Direct Incorporation of Fault Level Constraints in Optimal Power Flow As A Tool For Network Capacity Analysis

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO.

4, NOVEMBER 2005 2125

Direct Incorporation of Fault Level Constraints


in Optimal Power Flow as a Tool for
Network Capacity Analysis
Panagis N. Vovos and Janusz W. Bialek

Abstract—The aim of this paper is to present a method for the equipment is transferred to developers of new generation.
direct incorporation of fault level constraints (FLCs) in the optimal Clearly, efficient use of the existing protection infrastructure
power flow (OPF) as a tool for network capacity analysis, i.e., op- improves the investment conditions for new generation.
timal generation expansion planning within an existing network. A
mathematical methodology to convert constraints imposed by fault Usually, generation expansion planning is done in a heuristic
levels to simple nonlinear inequality constraints is developed. No manner. The candidate locations for new capacity are few, so
new variables are introduced in the OPF formulation to describe a small number of power flows for the suggested expansion
the additional constraints. Most common OPF-solving engines al- options is enough to determine if the future system will re-
ready have the computational capacity to handle numerous non- spect system constraints. Loads are considered equal to their
linear constraints, such as the ones described by the power bal-
ance equations on buses. Therefore, once FLCs are converted to peak values, because it is assumed (not always true) that this
nonlinear constraints described by OPF variables, they can be di- case corresponds to the highest power flows. However, as dis-
rectly introduced to any optimization process performing the OPF. tributed generation increases, the candidate locations become
A 12-bus/15-line test case demonstrates the advantages of the new numerous. So, a more concise method is needed to coordinate
method in comparison with a previously proposed iterative method the allocation of new capacity on the network in order to ex-
that converted them to restrictions on new capacity. It also proves
that when FLCs are ignored, the capacity of the network to absorb ploit the existing network capabilities to the maximum. Several
new generation is overestimated. coordination strategies have been suggested for the effective al-
location of new generation capacity [1]–[3]. In [4], a method
Index Terms—Fault currents, load flow analysis, optimization
methods, power generation planning. has been suggested that appropriately models new generation
as sources of power. Then, by solving the optimal power flow
(OPF) problem, rather than power flows, the optimal allocation
I. INTRODUCTION of new generation capacity to predefined connection points con-
sidering system and network constraints is determined. This is a
A BASIC axiom in power system operation states that gener-
ation must always meet demand. In the long term, demand
constantly increases; thus, new generation capacity is needed.
novel, unorthodox method of using OPF. Instead of using OPF
as an operation planning tool, it is used as a generation plan-
The equipment of the transmission and distribution network sets ning tool that determines optimal locations and capacities of
several technical limitations to the location and size of new ca- new generators that satisfy network constraints.
pacity or the expansion of the existing capacity. In addition, po- However, the conventional OPF formulation does not include
litical and environmental reasons often prohibit the expansion of the constraints imposed by protection equipment on expected
the existing network, especially over urban and suburban areas. fault levels. As a remedy, an iterative process has been sug-
Therefore, the efficient utilization of the existing network is not gested in [4] that converts fault level constraints (FLCs) to re-
only suggested for economy but also imposed by need. strictions on new capacity, and then, OPF reallocates capacity
Furthermore, the installation of new generation capacity according to these restrictions. The final solution respects both
brings the network closer to its operational limits. The main network/system constraints and FLCs. In this paper, we develop
network limits are the thermal limits, voltage limits, and fault a mathematical technique to convert FLCs to simple nonlinear
level limits. Especially, the introduction of new generation inequality constraints described by the existing OPF variables.
increases bus fault levels, thus, the expected short circuit cur- Henceforth, we can use the OPF directly to allocate new ca-
rents faced by protection. This may be especially applicable pacity with respect to both system and FLCs. Additionally, as
to renewable generation embedded in distribution networks. in [4], the allocation results can be easily adapted by indepen-
Currently, the significant cost of upgrading the protection dent system operators (ISOs) for new capacity planning mech-
anisms that direct new capacity to specific locations through fi-
nancial incentives. A 12-bus/15-line test case demonstrates the
advantages of the new method in terms of efficiency and speed
of convergence.
Manuscript received November 16, 2004; revised June 13, 2005. Paper no. The advantage of direct inclusion of FLCs is that during
TPWRS-00606-2004. the solution of the nonlinear OPF problem, shadow costs are
The authors are with the School of Engineering and Electronics, University
of Edinburgh, Edinburgh EH9 3JL, U.K. (e-mail: P.Vovos@ed.ac.uk). produced, connecting the overall optimum of the objective
Digital Object Identifier 10.1109/TPWRS.2005.856975 function with each constraint. This feature of the new method
0885-8950/$20.00 © 2005 IEEE
2126 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 4, NOVEMBER 2005

