Operational Constraints

Optimization in Planning and Operation of Electric Power Systems. Physica-Verlag (Springer), Heidelberg, May 93, pp.
217-264
Power System Models, Objectives and Constraints in Optimal

Power Flow Calculations
Rainer Bacher
Swiss Federal Institute of Technology (ETH)

CH-8092 Zürich, Switzerland
1
Optimization in Planning and Operation of Electric Power Systems. Physica-Verlag (Springer), Heidelberg, May 93, pp. 217-264
1 Introduction
The electric power system is defined as the system with the goal to generate and to transmit
the electric power user via the transmission system to the electric power end users. In order to
achieve a practically useful power system engineers have come up with principal components
and characteristics of the power system: These components are pieces of hardware like high
voltage transmission lines, underground cables, high voltage transformers, nuclear, hydro or
hydro-thermal power stations, compensators, etc.
Electric power can be provided in direct or alternating current form, however, in the be-
ginning of the 20th century alternating current has been determined as the more economical
and technically feasible solution. An electric power system where the sources are alternating
currents is principally of a dynamic nature, i.e. all current and dependent quantities like vol-
tage and electric power itself are varying over time. A physical law exists among these three
quantities:
P ower(t) = V oltage(t) ∗ Current(t) (1)
In (1) the variable (t) refers to time.

From a control point of view it is the electric power consumption of end users which is
hard to control: Today in most civilized countries the end user consumes the electric power
with almost no restrictions. Fortunately, due to a regular daily behavior of the businesses and
households electric power consumption follows more or less regular patterns. Due to the regular
behavior it is possible to use some kind of prediction method to get good approximative values
for power consumption at any future time.
Electric power is of such a nature that whatever electric power is needed must be generated
by generators at any time t. This generated power includes the power losses in the resistances
of the electrical transmission network.
Thus the approximate total sum of generated power is always defined for any given total
end user power consumption. Since the total generation can be produced by a high number of
generators, a power system could physically be operated in a huge number of states. However,
many of them are not tolerable due to physical or operational constraints. Examples are extreme
material usage due to e.g. excessive power flow through a cable, excessive voltage at an electrical
busbar, etc.
Quality standards have quickly been identified and standards have been formulated which
are valid world-wide with minor variations. These standards say that the power system opera-
tion must be safe, reliable and economical. Safety relates mainly to life threatening aspects
for people involved in the usage of the electric power. Reliability standards determine the total
time per year during which an electric power end user can be without electric power. Today, in
the western world, the standard for electric power end user availability at any time during the
year is very high at almost 100%. Since the generation and transport of power is very costly
and since this high cost must mainly be paid by the electric power consumers the importance
of economical aspects related to all aspects of power system operation are obvious.
This very complex dynamic behavior, the wide geographical distribution of the power sy-
stem and also the high cost and economical risk of planning, building and maintaining power
2
system components and to achieve the above mentioned power system goals have led to a most-
ly regional organization of the power system. The power system apparatus and power system
operation for these regional areas (’the power system control areas’) are coordinated by electric
utilities which are mostly privately held or mixed privately held / state owned companies in
the western countries.
With the size of the interconnected power systems getting larger and larger in the early 50’s
manual ad hoc operation of the power system even with local control mechanisms has become
harder and harder and high level, often hierarchical control systems have been introduced into
the power system to have an automatic or semi-automatic operation of parts of the power
system satisfying new standards.
Since almost all major power system control areas are interconnected with mainly high
voltage transmission lines operating rule standards have been established in Western Europe
(UCPTE) and also in the U.S.A. around 1960. The standards are mainly related to the quality
of the power system frequency (Europe: 50 Hz, U.S.A.: 60 Hz) and also to the control of power
or energy contract based inter-utility transfers.
This automatic high level control mechanism is called Load Frequency Control (LFC) and
has quickly been established as a standard control mechanism to hold both the frequency and
also the inter-utility power transfers within given tolerances. LFC establishes some standards
for the cooperation of even foreign electric utilities but still allows each utility to operate and
maintain its power system part autonomously.
The LFC concept, respecting autonomous power system area control is accepted today by
all participating utilities. It is within this framework where the individual utility must satisfy
the goals of the power system: Safe, reliable and economic operation.
The above keyword ’safety’ meaning, has been discussed before. ’Reliability’ and ’Econono-
my’ can be realized or enforced by computer controlled operation. It is obvious that the result
of some optimization could bring the utility nearer to achieve the above goals. The Optimal
Power Flow (OPF) is one very important computer tool to help the operator achieve this relia-
ble and economic power system operation for which the utility has autonomous responsibility.
However, the OPF is only one of several tools within a real-time based hierarchical computer
assisted control system as will be briefly discussed in the next section.
Much literature can be found about the characteristics and modelling of the power system.
A good overview of the data acquisition and the generation control system is given in [1] and
[2]. These two papers give the reader a brief overview of how real-time data is transferred from
the power system to the computer where it is both displayed and used as input for many kinds
of algorithms.
2 The role of the optimal power flow (OPF) computation wi-

