Improved Generator Voltage Control in Power Flow

Solutions
Brandon M. Allison and Thomas J. Overbye James D. Weber
Department of Electrical Engineering PowerWorld Corporation
Texas A&M University Champaign, IL
College Station, TX [email protected]
[email protected], [email protected]

Abstract – Generator voltage control in power flow solutions In a physical system, generators are equipped with
may be better represented by a general Q(V) function instead of Automatic Voltage Regulators (AVR) that adjust their reactive
the current standard PV/PQ modeling. Voltage control modeling power output to regulate voltage. This control is based on the
in power flow simulations plays an important role in representing generator’s point of interconnect (POI) bus voltage and the
one of the more important aspects of physical power systems.
Historically, PV/PQ bus modeling has been used as a standard
generator’s assigned voltage setpoint [17] with an allowable
across simulation packages. However, physical systems provide tolerance band (with example values of ± 0.01 pu in [17]).
system generators with a setpoint tolerance, which may not be well Although each generator may have a different AVR
represented by the strict rule-based approach in PV/PQ modeling. implementation, in general they will operate with reactive
Instead, the voltage control may be better represented by a power as some inversely related function of voltage. Each
“reactive power is a function of voltage” control model, as this generator in the physical system must have an AVR that
better correlates with an actual AVR implementation. Some complies with its interconnect’s regulations.
system characteristics of PV/PQ modeling and two Q(V) function The Electric Reliability Council of Texas (ERCOT)
models are presented in the sections that follow. manages the electric grid covering most of the US state of
Index Terms—Generator Setpoint Tolerance, PV, PQ, Bus Texas and is responsible for maintaining its reliability. In the
ERCOT interconnect, generators must maintain voltage within
I. INTRODUCTION ± 0.02 pu of the voltage setpoint, or provide reactive power at
Solving power flow systems in simulation models has been their minimum or maximum capability [8] (depending on the
an area of interest and growth for at least 60 years [11], [15], sign of the difference). This response requirement is much more
[19], [23]. During this period, PV/PQ bus modeling [14] has relaxed than the “sharp” function that is implemented with
been the standard for mimicking physical AVR voltage control PV/PQ modeling.
among commercial simulation software packages, as well as in Although it may be logical to immediately implement
research and academic settings. some arbitrary voltage control function which theoretically
matches a physical AVR function, some method of metrics is
A. Traditional PV/PQ Switching Logic needed. The term “Dynamic Reactive Reserves” refers to
The PV/PQ modeling method labels each bus that has a “reserving” the ability to rapidly adjust the reactive power
real power injection (generators [+] and loads [-]) as either injection at any given bus, which may be necessary in a
known real power and voltage (PV), or known real power and dynamic event. This is typically done by ensuring that rapid
reactive power (PQ). When using a case that has no previous response (dynamic) reactive devices always have the capability
solution, the bus types are determined by setting all busses to to adjust their reactive injection in either direction [7]. These
type PV and solving. If a generator has a reactive power dynamic reactive devices can include many types of devices,
violation, its regulated bus is changed to a PQ bus and the power but the largest contributors to reserves are typically generators.
flow is resolved. Other bus type changes are made based on the For a generator to effectively provide dynamic reserves, it must
PV/PQ switching logic shown in Table 1. This process of not operate at either the minimum or maximum MVAR limit.
updating the bus types and resolving is repeated until no bus As Independent System Operators (ISOs) actively seek to
updates are required and all mismatches are within a small maintain system reliability and Dynamic Reactive Reserves,
tolerance [23]. The PV/PQ switching logic is as follows: they will adjust static reactive devices (such as discrete
capacitors) and voltage setpoints to minimize the number of
Table 1: PV/PQ switching logic [23]
generators operating near one of the two reactive power limits
Type Change PV, if BusType = PV and Qmin < Q < Qmax [7], [12]. Because of this, physical systems typically do not
PV, if BusType = PQ, Q = Qmax, and V < Vset have many generators which operate at their reactive power