could be used in the design of a reinforcement planning mech- TABLE I


anism, which would assess the return of investments on both MODEL SUMMARY OF SINKS AND SOURCES
USED IN OPF FOR CAPACITY ALLOCATION
network and protection equipment. Indirect, iterative inclusion
of FLCs does not produce shadow prices connected with those
constraints.

II. USING OPF FOR NETWORK CAPACITY ANALYSIS


Reference [5] describes the formulation of OPF, which is tra-
ditionally used as an operating tool in power systems. In [4],
the authors presented a method of modeling new generation ca-
pacity, so that OPF can also be used to optimally allocate new
generation capacity. In other words, OPF is used as generation
expansion planning, rather than an operation planning tool, in
the sense that it determines optimal size and location of new
generating plants. Here, an outline of this methodology will be
presented.

A. Sinks and Sources


New generation capacities are simulated as generators with
quadratic benefit functions with negative coefficients. These
generators are connected to predetermined locations in the
network, the “capacity expansion locations” (CELs), with the
output of generators simulating the allocated capacity at the
CEL. Benefit function coefficients represent preferences for the where all lines, all generators, and all loads
allocation of new capacity between CELs. They may reflect a connected to bus . , , , and , , are the
benefit for transmission (or distribution) network owner from active and reactive power injected into bus . If the bus is also an
locating a new generator in a given CEL. If the coefficients are E/IP, then the complex power transferred from/to bus from the
the same for all locations, the method simply maximizes the external grid must be added to (for import) or subtracted from
total capacity of new generation to be connected to the network, (for exports) the above sum.
without expressing a preference for any particular location. The Proper operation of the power system equipment and quality
capacity allocation presented here does not consider different of supply requires the maintenance of bus voltages close to their
costs of power plants themselves. We assume that the allocation nominal values
is done from the point of view of the system operator who only (2)
assesses the network effects of different generation locations.
An alternative interpretation is that all new plants are of the where and are the lower and upper bounds of the
same type, so they have the same costs. bus voltage around the rated value.
Energy transfers from/to external networks are also simulated The installation of new generation capacity is limited by
as generators with quadratic benefit functions. We will refer to statutory regulations on quality of supply, environmental
them as export/import points (E/IPs). The coefficients of the concerns, planning policies, technological limitations, or
benefit functions are negative for exports and positive for im- system constraints imposed by stability, fault, or other secu-
ports. The outputs of the generators are negative when they rep- rity analyses. Here, only restrictions resulting from statutory
resent exports and positive when they represent imports. regulations and fault analysis are used
Finally, existing generation capacities are simulated as gener- (3)
ators with constant active power output, equal to their maximum
capacity, and given reactive power injection capabilities. Loads where and are the lower and upper bounds, respec-
are simulated as sinks of constant active and reactive power tively, for the generation output at CEL .
set at their annual peak values. The overall simulation model In most distributed generation applications, synchronous gen-
is summarized in Table I. erators perform automatic voltage regulation in power factor
control mode [6]. Thus, in order to simplify our analysis, we
B. Constraints assume that CELs have constant power factors
The amount of active and reactive power injected into any const. (4)
system bus must equal the amount withdrawn from it. The
complex power balance on the buses is formulated
This assumption holds for most distributed generation (DG)
installations that interface to the network through an inverter
[7]. However, in more general cases, the production of reactive
(1) power is more flexibly controlled or additional reactive power
VOVOS AND BIALEK: DIRECT INCORPORATION OF FLCs IN OPF 2127