thin the overall power system control
Due to the complexity and very high degree of freedom of power system operation hierarchical
models have been established which are valid for different time frames. Usually, the output
of the higher level model based computations can be taken as input for the next lower level
model, etc.
3
Within each model type different optimization problems can be formulated. The distinction
from model type to model type is both the time frame for which the models are valid and also
the resulting solution algorithms.
By decoupling the power system into hierarchical models for different time frames global
mathematical optimality is lost. However, practical experience has shown that the chosen time
based hierarchical characterization leads to quite a practicable approach which is probably not
too far from a global optimum.
From an optimization point of view two different steady state power system models exist:
Category 1: In this category the models are derived in such a way to be applicable to power
system simulation from approx. 24 hours to 10 years from actual time. Algorithms based on
category 1 models are executed in a real-time environment at a rate of about one hour or slower.
Within this time frame the goal is to determine the least expensive subset of power generators
for each discrete time step based on uncertain and predicted total system load determined by
some method which constitutes the only hard equality constraint per discrete time step. Many
other inequality type constraints for the individual generation units are incorporated. The
objective function is usually to minimize the sum of the cost of all generators for all discrete
time steps.
The main mathematical solution problem comes from the mixed discrete, continuous varia-
ble problem formulation and also from the fact that inequality constraints for variable changes
from time step at time t to those from time step t + ∆t exist.
The problem area of category 1 is often called Optimal Unit Commitment (UC). Different
problem statements, mainly depending on the chosen time horizon and the discrete time steps
are possible, however, all have in common, that the power transmission system, i.e. the trans-
mission lines, the transformers, etc. are not modelled. In summary, all computations within
this category 1 have in common
• that the total end user power consumption is predicted by some method and is given
from now until some given future time in discrete intervals.
• that the power system active power transmission losses for each network state in the
future are zero or simply predicted by some heuristic approaches and
• that no power transmission based power system limits are violated.
Category 2: Model validity: 5 minutes to 1 hour from actual time into the future. All dynamic
quantities are assumed to be in sinusoidal wave form with constant frequency and constant wave
peak amplitude (a network state called ’steady state’) which allows the application of complex
numbers to the solution of the problem. The goal is to determine the optimal scheduled values
or control set points for a set of generators and other non-generator controls such that the
power system with dynamically varying end user power consumption is continuously operated
in or towards a reliable and economic network state.
The main algorithmic problem is given by the need to incorporate the model of the power
transmission system with elements like transmission lines, underground cables, transformers,
shunts and associated operational constraints. Also, simple models for the generators and loads
must be used. This is in contrast to the models used in category 1 where the generation and
load are the only power system elements and variables respectively.
4
The aspect of load behavior uncertainty in algorithms of this category 2 is much smaller
than in the algorithms of category 1, due to the relatively short period in the future for
which the load must be predicted. Today, due to the complexity given by the need to model
the operational constraints of the power transmission system and also due to the additional
complexity of the immediate application of the optimization result as control means, many
algorithms of this category 2 neglect the varying load behavior at all. In order to compensate
for this, a different, less accurate power system model which incorporates the changing load
behavior is used in some utilities and applied in real-time more often, e.g. at intervals of 1
minute or even faster.
The OPF which is the main theme of this paper, today must be put within
category 2. It is the function in an Energy Management System that schedules the
power system controls in some optimal way being at the same time constrained by
the power flow network model and power system operating limits.
The modelling challenges are mainly coming from two problem areas:
• Due to the necessity to model the transmission system and its operating constraints for
sinusoidal waves with constant frequency and amplitude (which means usage of complex
variables) the dimension of the mathematical formulation is enormous. This and the
non-linearity of the underlying power system model results in an optimization problem
statement whose solution is not at all straightforward.
• The fact that the optimization output should be applicable to the controllers of the power
system in real-time mandates very fast and robust algorithms solving a problem based
on models which represent the real power system behavior as closely as possible.
In this paper an OPF problem statement is derived independent on the actual computer
solution and the fact that the result must be transferred in real-time to the controllers. Thus
emphasis is given to the facts that the OPF formulation
• is derived in a form to get a model of very high quality of the power system for a steady
state power system;
• is precise in mathematical terms;
• satisfies most constraints given from an operator controlled real-time application of the
OPF output.
This is in constrast to the actual solution processes for the OPF problem formulation which
are not presented in this paper. However, it must always be kept in mind, that the solution
process itself can lead to necessary simplifications of the original problem formulation. Also it
is very hard if not impossible to prove that a closed form mathematical solution process with
an optimal solution exists for the given OPF problem formulation. It is known that at least
for certain defined subproblems given in the following sections, optimal solutions and clear,
straighforward solution algorithms exist (see paper by H. Glavitsch).
During the past 20 years much literature has been written about OPF problem formulations
and related solutions algorithms. In the appendix literature is cited which is mainly related to
the modelling of the power system in connection with the OPF (see [3] ... [7]). Most of these
papers have detailed literature references which are not repeated in this paper.
5
3 The power flow model as equality constraint set of the OPF

