Autonomous Voltage Security Regions to Prevent Cascading Trip Faults in Wind Turbine Generators

Cascading trip faults in large-scale wind power centralized integration areas bring new challenges to the secure operation of power systems. In order to deal with the complexity of voltage security regions and the computation difficulty, this paper proposes an autonomous voltage security region (AVSR) for each wind farm and the point of common coupling (PCC) substation, whose voltage can be controlled in a decoupled way. The computation of the AVSR can be completed using a stepwise search method exchanging voltage and power information between the control center and the wind farms. At each wind farm, an AVSR is determined to guarantee the normal operation of each wind turbine generator (WTG), while in the control center, each region is designed in order to guarantee secure operation both under normal conditions and after an N-1 contingency. A real system in Northern China was used to carry out case studies to verify the effectiveness of the AVSRs proposed, and good performance was demonstrated using the Monte Carlo method.

 Abstract-Cascading trip faults in large-scale wind power centralized integration areas bring new challenges to the secure operation of power systems. In order to deal with the complexity of voltage security regions and the computation difficulty, this paper proposes an autonomous voltage security region (AVSR) for each wind farm and the point of common coupling (PCC) substation, whose voltage can be controlled in a decoupled way. The computation of the AVSR can be completed using a stepwise search method exchanging voltage and power information between the control center and the wind farms. At each wind farm, an AVSR is determined to guarantee the normal operation of each wind turbine generator (WTG), while in the control center, each region is designed in order to guarantee secure operation both under normal conditions and after an N-1 contingency. A real system in Northern China was used to carry out case studies to verify the effectiveness of the AVSRs proposed, and good performance was demonstrated using the Monte Carlo method. S the most promising renewable energy source (RES), wind power is widely used over the world. There are several voltage-related challenges for accommodating large-scale wind power such as voltage fluctuations and the voltage stability under disturbances. In order to address these operation issues in wind power grid, a number of techniques have been developed to enhance the wind power hosting capacity and stability of the power system [1][2][3][4][5], maintain the voltage of the wind power integration area within limits [6][7][8][9][10][11] and maintain an appropriate voltage profile with the help of on-load tap changes (OLTCs) and capacitor/reactor banks [25][26][27]. Furthermore, static methods such as PV curves and continuation power flow (CPF) are used to analyze the risk of voltage instability from the perspective of voltage stability in wind systems. There are also some researches focus on improving the voltage stability [12][13][14] of the power system with wind power.