sources are utilized (e.g., FACTS). Then, this restriction can be However, the original OPF formulation does not include the
relaxed providing higher generation capacities. constraints imposed by fault levels, such as the breaking ca-
The thermal capacity of a line sets a limit to the maximum pability or capacity of switchgear. Consequently, the conven-
apparent power (MVA) transfer tional form of OPF as a tool for network capacity analysis also
ignores the impact new generation has on fault levels, and, as it
(5) is proven by our test case results later on, network capacity may
where is the apparent power, and is the thermal limit be overestimated.
of line .
Each E/IP represents a physical connection to an external net-
IV. ITERATIVE INCORPORATION OF FLCS
work. The capacity of the connection sets a limit to the max-
IN CAPACITY ALLOCATION
imum amount of power that can be transferred to and from the
external network. Furthermore, in cases where the quantity of Reference [4] presented a three-step iterative process to con-
the exported or imported power has a significant impact on the sider FLCs in the final capacity allocation. It was termed fault
operation of the external grid, more conservative bounds than level-constrained optimal power flow (FLCOPF).
the connection capacity must be applied to limit the voltage rise In the first step, OPF allocated new capacity, ignoring fault
or drop at buses within the external network. These limits are levels. Then, fault analyzes determined which power system
expressed as equipment would exceed its operational specifications under a
possible fault at each system bus.
(6)
In the second step, a generation reduction optimization algo-
where for imports, and for exports. rithm (GROA) reduced new capacity at CELs in order to max-
In addition, we must provide the maximum reactive power imize total new generation capacity with respect to the FLCs
, the external network can feed into the system, and imposed by that equipment. The subtransient reactance of new
the minimum it can absorb generators was estimated as a function of their capacity
using the following formula:
(7)
(9)
C. OPF Objective Function
In a traditional OPF, the objective function is equal to the sum where function calculates the reactance of a generator with
of fuel costs. In a generation planning OPF, the objective func- size on the generator’s reactance base, and is the system
tion is equal to the total benefit from new generation capacity MVA base. The lower the new generation capacity, the higher
and expected exports/imports the subtransient reactance and the lower the fault current. The
opposite is also true.
However, the authors assumed that bus voltage patterns do
(8)
not change much during the optimization procedure of GROA.
Therefore, GROA did not directly set the new upper bounds for
where , are defined in Table I, is the number of CELs, capacity in OPF but pointed the direction that these bounds had
and is the number of E/IPs. to be reduced. In the last step, the new upper capacity bounds at
The minimization of (8) maximizes weighted capacity at CELs were estimated as a function of the bounds identified by
CELs and exports at EPs, due to the negative benefit function GROA and the last OPF allocation.
coefficients attached to them. This is done subject to (1)–(7), so At the next iteration, the OPF reallocated capacity, subject to
that the solution does not exceed the capabilities of the existing the new bounds. The iterative process converged to the optimum
network infrastructure. when there was no significant change of the capacity bounds be-
By selecting the appropriate coefficients, this objective func- tween iterations, and no FLCs were violated. However, the iter-
tion could contain the monetary/communal benefit (expressed ative nature of this method cannot guarantee global optimality
as negative cost) of the network planning authority from ex- for the final solution.
ploiting the capabilities of the existing network to absorb new
generation. Its minimization reflects the optimal allocation of
new generation capacity at CELs and the setting of energy trans- V. DIRECT INCORPORATION OF FLCS IN THE OPF
fers at E/IPs, which maximize this benefit.
In this paper, a method for the direct incorporation of FLCs in
OPF is presented. We develop a mathematical methodology to
III. IMPORTANCE OF CONSIDERING FLCS convert constraints imposed by fault levels to simple nonlinear
The equality and inequality constraints included in the OPF inequality constraints, described by the usual OPF variables.
formulation guarantee that the capacity allocation will not vio- Henceforth, OPF can be used directly to allocate new capacity
late any of the network or system constraints. Such constraints with respect to both network/system constraints and FLCs. This
are imposed by lines thermal limits, the active and reactive one-step optimization procedure improves the capacity alloca-
power capability of generators, and statutory regulations on bus tion properties when compared with the previous (iterative) ap-
voltage fluctuations. proach based on OPF and has important implications for eco-
2128 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 4, NOVEMBER 2005