3.1 Basic model assumptions
3.1.1 Power system loads
The OPF is based on a power system model for category 2, see the preceding section. The end
user power consumption is varying permanently over time and this has to be (or should be)
considered in any power system model, especially when applying the model based algorithmic
output back into the power system via control mechanisms. As mentioned in the preceding
section the validity for the OPF model can be found in the 5 minute to approx. 1 hour time
frame. The change of load within this time frame can result in different power system states
at time t and time t+5 Min., t+15 Min. or t+1 hour which again could lead to quite different
optimal controller settings. Two possibilities exist to handle this situation, knowing that the
OPF is executed only once during the chosen time interval:
• Either the utility moves the controller setpoints only once at every discrete time interval,
i.e. after the OPF algorithm output is obtained, or
• an additional mechanism with the incorporation of measured or accurately predicted end

user power consumption is used to adapt or recompute the optimal controller settings at
faster intervals than the main OPF execution rates.
The first approach leads to a non-optimal or only near-optimal network operation. The
second approach seems to be better from an optimal operation point of view, however, a more
complicated, probably hierarchical algorithmic usage is needed.
In conclusion electric end user power consumption is discretized at pretermined intervals.
The individual end user power consumption at a discrete time step is called a load in OPF
models and computations.
3.1.2 Power system model: Steady state and symmetric power system operation
All electric quantities power system like current, power and voltages are quantities varying over
time. For the power system model it is assumed that
• All currents, powers and voltages are quantities with sinusoidal wave form with constant
amplitude. This behavior is called steady state power system operation and leads to
the very important fact that the power system states can be modelled with complex
variables. The complex variables for voltages, currents and powers are transformations of
the corresponding steady state power system quantities.
• The power flow model of the three-phase power system assumes so-called (phase-) sym-
metric power system operation. Without going into details this assumption allows the
modelling of the power transmission system with electric two-ports.
Note that both above assumptions are only models of the real power system. Thus model
errors (errors with respect to the real-world power system) are introduced. However, practical
experience shows that this assumption is valid for a wide variety of cases within category 2.
6
3.1.3 Power system model: Generation, load and the transmission system
The main components of the power system which must be modelled under the assumption
of known or predicted loads at geographical locations and under steady state and symmetric
power system operation are the following:
Overhead transmission lines, underground cables, transformers, shunt elements.
Each of these passive power system transmission elements is modelled as a two-port mathe-
matical element, situated between electrical nodes i and j (shunt elements are only associated
with one electrical node i). Thus the power system model is composed of many passive ele-
ments placed between nodes i and j (i 6= j) yielding a network of branch elements. The special
properties of this passive network are summarized as follows:
An electrical node (called from now on ’node’) is connected via passive elements only to
about 2 or 3 other nodes (on the average). The resulting connectivity matrix is very sparse:
Matrix element (i,j) is non-zero only if there is a connection between i and j.
Since the power system is divided into power system control areas, each operated by an
electric utility, each utility can model its own control area with high accuracy.
Often the individual utilities split the model of their control area into submodels to reduce
the size and the complexity of the power system model. This splitting is possible especially
in the lower high voltage levels (e.g. below 110 kV or 60 kV), due to the organization of the
electric transmission system: Often the lower voltage network parts are only connected at one
point via a transformer to the highest voltage level network parts. Thus a split into parts at
these transformers is possible and does not lead to very high modelling inaccuracies.
At the highest voltage level, however, other circumstances exist: Here, power system control
areas of different utilities are interconnected. The modelling problem is obvious: Due to the high
degree of connectivity of highest voltage networks, each individual utility should also model at
least parts of the highest voltage levels and associated network elements of neighboring utilities.
However, especially in the network parts representing the highest voltage levels, ownership and
mostly profit-based economic reasons often prevent one utility to know the exact data of the
network components of the neighboring utilities.
Thus assumptions about passive network elements of mainly the highest voltage levels of
the neighboring or even more distant electric utilities must be made by each individual utility.
It must be emphasized that modelling only the data of the highest voltage levels of the own
control area usually leads to very inaccurate simulation results which might not be useful in
practice. This modelling problem has been recognized by the utilities and data exchange for the
most sensitive power system elements and their status (on-off) has been established. Thus for
the power system model of the transmission network operated by a utility, it can be assumed
that the model comprises at least parts of the neighboring utilities. Some theoretical approaches
which determine some criteria for the neecessary size of neighboring areas exist today, although
not in perfect form. Often they are based on heuristic assumptions.
For power system models (category 2) it must be assumed that the chosen network comprises
an expanded geographical area and that the most important power system elements of the own
area and neighboring areas are modelled with high accuracy. In this text, the term ’network’
refers to this expanded power system network, at least for highest voltage levels networks.
Depending on the size of the utility and the data-exchange related cooperation between
7
neighboring utilities, networks can have varying sizes, starting at about 100 nodes up to 5000
or more nodes with about 150 to more than 10000 passive network elements, each modelled by
two-ports, derived in the next subsection.
Generators: They are modelled individually, i.e. at their geographical location at a node
of the electrical network. For OPF models it can be assumed that it is known which ones of the
individual, geographically distributed generators are ’on’ and which ones are ’off’. Only those
generators which have a status of ’on’ can deliver power into the network and are important in
the category 2 models. Certain new modelling aspects are introduced per generator as compared
to the category 1 model, such as constraints on the upper and lower reactive generator power
(to be defined later in this paper) and upper and lower voltage magnitude levels.
Loads: It is assumed that loads are modelled individually at their geographical location at
a node of the modelled network. If passive elements for subareas of the power system control
area are not modelled or if they are split apart to be treated in separate models (discussion
see above), the individual loads of such an area are assumed to be collected together at one
precisely known node. Also, it is assumed that the power values for all the loads are known,
either because they are measured precisely or because they are predicted by some method.
DC-Lines: DC-Lines and associated control equipment represent important power system
elements in certain power system control areas. A detailed model description is not given in this
paper. However, a simple model for a DC line with given MW flow at each end is a generator
with given MW. The generator model for OPF computations is described in detail in this
paper.
3.2 Power flow: Mathematical model of the power system for steady state
simulation
3.2.1 Passive power system elements
Branches
Branches are passive network elements which can all be modelled by the same type of
two-port equation, given below. Branch elements always refer to two different nodes i and j:
For each branch-element a relationship between the current, the voltage and the line para-
meters, all in complex quantities, exists:
#   #
I el−i−j i y el−i−j yel−i−j Vi
= ii ij  (2)
I el−i−j j y el−i−j
ji
yel−i−j
jj
Vj
(2) is the two-port equation for a branch in the model for category 2. The meaning of the
variables is summarized in the appendix of this paper.
Branch elements can be subdivided into two major subcategories: Lines and transfor-
mers:
Lines can be categorized into two main classes:
• Overhead transmission lines and
• underground cables.
8
The four complex matrix terms of (2) for a line are computed with the line parameters
(variables with capital Y ) as follows:
y el−i−j:Line
ii
= Y el−i−j
iio + Y el−i−j
ij
el−i−j
y el−i−j:Line
ij
= −Y ij
(3)
y el−i−j:Line
ji
= −Y el−i−j
ij
el−i−j el−i−j
y el−i−j:Line
jj
= Y jjo + Y ij
Transformers are passive network elements which allow the transformation of one voltage
to another (or some even to two or even three other voltage levels). Transformers represent a
(rather small) subset of the branch-elements defined with (2).
The four complex matrix terms of (2) for a transformer are computed as follows:
el−i−j el−i−j
y el−i−j:T
ii
raf o
= Y iio + Y ij
el−i−j:T
y ij raf o = −t el−i−j Y el−i−j
ij
raf o = −(tel−i−j )∗ Y el−i−j (4)
y el−i−j:T
ji ij
y el−i−j:T
jj
raf o
= (|tel−i−j |)2(Y el−i−j
jjo + Y el−i−j
ij )
The elements denoted by capital Y in (3) and (4) can be assumed to be numerically known
at precise values for each line or transformer from node i to another node j (although this is
not perfectly true due to data uncertainties in the range of plus or minus 10 %). This data
uncertainty comes from the fact that the variables are dependent on many effects which are
hard to quantify precisely. For example, a complex geometry of the transmission line tower,
earth resistance dependency on ground condition which again is weather dependent, geometric
construction of the transformer, etc.
The complex transformer-related variable tel−i−j (t: tap) represents the measure for varying
the transformer two-port parameters which in the power system allows to have a variable voltage
relationship of side i to side j of the transformer.
In practice, t represents a discrete control of the power system. However, in OPF algorithms,
t is taken as a continuous variable within upper and lower bounds and is set to the next
practically possible discrete step after the optimization.
Shunts
Shunts are passive network elements which can all be modelled by the same type of equation,
given below. Shunt-elements always refer two one node i:
For each shunt-element a relationship between the current, the voltage and the shunt pa-
rameters, all in complex quantities, exists:
i h i i
I el−i−o = yel−i−o
ii
Vi (5)
(5) is the equation for a shunt in the model for category 2.