New Bus Type = { PV,

PQ,
if BusType = PQ, Q = Qmin, and V > Vset
if BusType = PQ, Q = Qmax, and V ≥ Vset
limits. By contrast, many steady-state power flow cases, which
are employed by various ISOs in planning studies, inaccurately
solve with additional generators operating at a reactive limit.
PQ, if busType = PQ, Q = Qmin, and V ≤ Vset
PQ, if BusType = PV, and Q < Qmin, or Q > Qmax

978-1-7281-8192-9/21/$31.00 ©2021 IEEE

 Reactive Power Response Reactive Power Response

Qmax Qmax
Reactive Power

Reactive Power
Allowable Range
Qmin
Qmin
Vset - 2% Vset Vset + 2%
Voltage Voltage
Figure 1. Setpoint Tolerance Allowable Functions Figure 2. PV/PQ modeling Q(V)

In sections that follow, two new voltage control modeling deadband function. They are intended to implement a simple
methods will be presented along with several case studies. linear AVR response as well as an AVR response with a
deadband.
II. METHOD The linear function is the result of “drawing a line”
As noted, in a physical system, each system generator has diagonally through the Allowable Range in Figure 1, or through
an obligation to deploy and operate an Automatic Voltage the two points:
Regulator (AVR) which will adjust the units reactive power (Vmin, Qmax) | (Vmax, Qmin)
output and maintain voltage within regulations. Most where:
interconnects provide a setpoint tolerance, as well as the Vmin = Vset – Setpoint Tolerance
minimum reactive power absorption and production levels Vmax = Vset + Setpoint Tolerance
required during boundary voltage events [8], [24]. Using a
tolerance of ± 0.02 pu, the allowed reactive power functions And the function becomes:
include anything that fits in the “box” of Figure 1.
Most power flow steady state simulation packages solve a Qmax, for V < Vset – ST. .
power flow using an iterative approach, such as Newton’s &'&()"
Q(V) = { Qmin + (Qmax – Qmin) &(*+'&()" ,
method [19]. The power flow equations that are typically used
for Vset – ST < V < Vset + ST
in steady state simulation have the form of [14]: (3)
Qmin, for V > Vset + ST
#

𝑃! = 𝑉! $ 𝑉" [𝐺!" cos(𝛿! − 𝛿" ) + 𝐵!" 𝑠𝑖𝑛(𝛿! − 𝛿" )] (1) where:

"$%
Qmax is the generator high reactive power limit
#
Qmin is the generator low reactive power limit
Vset is the Voltage Setpoint
𝑄! = 𝑉! $ 𝑉" [𝐺!" sin(𝛿! − 𝛿" ) + 𝐵!" 𝑐𝑜𝑠(𝛿! − 𝛿" )] (2) ST is the setpoint tolerance
"$%

This AVR response of this function is shown in Figure 3. As

Here, Pk and Qk are the real and reactive power injections
noted in [17] the Qmax and Qmin values may actually be a
at each bus k. Gkn and Bkn values for each pair of buses (k and
function of the generator’s real power output. However, in the
n) are known from the Y bus, which is defined as an input. In
algorithm being presented here for reactive power control, the
order to uniquely define a (stable) solution (which consists of
real power and hence these limits, may be considered fixed.
solving for V and 𝛿 at each bus), at least two of Pk, Qk, Vk and
dk must be defined at each bus k. The PV/PQ modeling method Next a piecewise linear function is considered, which adds a
directly defines Pk and Vk, or Pk and Qk respectively. Another dead-band near the voltage setpoint where there is no reactive
method may to define one unknown directly and define a .
second in relation to a solvable value (like defining P and
Q(V)). Reactive Power Response
The traditional standard of PV/PQ modeling will result in
the response shown in Figure 2. PV buses correspond with any Qmax
Reactive Power