nomic analysis of the final solution. The new approach was also a function of the same elements according to (11). Conse-
termed direct fault level-constrained OPF (D-FLCOPF). quently, the problem of direct inclusion of fault level constraints
in OPF focuses on the expression of as a function of new
A. Connecting Fault Level Constraints With the OPF Variables capacity.
FLCs refer to the operational limitations of protection equip- The Sherman–Morrison–Woodbury (S–M–W) formula [8],
ment (e.g., switchgear) during a fault. If the specifications of the [9] calculates the elements of the new inverse of a matrix after
protection equipment are not adequate to clear or isolate a fault, one element is modified, given the inverse of the initial matrix
then not only the equipment itself will be possibly damaged, but and the modification
the operation of a broader part of the power system will become
insecure. Generally, it is the magnitude of a fault current that is (12)
compared with the specifications of the protection equipment.
where is the element of the new inverse matrix ,
For instance, two basic specifications of switchgear are capacity
is the old inverse matrix, is the modification of element
and breaking capability. They both set limits to the magnitudes
of the initial matrix where , with
of fault currents that the switchgear can securely handle. There-
.
fore, we will focus our analysis on the magnitudes of fault cur-
Let us assume that is the impedance matrix of the ex-
rents, rather than their complex values.
isting power system (without any new capacity added), built el-
The magnitude of the expected fault current flowing ement-by-element or calculated as the inverse of the initial ad-
through line , must comply with the maximum allowed by mittance matrix . The reactance of a new generator at bus
the specifications of the switchgear at the terminal buses affects the diagonal element of the
and :
(13)
(10)
In order to include FLCs in OPF, a mathematical expression where is the reactance of the new generator.
linking the expected fault currents with the OPF variables must Equation (13) describes the modification of an element of the
first be found. These variables are the voltage magnitude initial matrix . Therefore, we can use (12) to calculate the
and angle of any bus and the real and reactive power elements of the new impedance matrix for a new generator
of any generator . The fault current flowing through a line with at bus with subtransient reactance
series impedance for a fault on bus is given by the equation
(14)

(11) Whenever a new generator is added at bus ,


similar numerical calculations to (14) directly give the impact
where , , are the voltages of buses , , . They are de- of the new reactance on the , with respect to the previously
scribed in magnitude and angle by the OPF variables , , modified from the addition of generator at bus :
where . , where ,
and are the elements of the impedance matrix . (15)
The is not directly connected with any of the OPF vari-
ables. If OPF is used in its conventional form to identify the most If is replaced in (15) with the estimated reactance
economic operating point of generation to cover the demand, given by (9), then
then generators have specific impedances, is constant and
includes the reactances of previously existing generators and (16)
Thevenin reactances of E/IPs (estimated as the reciprocal of the
short circuit capacity at the connection point). However, if OPF This way of updating the , each time OPF allocates new
is utilized to allocate new capacity, then the generators’ size is capacity at a bus, shares two important features. First, we have
not specific. is a function of new capacity, and so is mathematically formulated the impact new capacity has on the
and the expected fault currents. , since the S–M–W formula computes any element of the
Equation (9) estimates the subtransient reactance of new gen- new impedance matrix with respect to the size of new genera-
erators according to their capacity. While a change in a gener- tion. Second, we avoid a serious computational burden, espe-
ator’s reactance linearly affects one element of the admittance cially for largely interconnected networks consisting of thou-
matrix (the one connecting the bus of the generator with sands of buses and tens of thousands of transmission lines. There
the reference bus ), it nonlinearly alters all is no need to invert the new or building the ele-
elements of the . If FLCs are ignored, changes of ment-by-element from scratch each time we change the capacity
the reactances of new generators have no impact in the OPF op- allocation and want to estimate the impact on the expected fault
eration for the allocation of new capacity. In this case, OPF uses currents.
only the nondiagonal elements of the to solve the sequence We have managed to link new generation capacity to the ,
of power flows leading to the optimum. Conversely, since the hence to fault currents, but capacity is not a variable in OPF.
allocation of new capacity changes the and is a We suggest two ways to express capacity as a function of OPF
function of elements of the , the expected fault currents are variables. The reactive power capability curves of several types
VOVOS AND BIALEK: DIRECT INCORPORATION OF FLCs IN OPF 2129

and sizes of generators that will be possibly preferred by the in- • If , then
vestors can be collected from major manufacturers. From each
curve, the MVA rating of the generator can be extracted and a
new curve can be created, which will connect roughly
the MVA rating of new generators with their rated output
in megawatts (MW). Alternatively, the MVA rating of gen-
erators can be specified from their rated power and rated power
factor p.f. The rated p.f. does not change much between different
sizes and models (for most synchronous generators, the rated p.f.
is around 0.8). Therefore, a function of MVA ratings of genera- (21)
tors with respect to their rated MW output divided by a constant
typical p.f. can be created • If , then