Index Terms-autonomous voltage security region (AVSR), N-1 contingency, voltage control, wind power integration
One of the major challenges for large-scale wind power in China is cascading trip faults. During 2011~2014, several severe cascading trip incidents occurred in China due to the normal but not safe operation state [19]. Once a trip fault occurs, other wind farms' voltage will significantly increase and cannot be kept within a normal voltage limit. As a consequence, wind turbine generators (WTGs) in other wind farms are tripped by high-voltage protection systems. As more WTGs trip, wind farm voltages become higher, resulting in cascading trip faults. Here, a cascading trip incident in Zhangbei Wind Power Base in Northern China on Feb 26 th 2012, was recorded in Fig. 1 by synchronized measurements from deployed phasor measurement units (PMUs). As shown in Fig. 1, the cascading faults were triggered by short-circuit faults in wind farm GT, which caused the very low voltage. Unfortunately, most of the WTGs in China were not equipped with effective low-voltage ride through (LVRT) control, so these WTGs were shut down. Combined with the capacitors that were not switched off in time, this led to a sudden large amount of redundant reactive power. Due to the fact that the wind power pool area was connected with a relatively weak power grid, afterwards, the voltage profile in this integration area significantly increases during 0.4s~2.0s, resulting in great wind generation loss. According to the data from State Electricity Regulatory Commission of China, there are 193 large-scale cascading trip incidents during January to August in 2011 in Northern China, and the most severe wind power loss in an incident was 500 MW, which brings great challenges to power system operation. We can conclude from the cascading trip process that the static voltage profiles are crucial to secure operations and the most important reason leading to cascading trip is the improper static voltage magnitude profiles.
Thus in order to deal with the large-scale cascading trip problems and keep the wind farms working under normal and safe state, voltage security region is then proposed to determine voltage operation ranges for all the important buses in the wind pool area. When the wind farms operate within these regions, once a wind farm trips, other wind farms can still operate at an acceptable voltage level, and will not lead to cascading trip. There is some preliminary research on static voltage security region [15][16][17], however, some challenges still remain.
Firstly, large-scale wind pool areas usually include dozens of wind farms with thousands of WTGs, and all the voltage of the area shall be taken into account. Otherwise, any trip incident may result in cascading trip faults. A control center can hardly model all the details to guarantee each WTG's operation constraints both under normal conditions and N-1 contingencies.
Secondly, the boundary of static voltage security region is complex due to the intermittent and stochastic characteristics of wind power. Therefore, the computation of voltage security region for online application is also a great challenge. Several studies have proposed methods for reducing the computational burden of the boundary. Based on the two-level wind automatic voltage control (AVC) system in [18][19], an approximate N-1 voltage security region boundary encompassed by cutting planes for centralized multiple wind farms is presented in [20] and [21]. However, with the linearization assumption, the accuracy is sacrificed to reduce the computation burden.
Last, but not least, for practical applications, the hierarchical wind-AVC system uses an autonomous voltage controller in each wind farm and a synergic voltage controller in the control center [19]. It is more promising and practical for wind farms to independently control themselves without considering the details of other wind farms' operation. The previous research [22] didn't produce decoupled voltage operation ranges for wind farms.
Therefore, a concept of AVSR (autonomous voltage security region) is proposed in this paper. The definition of the AVSR in the wind power grids is: If the control center controls the point of common coupling (PCC) bus and wind farms control their own point of coupling (POC) buses in their own certain ranges, each wind farm can control all the WTGs within normal operation ranges under both normal conditions and any N-1 contingency, without considering the operation details of other wind farms. The set of these voltage operation ranges for PCC bus and POC buses is defined as the AVSR. The concept of "autonomous" means each wind farm can control their voltages by themselves without considering other wind farms, i.e., the AVSR produces decoupled voltage control strategies for each wind farm. The AVSR in this paper is proposed from the perspective of security, aims to deal with the large-scale cascading trip faults caused by high-voltage protection systems of WTGs when the terminal voltages of WTGs exceed their upper bounds after an N-1 contingency. According to the analysis of cascading trip process in Fig. 1, we can conclude that the how to compute autonomous voltage security region (AVSR) to acquire voltage security control ranges of each wind farm and avoid cascading trip is a static voltage security problem, and the computation of AVSR should be based on static power flow equations. The ASVR computed here could be adopted as constraints for the voltage controllers that are deployed in wind farms and control center. During the control process, of course, the dynamic models are crucial, especially the dynamic reactive power reserve of SVCs/SVGs will greatly influence the voltage profiles when contingencies happen [30]- [31]. In future work, we will further research how dynamics models influence the AVSR (e.g. the optimal allocation of SVC/SVGs) in wind farms.
This paper aims to extend the previous work to propose the static AVSR and focus on accelerating the computation. And the contributions of the paper can be summarized as follows.
(1) The AVSR for the PCC bus and the POC bus in each wind farm is proposed and can be used for decoupled voltage control among all wind farms and the PCC bus. That's the most important contribution.
(2) In previous studies [22], an iterative method for the system-wide computation was proposed. The stepwise search method proposed in this paper does not require iteration, and the necessary constraint information is exchanged only once between the control center and each wind farm, resulting in less necessary computation time. (4) DistFlow format [23] is used to compute optimal power flow (OPF) in each wind farm. This model is completely equivalent to power flow model with polar coordinates format in radial networks [23], which guarantees the computation accuracy. The stepwise method proposed in this work needs repeated computation, and the DistFlow model significantly saves computation time due to the higher linearity.
The paper is organized as follows. The voltage feasible region (VFR) for an individual wind farm is first described in Section II. Based on these VFRs, the AVSR is computed. Six necessary steps to acquire the AVSR and the information exchange between wind farms and the control center are given at the end of Section II. Section III presents case studies of a simple system and a real system. The effectiveness of the proposed stepwise method is demonstrated by Monte Carlo simulation in Section III followed by conclusions.