bus whose voltage and reactive power fall on the vertical line,
and PQ buses correspond with the two ending horizontal lines.
Many other methods of modeling voltage control are
possible, provided that they specify real power, along with Qmin
voltage or reactive power; or some relation between the two
(such as Q(V)). Here, two functions using a method with Vset - 2% Vset Vset + 2%
reactive power as a function of voltage are presented. These Voltage
include an inverse linear function, and a piecewise linear. Figure 3. Inverse Linear Q(V)
.
power supplied. With a 0.01 pu dead-band and a 0.02 pu geographical average voltage distribution over each of the
setpoint tolerance, the function is as shown in Figure 4. individual buses [13]. Here, blue is used for high voltages, and
red for low voltages.
Qmax, for V < Vset – ST . As the cases change from the PV/PQ model, to the linear
!"#$% (,-./!&')"#$% model, to the piecewise deadband model, system voltages
𝑉−
&'!() &'!() approach more extreme levels. However, there are no base case
for Vset – ST < V < Vset – DB . voltage or branch MVA limit violations in any of the three
Q(V) = { 0, for Vset – DB < Vset < V + DB . cases.
"#12 (,-./3&')"#12
𝑉− Table 2 presents the number of generators in each case
&'!() &'!()
for Vset + DB < V < Vset+ ST . which are operating at either one of their reactive power limits.
The PV/PQ case has the most limit operating generators.
Qmin, for V > Vset + ST (4)
Because physical systems typically maintain significant
where:
Dynamic Reactive Reserves, the linear Q(V) function method
Qmax is the generator high reactive power limit best represents a physical system in regards to this metric.
Qmin is the generator low reactive power limit
Vset is the Voltage Setpoint
DB is the deadband
ST is the setpoint tolerance
0 < DB < ST, Qmax > 0, Qmin < 0

This function’s reactive power response is significantly

relaxed from the PV/PQ function, and it may cause a significant
difference in the solution voltage profile. The AVR response of
this function is shown in Figure 4. The differences in physical
AVR implementations and the typical PV/PQ simulation
voltage control modeling likely contribute to the discrepancy
seen in the number of generators which operate at reactive
power in simulations.
Figure 5, PV/PQ modeling
III. CASE STUDY
In this section, the different voltage control modeling
methods are explored in detail. Each model’s effect on the
voltage profile, contingency analysis and dynamic reactive
reserves are presented.
For the case study, three synthetic grid cases are used [21].
These grid cases are entirely fictional, but are sufficiently
complex and representative of the characteristics of physical
actual electric grids to allow for research studies. Three
synthetic grids have 200, 2000 and 10,000 buses respectively.
A. 200-bus case
The figures that follow show the voltage profile,
contingency analysis, and reactive reserves results from the
200-bus case. The most obvious difference in the three methods Figure 6, Linear Q(V) Modeling
comes from the voltage contours. These are shown in Figures 5
through 7. The voltage contour visualization represents the
.
Reactive Power Reponse

Max Q
Reactive Power

dead band

Min Q

Setpoint - 2% Setpoint Setpoint + 2%

Voltage
Figure 4. Piecewise Linear Function Q(V) Figure 7, Piecewise Linear Q(V) Modeling
 In the contingency analysis, shown in Table 3, the
piecewise deadband function had the most violations. These
results again show that the different solutions from the three
modeling methods may affect system planning decisions.