The complex network element of (5) for a shunt is computed as follows:
y el−i−o:Shunt
ii
= sel−i−o Y el−i−o
iio (6)
9
The element with capital Y in (6) can be assumed to be numerically known at precise values
for each shunt at node i. However, also here, data uncertainties exist for the same reasons as
given for branch elements.
This data uncertainty, however, is not considered in most OPF models. Thus all variables
written with capital Y can be assumed to be precisely known.
The real variable sel−i−o represents the measure for varying the shunt admittance. Practi-
cally this must be seen as a measure indicating how many individual shunts of a shunt bank
at a node i must be switched in and how many are in an ’out’-status. Thus, in practice, s
represents a discrete control of the power system. However, in OPF algorithms, s is taken as a
continuous variable within upper and lower bounds and is set to the next practically possible
discrete step only after the optimization.
3.2.2 Kirchhoff-law: All currents at a node must add up to zero
A fundamental law - one of the Kirchhoff laws - says that all currents flowing into an electrical
node must sum up to exactly zero. This together with conventions on the direction of currents
is represented in Fig. 1.

Branch to node k
I:

Generator
-
el−i−k i
G
I Gi
- Branch to node l
I el−i−l i
L
I Li XXXXXXXXIXz
XXXXX
el−i−mi
Load
?I
Branch to node m
el−i−o
Node i Shunt
Abbildung 1: Currents at a node i: Conventions
Note the conventions for the generator currents I Gi , going into the node, the load currents
I Li , the shunt currents I el−i−o and the branch currents for a branch between nodes i and j
computed at node i I el−i−j i leaving the node i.
For these currents the Kirchhoff Law is as follows:
X
N
I Gi − I Li − I el−i−j i − I el−i−o = 0 ; i = 1 ... N (7)
j=1
In both (7) and Fig. 1 and also throughout the rest of this paper the following assumptions
are made:
• N represents the total number of electrical nodes
10
• All generators at a node i are summarized in one generator (I Gi ).
• All loads at a node i are summarized in one load (I Li ).
• All shunts at a node i are summarized in one shunt (I el−i−o ).
• All parallel branches from node i to node j are summarized into 1 branch element with
one two-port equation (I el−i−j i and I el−i−j j ).
• If there is no generator at a node i then I Gi = 0.
• If there is no load at a node i then I Li = 0.
• If there is no shunt at a node i then I el−i−o = 0.
• If there is no branch-type connection between nodes i and j then I el−i−j i = 0 and