(17)

In either case, (16) becomes a function of the rated power of


new generators
(22)

(18) where
is a variable in the OPF when it is used to allocate new =
generation capacity. In other words, we have managed to link
and= .
, a function of , to the OPF variables , (since
capacity reactive power ). Equation Superscripts , denote real and imaginary part of the respec-
(11) is now fully defined by the OPF variables. Given that fault tive complex variables.
levels are now a function of the OPF variables, they can be di- Similarly, we calculate , , and
rectly added to the nonlinear system constraints. Equation (10)
becomes , .
According to (11), fault currents are connected only with the
OPF variables describing the voltages across the line under con-
sideration and the faulted bus. Therefore
(19)

where is the number of system buses, and is the number of bus (23)
generators.

We calculate the derivatives of the fault currents with respect to


B. Derivatives of FLCs
real and reactive power of the virtual generators at CELs using
The solution of nonlinear programming problems, such as the chain rule
OPF, is generally reached using an iterative procedure that needs
a direction of search at each step toward the global optimum.
This requires the calculation of the derivatives of all nonlinear and
equality and inequality constraints with respect to each OPF (24)
variable. In this section, the derivatives for the new constraints where
only (FLCs) are presented. The way they are calculated is sum-
marized in the Appendix.
The derivatives with respect to bus voltages depend on fault
location.
• If and , then and

and

(20) and
2130 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 4, NOVEMBER 2005

, , is the apparent, real, and reactive power of gener-


ator and its admittance

C. Reducing the Number of FLCs


Fig. 1. FLCs reduction algorithm.
Theoretically, there are as many FLCs as the combination of
the number of buses that can accommodate a fault and the
pairs of fault breaking equipment at the end of each line : TABLE II
BENEFIT FUNCTION COEFFICIENTS USED IN TEST CASES
. In order to reduce the computational burden
caused by the large number of additional constraints during the
solution of the OPF, an algorithm (see Fig. 1) that identifies the
“active” constraints for the final solution was developed. It was
termed fault level constraints reduction algorithm (FLCRA).
The target is to create a list of binding FLCs for the OPF.
Binding FLCs are added to the existing set of OPF nonlinear
constraints, which consists of the active and reactive power bal-
ance on each bus. The nonbinding FLCs only limit the solution
space without affecting the optimum. B. Constraints
Switchgear imposing a constraint on the allocation of new Line 2–5 is constrained by a thermal limit of 14 MVA, 4–9 by
capacity is expected to be located near the CELs, since this is a thermal limit of 40 MVA, while all other lines are considered
where fault currents are expected to raise more. Usually, the to be unconstrained. It was assumed that the E/IP can exchange
number of system buses is much higher than CELs and their up to 100 MW with the external network without affecting its
nearby buses, making binding FLCs only a small fraction of secure operation. The external network is also capable of pro-
total nonlinear constraints. Thus, D-FLCOPF is expected to viding up to 60 MVAr of reactive power to the local network and
share similar convergence properties with the traditional OPF, consuming up to 50 MVAr. A hypothetical government policy
which is indifferent to system size. restricts the maximum allocated capacity to 50 MW at each
CEL. Finally, statutory regulations limit bus voltage fluctuations
to around the nominal values. Switchgear is tested only
VI. EXAMPLE for capacity adequacy, assuming 250 MVA at 11 kV, 1000 MVA
In order to demonstrate the new method (D-FLCOPF) and at 33 kV, and 3500 MVA at 132 kV, which are typical U.K.
compare it with the previous approach (FLCOPF), the same ratings.
test case will be used. A meshed distribution network has been
C. Simulation Model
chosen as a test case, as a more general example of a distribu-
tion or subtransmission network. New generators and exports/imports to external networks are
modeled as described in Table I. The coefficients of the ben-
A. Topology efit functions attached to each virtual generator simulating the
CELs, and the E/IP are shown in Table II. The performance of
The 12-bus 14-line network presented in Fig. 2 has three the two methods was examined with and without preferences
available CELs at buses 1, 10, and 11. It also has an E/IP to expressed for the allocation of new capacity at CEL 1.
an external network at bus 12. A 15-MW generator is installed
on bus 5. It can consume or provide up to 10 MVAr of reactive D. Assumptions About Capacity Expansion Locations
power. The network has a common rated bus voltage level at
In order to simplify our analysis, we assumed that new gener-
33 kV, except for the CEL buses, which have a rated voltage of
ators at CELs have constant lagging power factors set at 0.9 (see
11 kV and the E/IP bus at 132 kV.
Section II-B). It was also assumed that their internal subtran-
The CEL buses connect to the network through 30-MVA
sient reactance increases proportionally to their size between
transformers with fixed taps. The E/IP bus connects through
and on the generator reactance base
a 90-MVA transformer with automatic tap changer, which
[10]. Generators are considered to be small when their capacity
regulates the voltage within a range of the rated voltage at
is well below MVA. An example of how this
the low voltage side with a tap range around the nominal
assumption can be mathematically approximated by (25), which
tap ratio. The electric characteristics of transformers and lines
is graphically presented in Fig. 3
are presented in the table next to the network topology in Fig. 2.
Loads consuming constant complex power on buses 1, 3, 5, 6,
8, 10, and 11 have been assumed. Their size is depicted in the
same figure. (25)
VOVOS AND BIALEK: DIRECT INCORPORATION OF FLCs IN OPF 2131