Table 2: 200 Bus Case Dynamic Reactive Reserves

Number of Low Q
High Q limit Total
Generators Limit
PV/PQ 13 0 13
Linear Q(V) 2 0 2
Piecewise Q(V) 6 0 6

Figure 11, PV/PQ Voltage Contour

Table 3: 200 Bus Case Contingency Analysis
Contingency Max Low High Total
Violations Branch Bus V Bus V Violations It is apparent that the three modeling methods each result
PV/PQ 0 2 0 2 in significantly different results. The voltage profiles and
Linear Q(V) 0 2 0 2 contingency results from steady state solutions, such as these,
Piecewise Q(V) 0 5 0 5 have the potential to affect system planning changes in a
physical system. Inaccurate results during the planning stage
could cause future system complications to go undetected.
200-Bus Case Voltage Profiles Although it is clear that each of the modeling methods have
their differences, it is not yet clear which, if any, provide the
most accurate results.
B. 2000 Bus Synthetic Texas Case
The 2000 bus synthetic Texas case, is overlaid onto a map
of Texas and has a load profile which is similar to the actual
state load. However, it is not representative of any physical grid
system. Nevertheless, comparisons between steady state results
from this system and the actual operating point of ERCOT’s
system do have merit.
Figures 14 through 16 show the voltage contour for each
Figure 8: PV/PQ Modeling of the models in the 2000 bus case. Again, these show more
extreme voltages as each the voltage control model moves from
PV/PQ to the linear Q(V) function, to the piecewise Q(V)
function. Unfortunately, any most voltage information from
ERCOT’s physical system is considered CEII (Critical Energy
Infrastructure Information) and cannot be presented here.
However, ERCOT does have a publicly available aggregate
voltage profile, which is reproduced in Figure 17 below [10].
Figures 14 through 16 show the voltage profile of each of
the synthetic cases, and Figure 17 shows ERCOT’s physical
voltage profile. Comparison of ERCOT’s profile with each of
the synthetic cases shows that the case using the linear Q(V)
modeling method most closely resembles the physical system.
Figure 9: Linear Q(V) Modeling

Figure 10: Piecewise Linear Q(V) Modeling Figure 12, Setpoint Tolerance Voltage Contour
 Table 4: 2000 Bus Case Dynamic Reactive Reserves
Number of High Q
Low Q Limit Total
Generators limit
PV/PQ 14 42 56
Linear Q(V) 0 11 11
Piecewise Q(V) 0 23 23

Table 4 shows the number of generators which are

operating near one of their reactive power limits. Similar to the
200-bus case, the PV/PQ modeling method resulted in the most
generators which operate at a reactive power limit.
Figure 13, Droop Control Voltag Contour The strictness of PV/PQ voltage regulation causes many
generators to operate at one of their reactive limits. The linear
2000-Bus Case Voltage Profiles function in the 0.02 setpoint tolerance case results in only 11 of
544 generators operating within 5% of the upper or lower
limits. However, the Droop Control Case’s deadband may be
too “loose” leaving some generators without adequate support
from their neighbors and several additional generators at their
high limits.
C. 10,000 bus case
The final case considered is the 10,000-bus synthetic case
which is overlaid on the Western United States. The voltage
profiles, shown in Figures 18 through 20, illustrate the expected
Figure 14: PV/PQ Modeling aggregate voltage response across the many buses. Dynamic
reactive reserves results are shown in Table 5.

Table 5: 10,000 Bus Case Dynamic Reactive Reserves

Number of Low Q
High Q limit Total
Generators Limit
PV/PQ 485 544 1029
Linear Q(V) 82 69 151
Piecewise Q(V) 96 60 156