I el−i−j j = 0.
With (2), (5) and (7) a mathematical relationship between the nodal voltage related va-
riables V (node-related) and the element related current variable I Gi (generator-related), I Li
(load-related), I el−i−o (shunt-related) and I el−i−j i (branch-related) is given.
3.2.3 Power - voltage - current - relationship

For every network element (generator, load, branch, shunt) relationship between the complex
power (S), the complex element-related voltage (V ) and the complex element-related current
(I) is valid:
Generators
S Gi = V i I ∗Gi ; i = 1 ... N (8)
Loads
S Li = V i I ∗Li ; i = 1 ... N (9)
Branch-Elements
S el−i−j i = V i I ∗el−i−j i ; i-j = all branch elements (10)
S el−i−j j = V j I ∗el−i−j j ; i-j = all branch elements (11)
Shunts
S el−i−o = V i I ∗el−i−o ; i = 1 ... N (12)
I.e. the complex power of an element is always the product of the complex voltage times
the conjugate complex current of the corresponding element.
11
3.3 Mathematical formulation of the various equality constraint sets

3.3.1 General characteristics
The equations (2) to (12) written for all corresponding network elements, represent the set of
equality constraints for the power flow model. Its characteristics are:
• All variables are complex with the exception of the shunt taps.
• There is an enormous number of variables and equality constraints which define non-linear
relations among the variables.
• The connectivity of the power system elements (the branch elements) determines the
number of terms in the equality constraints, especially in (7).
3.3.2 Classification of the variables and equality constraints
Since this set of equations represents a subset of the constraint set a classification of the already
used variables in subsets is advantageous:
The variables of the OPF problem will be called x from now on. Similarly behaving variables
will receive the same index and belong to the same group.
Subsets of variables can be characterized as follows (Sets of variables are put into a bracket
like { .. }).
• {xA1 } = {Y el−i−j
ii , Y el−i−j
ij , Y el−i−j
ji , Y el−i−j
jj , Y el−i−o
iio } (numerically given parameters of
all passive network elements)
• {xA2 } = {tel−i−j , sel−i−o } (transformer and shunt related variables)
• {xB } = {I el−i−j i , I el−i−j j , I el−i−o } (branch and shunt related current variables)
• {xC } = {I Gi , I Li } (generator and load related current variables)
• {xD } = {V i } (nodal voltage variables)
• {xE } = {S Gi , S Li } (generator and load related complex power variables)
• {xF } = {S el−i−j i , S el−i−j j , S el−i−o } (branch and shunt related complex power variables)
Note that all of the above variable sets are complex, except sel−i−o related to shunt elements.
Some of the variable sets created above can be further categorized: There is one variable
set whose elements can be assumed to be numerically known at fixed, given values:
• {xA1 }: Numerically given network parameters of the passive elements.
With this variable classification the equality constraints (2) to (12) can also be classified
into four categories:
(2) and (5) are characterized as follows:
gB = xB − fB (xA2 , xD ) = 0 (13)
12
(7) is characterized as follows:
gA = fA (xB , xC ) = 0 (14)
(8) and (9) are characterized as follows:
gE = xE − fE (xC , xD ) = 0 (15)
(10), (11) and (12) are characterized as follows:
gF = xF − fF (xB , xD ) = 0 (16)
(13) .. (16) represent the sets of equality constraints which must be satisfied for any numeric
power flow solution. They show certain properties:
• Each equality constraint set has its own properties. These properties have consequences
in the OPF solution process: The Jacobian submatrices, representing the first partial
derivatives of the equations with respect to all variables will have special and different
properties like sparse matrices, block-diagonal matrices, symmetric matrices, etc..
• The sets of equality constraints gE and gF could be eliminated from a mathematical

solution process if the set of free variables {xE } and {xF } do not appear in any other
equality and/or inequality constraint set. Note that as shown later in this paper, the
latter is the case, thus {xE } and {xF } cannot be eliminated from the equality constraint
set.
• {xB } of the equation set gB could immediately be replaced into gA and gF , thus elimina-
ting the variables {xB } in the equality constraints and reducing the number of equality
constraints. However, for certain mathematical formulations and because of OPF solution
based reasons, it is better not to eliminate the variables at this stage.
(13) .. (16) can be summarized into one compact equality constraint set as follows:
g(x) = 0 (17)
with xT = (xTA2 , xTB , xTC , xTD , xTE , xTF ) and g = (gA , gB , gE , gF ).
4 Mathematical formulation of operational constraints