Fig. 2. Twelve-bus 14-line test case and the table of transformer/line characteristics.

TABLE IV
FLCOPF AND D-FLCOPF CAPACITY ALLOCATION

TABLE V
CONVERGENCE ATTRIBUTES OF FLCOPF AND D-FLCOPF

Fig. 3. P.u. reactance of new generators with respect to their MVA base.

TABLE III
ALLOCATION OF NEW CAPACITY IGNORING FAULT LEVELS
Table IV presents a summary of the capacity reallocation,
which considers FLCs, as it was performed by FLCOPF and
D-FLCOPF for both cases concerning preferences. Table IV
also includes the total negative cost (benefit) induced by each
allocation, resulting from the minimization of (8). For the cal-
culation of total benefit, besides the expected exports in MW,
The reactance from the generator’s MVA base was converted the MW values of new capacity are used. They can be obtained
to the system MVA base MVA using (9). The p.u. from the respective MVA values of the same table under the
reactance of the new generator with respect to its capacity is assumption that generators operate at 0.9 lagging power factor
given by (see Section VI-D).
Table V summarizes the number of iterations and total time
(26) (for a 1.7-GHz CPU) needed from each method to converge.
In order to compare the performance of the two methods in
terms of total connection capacity and benefit, impact of pref-
VII. RESULTS erences and convergence the results presented in Tables III–V
Initially, the external network provides 18 MW and were used.
6.2 MVAr, while the generator at bus 5 operates at full ca-
pacity (15 MW, 10 MVAr) to cover the existing demand. A. Connection Capacity and Benefit
The initial allocation of new generation capacity by the OPF, When there are no preferences, the allocation mecha-
which ignores FLCs, is shown in Table III. This allocation re- nism maximizes the possible total new capacity. According
sulted in the capacity violation of the switchgear connected on to Table IV, D-FLCOPF allocates more total capacity than
the primaries of the transformers 1–2 and 10–4. FLCOPF by MVA. The total benefit is also
2132 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 4, NOVEMBER 2005

Fig. 4. Meaning of “Iterations” for (a) FLCOPF and (b) D-FLCOPF.

higher by . Evidently, D-FLCOPF VIII. FUTURE WORK—INTEGRATION IN AN INFRASTRUCTURE