A. Computation structure of AVSR
Unlike hydro and thermal power plants, wind farms are often distributed over large areas and the wind powers are then connected to a high-voltage bus (110/220 kV-level) via several feeders (35 kV-level) in wind farms. The high-voltage bus in each wind farm is the POC bus. Several wind farms are then connected to a PCC bus in a substation at a higher voltage level (220/500 kV) via transmission lines, then centrally integrated into the power grid. A typical structure for centralized integration of wind farms in Northern China is shown in Fig. 2. According to the hierarchical wind-AVC system [19], the voltage of the PCC bus is controlled by the control center while the voltage of POC bus is controlled by wind farms. Therefore, it is feasible to find a decoupled voltage control range (the AVSR) for the control center and each wind farm. According to the AVC system, the AVSR's computing structure is designed as follows.
For each wind farm, an autonomous control strategy is designed to keep the POC bus within their own range, which takes into account the wind farm's parameters and aims to acquire both maximum and minimum voltage magnitudes for each POC bus in a centralized integration area.
For the control center, a synergistic control strategy is designed to coordinate all distributed POC buses, which uses the security-constrained optimal power flow (SCOPF) and ensures that the voltage of all wind farms satisfies the operational constraints under both normal conditions and after an N-1 contingency.

B. VFR for an individual wind farm
The typical structure of an individual wind farm is shown in Fig. 3. In each wind farm, the detailed network topology of different kinds of devices is taken into account. First, for an individual wind farm, it is supposed to find the VFR for the POC bus (V POC,i 0 ) and the PCC (V PCC 0 ) bus.
Note that the voltage of each WTG in a wind farm is a state variable, and the reactive power of each device is a control variable. In each wind farm, for a specified active power (1-f) and the given voltage of V PCC The radial network feature of wind farms is fully considered to improve computation performance. Here, (1-c) expresses the power flow model in DistFlow [24] format, which include three linear constraints (1-c-1)~(1-c-3) and one quadratic constraint (1-c-4). The higher linearity of the DistFlow model results in considerable time saving for the stepwise method below with repeated computation [23]. 22 , 1 , , ,  It should be noted that: 1) Decreasing the search step will produce more accurate results, but this will lead to a significant increase in computation time. Considering both accuracy and efficiency, search step is set as 0.01 p.u. 3) For each step, once V PCC 0, SP is given, V POC,i 0 is increasing in wind farm's total reactive power Q i 0 (in Fig. 3). Thus for each given V PCC 0, SP , wind farm's corresponding total reactive control capability Q i 0 and Q i 0 , which are used below for system-side in (3) 4) On-load tap changers (OLTCs) and capacitor/reactor banks are not considered as optimal variables due to the following two reasons: First, in China, most tap changers in wind farms could not be online regulated in operation. Few OLTCs are used to optimize the voltage profile only 3~5 times a day. In terms of capacitor/reactor banks, one of the reason leading to the cascading trips is the improper switches capacitor/reactor banks, which are gradually replaced by SVC/SVGs in wind farms in China. Thus this paper mainly focuses on the coordination of WTGs (wind turbine generators) and SVC/SVGs. Second, in order to guarantee secure and economic operations of wind farms in China, some wind farms may use the OLTCs and capacitor/reactor banks to optimize the voltage scheme every 1~4 hours. But different from OLTCs and capacitor/reactor banks' optimization, the proposed AVSR is applied to real wind systems online for 1~5 minutes' level. Therefore, the control center will refresh the AVSR after every operation of the OLTCs and capacitor/reactor banks. When wind farm computes AVSR, the operation state of OLTCs and capacitor/reactor banks remains unchanged, thus they are not considered as optimal variables in AVSR computation model. 5) Due to the fact that the computation time of AVSR is always less than 20 seconds and the AVSR is applied to real wind systems for 1~5mins level, the computation of AVSR is based on the assumption that a short-term wind speed forecasting model is known with sufficient accuracy [28]- [29]. Thus the AVSR model uses the current active power interface. Thus the AVSR model uses the current active power interface. On the other hand, the control center will refresh the AVSR every 1~5mins online in wind operation, which also guarantees the accuracy of AVSR in a real wind system.

