power absorbed by the DRs and ΔH_OL14 was also the same, as shown in Figure 17c,d. After the DRs automatically exited, the fault property identification logic started. Although the CDH2 value, shown in Figure 17e, was always below the threshold of CDH_lim2 = 200, the capacitor voltage of WFMMC2 exceeded V_{cavg_lim} = 1.2 p.u (see Figure 17b). Thus, the fault was classified as permanent. The DRs were re-input again to gain time for wind power reduction. After 500 ms, the output of the wind farm dropped to 0, and the DRs were withdrawn (see Figure 17f). Then, the system resumed normal operation after 2.1 s but was missing 3000 MW of wind power compared with the rated operation.

6. Conclusions

In this paper, a smart non-communication AC fault-tackling strategy including fault location and property identification is proposed for the MTDC wind power integration system. Additionally, an adaptive coordination strategy of wind farms and energy dissipation resistors is proposed to deal with different AC faults. Under the proposed smart fault-tackling strategy, the system can operate uninterrupted during any AC TPG fault at the receiving end. For instantaneous AC faults, the system can resume normal-rated operations within 600 ms of the fault occurrence. For permanent AC faults, the system can achieve secondary power balance within 800 ms of the fault occurs, and it can maintain maximal power transmission during the fault. In summary, this strategy greatly improves the anti-interference and power transmission efficiency of the MTDC wind power integration system.

So far, the effectiveness of this method is only verified in the four-terminal DC power grid. When the system has more end-drops (such as hierarchical connection mode), the characteristics of different AC faults may be more similar, leading to the failure of the criterion. Therefore, the AC fault ride-through in a complex system will be further studied.