4.1 Introduction
In the preceding section a power system model and needed equality constraints have been
formulated. Satisfying these equality constraints with any numerical set of variables means that
the physical characteristics of the power system for a model of category 2 are satisfied. The
problem is that many of these physically possible states do not make operational sense or are
not operationally possible. Thus in order to model the power system behavior more realistically
additional constraints have to be formulated. Different types of operational constraints can be
formulated:
13
• Physical damage to network equipment must be prevented since power system equipment
is often very expensive and hard to repair.
• Laws dictate mandatory standards to be satisfied by all utilities, e.g. the voltage am-
plitude at a node must be within certain upper and lower limits, or: a power end user
must have power available at almost 100% during the year, or: if any network element is
unvoluntarily outaged the power system must be brought back to an acceptable network
state within a given time period.
• Physically given limits for power system sources (any generator has its upper power limit).
• Power consumption at certain discrete time steps force the power to flow from the ge-
nerators via the transmission system to predefined geographical places, i.e. the power
consumed by loads at certain nodes at certain times is given and must be considered.
• Contracts among utilites determine precisely at what times how much power must be
imported or exported from one utility to another. This imposes limitations on the power
system operation and also on its model.
• Operational limits being computed by other power system problem analysis areas like
network stability determine e.g. the maximum allowable complex voltage angle shift from
node i to another node j.
• Human operators cannot implement more than say 10 % of all possible controls manually
within a give short time period.
These examples show that a huge number of operational constraints exist which must be
translated into mathematical constraint types.
These mathematical constraints are derived in the following subsections. Three main cons-
traint groups are identified: The transmission constraints, representing all operational limits
of the actual network. The contingency constraints are related to all operational aspects
if any network element is outaged as compared to the actual network and its associated net-
work state. In the third constraint group the operational policy based constraints are
formulated. They represent e.g. limits of human operator based system control.
The correct mathematical representation of the three constraint groups is emphasized in
the following subsections, although this might lead to problems which are not solvable today
with classical optimization tools.
4.2 Transmission constraints

Transmission constraints are always related to the actual network, i.e. to a network with given
branch connectivity.
4.2.1 Given complex loads
As already discussed in a preceding section individual loads cannot be influenced by operating

policies and must be satisfied at any time by the corresponding generation. Thus for any given
14
discrete time step where the individual load values S 0Li are either measured or predicted with
some method, the following must be valid:
S Li = S oLi ; i = 1 .. N (18)
An upper index o means a numerically given value (complex if underlined). This equality
constraint set deals with a subset of the variable set xE .
4.2.2 Branch current magnitude - maximum limit
The maximum current magnitude values for transmission branches, i.e. lines and transformers,
are given due to limitation of the branch material. Excessive currents would damage the trans-
mission elements.
|I el−i−j i | ≤ Iel−i−j
max
; i-j: all branches
(19)
|I el−i−j j | ≤ Iel−i−j ; i-j: all branches
max
An upper index max means a numerically given maximum limit value. An upper index min
means a numerically given minimum limit value. The symbol |(.)| refers to the absolute value
of the variable in (.). This inequality constraint set deals with the variable set xB .
4.2.3 Branch MVA-power - maximum limit

The same reason as the one for branch maximum current limit, discussed before, is valid.
|S el−i−j i | ≤ Sel−i−j
max ; i-j: all branches
(20)
|S el−i−j j | ≤ Sel−i−j
max
; i-j: all branches
This inequality constraint set deals with the variable set xF .

Either (19) or (20) or both must be formulated for each branch of the network.
4.2.4 Lower and upper nodal voltage magnitude limits

These limits are often given by very strict standards. Too high or too low voltages could cause
problems with respect to end user power apparatus damage or instability in the power system.
This could lead to unwanted and economically expensive partial unavailability of power for end
users.
Vimin ≤ |V i | ≤ Vimax ; i = 1 .. N (21)
This inequality constraint set deals with the variable set xD .

(21) is valid for every node of the network. There are nodes (often a subset of the generator
nodes) where the upper and lower voltage limits are identical, i.e. the voltage magnitude of
this node is numerically given.
15
4.2.5 Lower and upper generator active power limits

The active power of a generator i is defined to be the real part of the complex variable S Gi .
This quantity is physically limited in each generator.
min
PGi ≤ Real(S Gi ) ≤ PGi
max
; i = 1 .. N (22)
This inequality constraint set deals with the variable set xE .

(22) must be formulated for every generator. Often the lower limit is zero.
4.2.6 Lower and upper generator reactive power limits

The reactive power of a generator i is defined to be the imaginary part of the complex variable
S Gi . It is an important measure of voltage magnitude quality, e.g. a low voltage indicates a
local shortage of reactive power.
Gi ≤ Imag(S Gi ) ≤ QGi
Qmin max
; i = 1 .. N (23)
This inequality constraint set deals with the variable set xE .

(23) must be formulated for every generator. The upper and lower reactive power limits are
often not given as numeric values but as functions of the active generator power:
Qmin
Gi = fmini (Real(SGi ))
max (24)
QGi = fmaxi (Real(SGi ))
4.2.7 Upper and lower transformer tap magnitude limit
These limits come from the fact that the range of a tap is limited by the physical construction
of each transformer. Thus clear limits exist for each transformer.
el−i−j ≤ |tel−i−j | ≤ tel−i−j ; i-j: all transformer branches

tmin max
(25)
This inequality constraint set deals with a subset of the variable set xA2.
4.2.8 Upper and lower transformer tap angle limit
The same reasons as discussed for the transformer tap magnitude limits are valid here.
δtmin
el−i−j ≤ 6 (tel−i−j ) ≤ δtmax
el−i−j
(26)
; i-j: all transformer branches
Note that the symbol 6 refers to the angle of the following complex variable. This inequality
constraint set deals with a subset of the variable set xA2 .
In the power system mainly two types of transformers can be found with variable tap
position:
• In-phase tap changing transformer: The transformer can vary or regulate the voltage
magnitude
• Phase shifting transformer: The transformer can vary or regulate the voltage angle.
For an in-phase tap changing transformer usually a constraint of type (25), for a phase
shifting transformer a constraint of type (26) must be formulated.
16
4.2.9 Upper and lower shunt tap limit