10,000-Bus Case Voltage Contours

Figure 15: Linear Q(V) Modeling

Figure 18 PV/PQ Modeling

Figure 16: Piecewise Linear Q(V) Modeling

Figure 19: Linear Q(V) Modeling

Figure 17, ERCOT Spring 2017 Voltage Profile [10]
 Optimization, Security and Control of the Emerging Power Grid,
Rethymno, 2013, pp. 1-11, doi: 10.1109/IREP.2013.6629370.
[7] ERCOT.com, ‘2018_OTS_Coordinated_Voltage_ Control”, 2020.
[Online]. Available: http://www.ercot.
com/content/wcm/training_courses/158273/2018_OTS_Coordinated_Vo
ltage_Control.pptx. [Accessed: 25- Apr- 2020].
[8] ERCOT.com, “ERCOT Nodal Operating Guide”, 2020. [Online].
Available: http://www.ercot.com/
content/wcm/libraries/202570/March_1_2020_Nodal_Operating_Guide.
pdf. [Accessed: 25- Apr- 2020].
[9] ERCOT.com, “ERCOT Nodal Protocols”, 2020. [Online]. Available:
http://www.ercot.com/content/wcm/libraries
Figure 20: Piecewise Linear Q(V) Modeling /204209/April_3__2020_Nodal_Protocols.pdf. [Accessed: 25- Apr-
2020].
IV. CONCLUSION [10] ERCOT.com, “ERCOT-NPRR_849_Overview”, 2020. [Online].
Available: http://www.ercot.com/
Voltage Control plays an important role in any physical content/wcm/key_documents_lists/148053/ERCOT-
electric system’s stability and sustained operation. Although it NPRR_849_Overview.pptx. [Accessed: 25- Apr- 2020].
is not possible to perfectly model any physical system in a [11] H. E. Brown, G. K. Carter, H. H. Happ and C. E. Person, "Power Flow
simulation, improvements to the current PV/PQ modeling Solution by Impedance Matrix Iterative Method," IEEE Trans. Power
App. and Sys., vol. 82, no. 65, pp. 1-10, April 1963, doi:
method are possible. New modeling methods, such as the linear 10.1109/TPAS.1963.291392.
Q(V) and piecewise deadband Q(V) function methods [12] J Adams, S. Sharma, S. H. Huang, C. Thompson, T. Mortensen, and E.
presented here, have the potential to better represent a physical A. Villanueva., "ERCOT ISO's experiences in handling voltage related
system. issues in the control center," 2011 IEEE Power and Energy Society
General Meeting, Detroit, MI, USA, 2011, pp. 1-4.
Although all cases presented here are synthetic, the [13] J. D. Weber and T. J. Overbye, "Voltage contours for power system
differences in the results from these cases are similar to the visualization," IEEE Trans. Power Systems, pp. 404-409, February, 2000.
differences that would arise from using the three modeling [14] J. Glover, T. Overbye, and M. Sarma, Power System Analysis and Design,
methods on any simulated system. Ultimately, inaccurate 6th ed. Boston, MA: Cengage, 2017.
[15] M. Bjelogrlic, M. S. Calovic, P. Ristanovic and B. S. Babic, "Application
results from steady state solutions, especially when related to of Newton's optimal power flow in voltage/reactive power control," IEEE
planning studies for physical systems, will have consequences Trans. Power Systems, vol. 5, no. 4, pp. 1447-1454, Nov. 1990, doi:
for the reliability of the grid which we depend on. 10.1109/59.99399.
The results from the 2000 synthetic case show that the [16] M. Okamura, Y. O-ura, S. Hayashi, K. Uemura and F. Ishiguro, "A new
power flow model and solution method; Including load and generator
linear Q(V) method better represents an actual grid’s voltage characteristics and effects of system control devices," IEEE Trans. Power
profile than the standard PV/PQ modeling method. This linear App. and Sys., vol. 94, no. 3, pp. 1042-1050, May 1975, doi: 10.1109/T-
Q(V) method has the potential to improve simulation results in PAS.1975.31938.
a wide variety of different systems. [17] NERC, “Reliability Guideline, Reactive Power Planning”, 2016,