traces the potential connection capacity better than its iterative REINFORCEMENT PLANNING MECHANISM
alternative.
If a nonlinear programming method is used to solve the OPF,
such as sequential quadratic programming, then shadow costs
B. Impact of Allocation Preferences on Total New Capacity are produced, connecting the overall optimum of the objective
function with each constraint [11]. They describe how much the
According to Table IV, FLCOPF allocates objective function would increase if the respective constraints
MVA less when a preference is expressed for CEL 1. were marginally relaxed. When OPF is utilized for generation
Similarly, D-FLCOPF allocates MVA less capacity allocation, a higher value of the objective function
for the same preference. Evidently, total capacity has a looser means higher total new generation capacity. Therefore, shadow
connection to preferences when the allocation is performed by costs could be used as economic signals in an investment plan-
FLCOPF. ning mechanism, since the constraints reflect the limitations of
The reduction of total new capacity is connected to the ef- the existing infrastructure.
fect preferences have on the OPF objective function. Capacity An iterative process would gradually relax the best (for each
on CEL at bus 1 induces higher benefit per MVA than other iteration) value/money investment, and OPF would reallocate
CELs and exports at E/IP. This explains the lower total new ca- capacity. An additional cost should be attached to each mar-
pacity but almost unaffected total benefit from the allocations ginal constraint relaxation, reflecting the cost of reinforcing the
of FLCOPF and D-FLCOPF when preferences are expressed. existing infrastructure in reality. Such a mechanism could pro-
Therefore, there is a tradeoff between higher allocation of ca- duce an investment/capacity graph and reinforcement plans for
pacity at the “preferred” CELs and total new capacity. the existing infrastructure.
This is the subject of future work. However, D-FLCOPF
groups FLCs together with the system constraints. Therefore,
C. Convergence the impact of FLCs on capacity allocation is comparable to the
one imposed by the rest of the system and network constraints.
Even though, “iterations” serve a different function in each of This overcomes the limitations of the FLCOPF, which incorpo-
the two methods (see Fig. 4) they provide a means of compar- rates FLCs indirectly and respective investment signals cannot
ison for the speed of convergence. They are independent from be produced, at least at the same scale with system constraints.
the mathematical tools or the computational power used to solve
the optimization problem in each case. Conversely, total time in
seconds gives a more quantitative scale for the actual conver-
IX. CONCLUSIONS
gence time of each algorithm.
D-FLCOPF converges much faster than FLCOPF in any case. This paper presents a new method (D-FLCOPF) for the
According to Table V, D-FLCOPF needs only a few seconds incorporation of FLCs in the OPF as an assessment tool for
and iterations to converge to the final solution, while FLCOPF the capacity of a network to absorb new generation. No new
requires several hundreds. variables are introduced in the OPF formulation, and FLCs are
Preferences have a significant impact on the performance of converted to simple nonlinear (inequality) constraints. Most
FLCOPF. The allocation preference for CEL 1 reduces both common OPF-solving engines already have the computational
time and number of iterations. On the contrary, with or without capacity to handle numerous nonlinear constraints, such as
allocation preferences over CELs, D-FLCOPF converges in less the ones described by the power balance equations on buses.
than 3 s. The FLCRA (see Section V-C) considers all active Therefore, once FLCs are converted to nonlinear constraints
FLCs in the first iteration. Since only two additional constraints described by OPF variables, they can be directly introduced to
(one for each violated switchgear capacity) are added in the any optimization process performing the OPF. Alternatively,
original OPF, the convergence time changes insignificantly: It FLCOPF can be implemented using widely available tools to
is about 1.5 s for both the initial OPF and the final D-FLCOPF. solve the OPF and fault analysis of the capacity allocation
VOVOS AND BIALEK: DIRECT INCORPORATION OF FLCs IN OPF 2133

process. Thus, engineers have the freedom to choose them- and . Combining
selves the implementation of an important part of the overall (32)–(34)
mechanism.
The underperformance (as proven from the test case results)
of the earlier iterative approach (FLCOPF) is based on the indi-
rect incorporation of FLCs by converting them to restrictions on
new capacity. Therefore, the one-step optimization procedure of (35)
D-FLCOPF improved the capacity allocation properties of the • If , then
previous approach based on OPF. Finally, the direct incorpo- and
ration of FLCs in the OPF constraints allows the optimization . Com-
procedure to produce shadow costs for those constraints as well. bining (32)–(34)
This was not possible with the previous method, since the addi-
tional constraints were converted to restrictions on new capacity.
Further research needs to be done toward the full exploitation of
the shadow costs as economic signals in a reinforcement plan-
ning mechanism. (36)
• If , then
APPENDIX and . Combining
(32)–(34):
In this section, the authors briefly present the calculations for
the derivatives of FLCs, with respect to the OPF variables.
Voltages , , and in (11) can be analyzed as