This limit has to be understood as follows: In practice a limited number of discrete shunt
elements is available at some electrical nodes. These individual shunt elements can either be
switched in or out. Thus the upper shunt tap limit corresponds to the state where all shunts
are switched in and the lower limit to the lowest number of possible switched-in shunts.
el−i−o ≤ s
smin ≤ smax
el−i−o
el−i−o ; i-o: all shunt elements (27)
This inequality constraint set deals with a subset of the variable set xA2 .
4.2.10 Upper and lower limits on branch voltage angle

This type of constraint must be formulated if some stability based criterion is formulated for
the branch angle. Violating some of the constraints can cause severe dynamic stability problems
which could lead to power outage and severe economic penalties.
min
δel−i−j ≤ |6 (V i ) − 6 (V j )| ≤ δel−i−j
max
(28)
; i-j: all critical branch elements
This inequality constraint set deals with a subset of the variable set xD and must be
formulated for every branch determined by some stability criterion.
4.2.11 Minimum generator active power spinning reserve

X
N
min
Preserve ≤ max
(PGi − Real(SGi )) (29)
i=1
This inequality constraint is necessary in order to force each power system control area to
have a certain amount of total active generator power available for unforseen cases. An example
is the generator outage in a neighbouring power system control area: Here the power system
control areas with intact power generation automatically help to provide the power for a certain
time after the outage.
4.2.12 Upper and lower limits on total active power of a given set of branches
min ≤ | P
(i−j)∈{BSk } Real(S el−i−j i )| ≤ PBSk
PBS max
k (30)
; BSk : set of branches with limited total active power
This inequality constraint set deals with a subset of the variable set xD and must be
formulated for every branch set for which a constraint on the sum of the active power is valid.
Often the upper and lower limits for this total branch set flow is identical and numerically
given. This is the case if the network comprises the network data of more than one utility and
if the utilities have contracts for power transfer from one into the other power control area.
17
4.2.13 Control variable time-related movement limits

In a power system the operator can only control certain quantities. These quantities are called
controls and represent a physical apparatus to implement any operationally feasible and ac-
ceptable network state. Each control has an associated control variable. The most important
control variables are the following:
• Generator active power control PGi : This control variable is the real part of the variable
S Gi .
PGi = Real(SGi ) ; i: all generators (31)
• Generator reactive power control QGi : This control variable is the imaginary part of the
variable S Gi .
QGi = Imag(SGi ) ; i: all generators (32)
• Generator voltage magnitude control VGi : This control variable is the voltage magnitude
|V |i only of the generator nodes.
VGi = |V i | ; i: all generators (33)
• Phase shifter transformer tap position control δtel−i−j : This control variable is the angle
of the complex tap of a transformer tel−i−j .
δtel−i−j = 6 (tel−i−j ) ; i-j: all phase shifter transformers (34)
• In-phase transformer tap position control tel−i−j : This control variable is the magnitude
of the transformer tap tel−i−j .
tel−i−j = |tel−i−j | ; i-j: all in-phase transformers (35)
• Shunt value control: This setpoint is usually the real-number shunt tap variable sel−i−o
itself.
sel−i−o = sel−i−o ; i: all shunts (36)
• Active power interchange transaction control: A utility connected to neighbouring utilities

can buy or sell active power via the tie-lines which connect the areas. The control variable
is called Pinterchange−k−l (The index k refers to area k and the index l to a neighbouring
area l):
P
Pinterchange−k−l = (i−j)∈{k−l} Real(S el−i−j i )
(37)
; k-l: branches connecting utility l to utility k
18
The above mentioned variables on the left hand side of the equality constraints are called
control variables and represent variables with real values which can be directly influenced by
the power system operator.
The mathematical representation of control variables as used in this paper is u. They can
always be derived from variables of the vector x as shown with (31) .. (37) and are summarized
as follows:
u = gu (xA2 , xE , xD ) = gu (x) (38)
It is important to understand that the power system cannot realize the state of the desired
values at the time t when the control variables and the associated physical controls are changed
by the operator: Due to the dynamic nature of the power system there is always a time delay
from changing the control setpoint values to the time when the power system actually shows
the desired values. Also, it does not make sense to change the controls to values which cannot
be realized in the power system at all or only after quite a long time.
Thus there is an additional set of inequalities, representing the maximum difference between
actual power system state and the state where the power system can be moved from this state
within the time frame for which the OPF result is valid. This time frame corresponds often to
the discrete time interval in which one OPF calculation is done, see sections 2 and 3 of this
paper.
These additional lower and upper control variable constraints are formulated as follows in
general form:
umin ≤ u ≤ umax (39)
In (39), the limit vectors are derived both from the values of the actual network state and
the ability to move controls within a given time period. They can be assumed to be numerically
given.
4.2.14 Summary: Transmission constraints