C. AVSR for multiple wind farms
Based on the VFRs for each individual wind farm above, the AVSR is then proposed for POC buses of all wind farms and the PCC bus in the control center.
The AVSR means: each wind farm can autonomously control all their WTGs within normal operation ranges not only under normal conditions but also under any N-1 contingency   To compute the AVSR, the upper bounds and lower bounds of all wind farms' practical VFRs (green regions) are first linearized by two lines respectively for practical application. As shown in Fig. 6, there are N operational points on VFR's upper bound (from V1 to VN). (the upper-left small subfigure in Fig. 6 is the voltage feasible region, i.e. Fig. 4) The upper bound is better approximately linearized with a larger square of the triangle V1VnVN. It is assumed that n=n* is the optimal of total N points (2-a), then the two linearized constraints of the upper bound (line V1Vn and VnVN) can be expressed as (2-c). Similar to the upper bound, this method can be also applied to the linearization of the lower bound. There are also N operational points on VFR's lower bound (from V1 to VN). The square of the triangle V1VmVN reaches the maximum when m=m* (2-b), then the two linearized constraints of the lower bound (line V1Vm and VmVN) can be expressed as (2-d). (Vm is on the lower bound) 1 1,2,...
,  (3-a)    (3-a)), which also guarantees that the operation points are within the VFR (3-g) after N-1 contingency, the violation of constraints (3-g) may result in a cascading trip. For a wind pool area, all individual wind farms' trip faults should be considered, otherwise any trip fault is likely to cause a cascading trip fault triggered by the first trip fault. Here S is the sensitivity coefficient, e.g. S wp,j 0,qv denotes the sensitivity of PCC's (p) voltage (v) to wind farm (w) j's reactive power (q) under normal conditions (superscript 0). It should be noted: 1) αi is the weight coefficient for each wind farm. The weight coefficient for wind farms with a larger capacity of reactive power compensators is greater.
Then the square of AVSR can be computed using (5), the superscript p and q represent step p and q, as shown in Fig. 8. The AVSR can be obtained from the maximum S(p,q) expressed in (6), where the stars follow p or q represent the maximum solution of all steps. Another point should be noted is the requirement of accuracy. In addition to decreasing the search step in Fig. 4 and Fig. 8, linear interpolation can be used between two adjacent steps in Fig. 8 to get more data points in order to obtain a larger AVSR.

D. The computation process of AVSR in a wind pool area
The process of acquiring AVSR can be divided into six steps, as shown in table I, the information exchange and six computation steps between system-side and wind-farm-side are shown in Fig. 9. Step 2 Each individual wind farm sends the data set Step 3 The control center uses (2)  Step 5 Normalize each V POC,i(k) 0 using (4) and obtain the largest AVSR using (5). The result of AVSR is expressed as (6). Step 0.01 p.u. Fig. 9. Information exchange and six computation steps of AVSR between system-side and wind-farm-side For more clear illustration and understanding, we put all regions and curves into Fig. 10, which also shows the corres ponding step in table I to compute or get these regions and curves.  Fig. 11. AVSR for two wind farms and PCC bus A simple system with two wind farms was studied to verify the proposed method using Monte Carlo simulation. First, the Monte Carlo simulation method was used to generate 10,000 simulation points that guarantee normal operation under N-0 conditions, thereafter all N-1 contingencies based on each operation point were computed. The "N-1 secure points" were plotted green while "N-1 not secure points" were plotted red in Fig. 11. Then, the AVSR (blue cube) was computed for this simple system and plotted in Fig. 11. The simulation results demonstrate three characteristics of AVSR: 1) Security: All operational points in the AVSR were green points ("N-1 secure points").