Available: www.nerc.com/comm/PC_Reliability_
V. ACKNOLEDGMENTS Guidelines_DL/Reliability%20Guideline%20%20Reactive%20Power%
20Planning.pdf [Accessed: 12- June- 2020].
This work was partially supported through funding [18] R. D. Youssef, "Implicit generator and SVC modelling for contingency
provided by the U.S. National Science Foundation in Award scheduling of reactive power dispatch," IEE Proc. - Generation,
Transmission and Distribution, vol. 142, no. 5, pp. 527-534, Sept. 1995,
1916142, US Department of Energy’s (DoE) Cybersecurity for doi: 10.1049/ip-gtd:19952044.
Energy Delivery Systems program under award DE- [19] R. J. Brown and W. F. Tinney, "Digital Solutions for Large Power
OE0000895, and the Texas A&M Smart Grid Center. Networks," Trans. AIEE, Part III: Power App. and Sys., vol. 76, no. 3,
pp. 347-351, April 1957, doi: 10.1109/AIEEPAS.1957.4499563.
VI. REFERENCES [20] S. Khushalani, J. M. Solanki and N. N. Schulz, "Development of Three-
Phase Unbalanced Power Flow Using PV and PQ Models for Distributed
[1] A.B. Birchfield, T. Xu, K. Gegner, K.S. Shetye, T.J. Overbye, "Grid Generation and Study of the Impact of DG Models," IEEE Trans. Power
Structural Characteristics as Validation Criteria for Synthetic Networks," Systems, vol. 22, no. 3, pp. 1019-1025, Aug. 2007, doi:
IEEE Trans. Power Systems, vol. 32, pp. 3258-3265, July 2017 10.1109/TPWRS.2007.901476.
[2] A. Birchfield, T. Xu and T. J. Overbye, "Power Flow Convergence and [21] TAMU.edu. “Electric Grid Test Case Repository” 2020. [Online].
Reactive Power Planning in the Creation of Large Synthetic Grids," IEEE Available: https://www.electricgrids.engr.tamu.edu
Trans. Power Systems, vol. 33, no. 6, pp. 6667-6674, Nov. 2018. [22] W. -. E. Liu, A. D. Papalexopoulos and W. F. Tinney, "Discrete shunt
[3] A. Capasso and E. Mariani, "Influence of Generator Capability Curves controls in a Newton optimal power flow," IEEE Trans. Power Systems,
Representation on System Voltage and Reactive Power Control Studies," vol. 7, no. 4, pp. 1509-1518, Nov. 1992, doi: 10.1109/59.207375.
IEEE Trans. Power App. and Sys., vol. PAS-97, no. 4, pp. 1036-1041, [23] W. F. Tinney and C. E. Hart, "Power Flow Solution by Newton's
July 1978, doi: 10.1109/TPAS.1978.354582. Method," in IEEE Trans. on Power App. and Sys., vol. PAS-86, no. 11,
[4] A. W. Azizan, V. K. Ramachandaramurthy and C. K. Loo, "The influence pp. 1449-1460, Nov. 1967, doi: 10.1109/TPAS.1967.291823.
of embedded generator control modes on an electrical distribution [24] WECC.org, ‘Guide to WECC/NERC Planning Standards I.D: Voltage
network power flows and voltage profile," 2009 IEEE Bucharest Support and Reactive Power”, 2006. [Online]. Available:
PowerTech, Bucharest, 2009, pp. 1-7, doi: 10.1109/PTC.2009.5281867. https://www.wecc.org/Reliability/Voltage%20Stability%20Guide.pdf.
[5] B. Allison, D. Wallison, T. Overbye and J. Weber, "Voltage Droop [Accessed: 12- May- 2020].
Controls in Power Flow Simulation," 2019 IEEE Texas Power and Energy [25] Y. Lei, R. Wang, T. Li, Q. Tang, Y. Wang and J. Li, "Modeling PV/PQ
Conference (TPEC), College Station, TX, USA, 2019, pp. 1-6. switching in security constrained optimal power flow," 2019 IEEE
[6] D. K. Molzahn, V. Dawar, B. C. Lesieutre and C. L. DeMarco, "Sufficient Innovative Smart Grid Technologies - Asia (ISGT Asia), Chengdu,
conditions for power flow insolvability considering reactive power China, 2019, pp. 84-88, doi: 10.1109/ISGT-Asia.2019.8881086.
limited generators with applications to voltage stability margins," 2013
IREP Symposium Bulk Power System Dynamics and Control - IX