(27)
(37)
(28) Similarly, we calculate , , and
Then, (11) from (27) and (28) becomes , .
If is the admittance of generator , then
(29)

similarly to (38)
where

(30) (39)
and
(40)
(31)
If and are the real and imaginary parts of a function
We use the chain rule to calculate the desired derivatives
and
and (41)
In order to calculate , we assume a small change
(32) to the generator’s admittance . According to the S–M–W
formula, any element of the new inverse matrix is
The first derivative of each product is calculated using (29)
given by the equation

(42)
(33)

If FSF’ is the FSF after the change , then


The second factor in the products of (32) is easily obtained as
follows:
for
Since and
(43)
(34) where
The analysis differentiates according to the fault location. = ,
• If and , then and = , , .
2134 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 4, NOVEMBER 2005

According to the definition of derivation [3] K. Nara, Y. Hayashi, K. Ikeda, and T. Ashizawa, “Application of tabu
search to optimal placement of distributed generators,” in Proc. IEEE
Power Eng. Soc. Winter Meeting, 2001, pp. 918–923.
[4] P. N. Vovos, G. P. Harrison, A. R. Wallace, and J. W. Bialek, “Optimal
power flow as a tool for fault level constrained network capacity anal-
(44) ysis,” IEEE Trans. Power Syst., vol. 20, no. 2, pp. 734–741, May 2005.
where , , , are calculated for the current generator set-up [5] H. W. Dommel and W. F. Tinney, “Optimal power flow solutions,” IEEE
, so is a constant. Using (39), (40), (44) in (38) Trans. Power App. Syst., vol. PAS-87, pp. 1866–1876, 1968.
[6] T. W. Eberly and R. C. Schaefer, “Voltage versus VAr/power-factor reg-
ulation on synchronous generators,” IEEE Trans. Ind. Appl., vol. 38, no.
6, pp. 1682–1687, Nov./Dec. 2002.
[7] S. R. Wall, “Performance of inverter interfaced distributed generation,”
in Proc. IEEE/Power Eng. Soc. Transm. Distrib. Conf. Expo., 2001, pp.
(45) 945–950.
[8] J. Sherman and W. J. Morrison, “Adjustment of an inverse matrix corre-
If , , is the apparent, real, and reactive power of sponding to changes in the elements of a given column or a given row
of the original matrix,” Ann. Math. Statist., vol. 20, p. 621, 1949.
[9] M. A. Woodbury, Inverting Modified Matrices. Princeton, N.J.:
Princeton Univ. Press, 1950.
[10] J. M. Fogarty, “Connections between generator specifications and funda-
mental design principles,” in Proc. IEEE Int. Electric Machines Drives
Conf., Cambridge, MA, 2001, pp. 51–56.
(46) [11] P. N. Vovos, J. W. Bialek, and G. P. Harrison, “Optimal generation
capacity allocation and network expansion signalling using OPF,” in
Proc. 39th Int. Universities Power Eng. Conf., Bristol, U.K., 2004, pp.
1327–1331.
(47)

(48)
Panagis N. Vovos was born in Athens, Greece, on October 23, 1978. He re-
Using (44), (46), (47) and (44), (46), (48), we calculate ceived the Diploma degree from the Electrical and Computer Engineering De-
partment, University of Patras, Patras, Greece, in 2002. He is currently working
toward the Ph.D. degree from the University of Edinburgh, Edinburgh, U.K.
His special fields of interest include cross-border congestion management,
and efficient capacity allocation, and enhanced OPF.
(49)

REFERENCES Janusz W. Bialek received the M.Eng. and Ph.D. degrees from Warsaw Univer-
[1] K.-H. Kim, Y.-J. Lee, S.-B. Rhee, S.-K. Lee, and S.-K. You, “Dispersed sity of Technology, Warsaw, Poland, where he worked until 1989. From 1989
generator placement using fuzzy-GA in distribution systems,” in Proc. until 2001, he was with the University of Durham, Durham, U.K.
IEEE Power Eng. Soc. Summer Meeting, 2002, pp. 1148–1153. Currently, he holds the Chair of Electrical Engineering at the University of
[2] T. Griffin, K. Tomsovic, D. Secrest, and A. Law, “Placement of dispersed Edinburgh, Edinburgh, U.K. His main research interests are in the range of is-
generation systems for reduced losses,” in Proc. 33rd Ann. Hawaii Int. sues connected with liberalization of electricity supply industry and in power
Conf. Syst. Sci. , 2000, pp. 1–9. system dynamics. He has authored a book and about 70 research papers.

You might also like