A. A simple system with two wind farms
2) Accuracy: N-1 security cannot be guaranteed if operational points are located outside the blue cube in Fig. 11. Some red points ("N-1 not secure points") exist outside but near to the cube boundary, which shows that the boundary of computed AVSR is near to the real boundary of security region.
3) Autonomous ( Therefore, the AVSR proposed in this paper can be used for voltage control in a decoupled way.

B. A real system with eight wind farms in Northern China
A real system with eight wind farms comprising the ZB Wind Power Base in North China, as Fig. 2 shows, was studied to verify the effectiveness of the AVSRs proposed in this study. Firstly, the time and frequency computed once for wind farms are recorded in table II. Due to the higher linearity DistFlow model in wind farms and only once necessary information exchange, the computation does not take a lot of time. Fig. 12 shows the voltage deviation in the ZB wind power base after different N-1 contingencies. There are eight black lines, which represent eight different contingencies for eight wind farms. If eight wind farms and the PCC substation operate within their own AVSR, then after an N-1 contingency, even if the voltages of other wind farms and the PCC substation increase, they will not violate the voltages' upper and lower bounds of normal operation.

Wind Farm N w
Step 2 Step 6  Fig. 13 shows the voltage of each WTG in an individual wind farm, in which each white star represents a WTG. If a wind farm operates within its AVSR, as shown in Fig. 13 (a.1) and (a.2), then the WTGs operate at a lower voltage level before a trip fault. After other wind farms' trip fault, the voltage of all WTGs will increase, but they will not exceed the upper bound (1.10 p.u.). However, the situation is different if the wind farm operates without the AVSR, which is shown in Fig. 13

C. Comparison with other works
Here we compare the proposed AVSR with recent relevant researches from eight aspects in  Fig. 14) 2) The proposed AVSR also provided decoupled security voltage ranges [V PCC AVSR , V PCC AVSR ] for PCC substation, but other works didn't. 3) In each wind farm, the power flow model (in DistFlow format) is used. Thus the result of the proposed method will be more accurate. 4) The stepwise search method proposed in this paper does not require iteration, and the necessary constraint information is exchanged only once between the control center and each wind farm, resulting in less necessary computation time. Here we use 100 different power flow interfaces to compute VSR of the real system (as Fig. 2 shows) using different methods. The average total computation time was recorded in table III and the proposed method observes faster computations than other works.

IV. CONCLUSIONS
Due to the complexity of voltage security regions and the difficulty of computation, this paper proposes the AVSR and a stepwise search method for its computation. At each wind farm, an AVSR is designed to guarantee the normal operation of each WTG, while in the control center, each region is designed in order to guarantee normal operation both under normal conditions and after N-1 contingencies. If the control center controls PCC bus within its AVSR and all the wind farms' POC buses operate in their own AVSR, each wind farm can realize decoupled control of all the WTGs within normal operation ranges both under normal conditions and under any N-1 contingency without considering the operation details of other wind farms.
Compared with the previous methods, the results are more accurate because the power flow model is used instead of a traditional sensitive model in each wind farm. It will not take a lot of computation time because this method does not require iteration and uses power flow model with the DistFlow format in each wind farm. Case studies with s simple system and a real system verify the accuracy and effectiveness of the method, and good performance using a Monte Carlo simulation.
With increased wind power penetration, AVSRs may not exist. Thus, how to curtail wind power and acquire maximum security regions will be studied in a future study. Indeed, it's also important to deal with the randomness of active power due to the forecast errors. Compared with the AVSR proposed in this paper, the AVSR robust to active wind power randomness is much more complicated and the computation must take more time. Based on the proposed AVSR using the specific current active power interfaces in this paper, the AVSR robust to active wind power randomness will be studied in the future works. As we all known, systems dynamics play an important role in wind security. How dynamics influences the AVSR (e.g. the optimal allocation of SVC/SVGs) in wind farms, which is more complicated, which will be studied in future work.