Using the variable characterization of the preceding section the additional equality (18) and
inequality constraints (19) .. (30) can be setup as follows:
(18) {gE1 (xE1) = 0}
(38) {gu (xA2 , xE, xD ) − u = 0}
(19) {hB (xB ) ≤ 0}
(20), (22), (23), (30) {hE (xE ) ≤ 0}
(21), (28) {hD (xD ) ≤ 0}
(25), (26), (27) {hA (xA2 ) ≤ 0}
(39) {hu (u) ≤ 0}
All of the above equality constraints and those of (17) and all inequality constraints above
can be summarized in compact form as follows:
g(x, u) = 0 (40)
h(x, u) ≤ 0 (41)
19
Both (40) and also (41) are functions of mixed complex and real variables. However it is
always possible to formulate the problem with real variables by using the simple rectangular
or polar coordinate formulation.
From a mathematical point of view these equality and inequality constraints are relatively
simple functions of related complex OPF variables. At this point the choice of the variable
coordinate system for the OPF variables x has a direct influence on the complexity of the
real-variable formulation of the constraints:
When choosing the polar coordinate system then all angle related inequality constraints
6
( (.)) (like (26), (28), (34)) and all magnitude related functions (|(.)|) (like (19), (20), (21),
(25), (27), (30), (33), (35) and (37)) are very simple 1:1 functions (meaning xmagnitudei or
xanglei ) of the corresponding polar magnitude and polar angle related variables.
These functions are more complicated if using the rectangular
q coordinate system: Here
the functions for magnitude related functions take the form (xreali + ximagi ) and for angle
x
related functions arctan ximagreali
i
.
For functions where the real or imaginary part of a complex variable is desired (like (22),
(23), (31) and (32)), the polar coordinate formulation is more complex and takes the form
cos(xanglei ) · xmagnitudei for ’Real’- type functions and for ’Imag’ - type functions sin(xanglei ) ·
xmagnitudei .
In rectangular coordinates the corresponding functions are simple: For ’Real’- type functions
xreali and for ’Imag’ - type functions ximagi .
(40) and (41) represent the set of equality and inequality constraints. Both together are
called the transmission constraints within the complete OPF formulation.
Note that for each discrete time step for which an OPF is valid the transmission constraints
can be slightly different with respect to the symbolic formulation, but significantly different
with respect to the numerically given limit and load values.
4.3 Contingency constraints

In the preceding subsection all equality and inequality constraints refer to the actual network
state. This means that the OPF transmission constraints are based on a given network structure
with known branch, shunt, and generator statuses. It has been assumed that all these elements
are in a ’switched-in’ status. The network where all elements are in this ’switched-in’ status is
called the ’actual network’.
In addition to satisfying the transmission constraints for this ’actual network’, law or econo-
mic reasons force utilities not only to satisfy all transmission constraints of the ’actual network’,
but also to satisfy the so-called ’n-1-security’ based constraints which can be defined as follows:
The state of the ’actual network’ is defined as ’n-1-secure’ with post-contingency reschedu-
ling at any time T if
1. all operational constraints of the ’actual network’ are satisfied, i.e. the transmission
constraints and if
2. any one of the branches, shunts or generators is outaged (representing a contingency),

the new network state must be such that within a given short time period τshort , the
’new actual network’ can be shifted to a transmission constrained state by control means.
20
Since this control variable movement occurs immediately after the contingency happens
it is called post-contingency control movement.
Figure 2 shows typical network state changes caused by the outage on an element:
Network state
Outage of a network element at time T
After a short time, the network state
is recovered by operator control
Network state A
Network state B
Network state C
Time
t
Comments:
network state changing over time
Network state A: "n-1" secure network state with post-contingency control movements
Network state B: Transmission-constrained network state
Network state C: State with violated transmission constraints
Abbildung 2: Network state changes due to the outage of a network element at time t
Many utilities, e.g. at the east coast in the U.S.A., choose the time for recovering the
network state after the contingency has occured to a transmission constrained network state,
i.e. τshort as zero. This means that the utility wants to operate its actual network in such a
state that if any contingency occurs the new network state after the contingency is immediately
such that no transmission constraints are violated.
It is important to note that it can be assumed that the control variables u do not change
from the network state immediately before to immediately after the contingency occurs. Only
the OPF variables x change immediately due to the physical behavior of the power system.
Thus a contingency can immediately lead to operational constraint violations which must be
corrected by moving the control variables u within a short time τshort .
A dynamic simulation not being the objective, the recovery time effect is translated into a
maximum control variable move after the outage occurs (the so-called post-contingeny control
variable movement):
It is known how long it takes to move a certain power system control by a certain amount,
e.g. the active power of a generator from a generation of 100 MW to 130 MW. Thus it is
assumed that the maximum possible and operationally acceptable movements of all control
variables are known. The resulting new inequality constraint set representing the maximum
allowable upward and downward movements of all control variables per outage is as follows:
First the maximum movements for all control variables are computed numerically. Note
21
that different values can be chosen per contingency i:

max max
∆u(i) = τshort · ∆u/sec.(i) (42)
max
The scalar value τshort is given (in seconds) and the vector ∆u/sec.(i) represents the
numerically given maximum possible movements per second for each control variable.
max max
u − ∆u(i) ≤ u(i) ≤ u + ∆u(i)
(43)
i: all possible contingencies
In (43) the vector u(i) refers to the control variable set valid at time t+τshort , assuming that
the outage of network element i has occured at time t. I.e. for every possibly outage network
element i such a vector is created, representing the new state at time t + τshort . The size of the
vector u(i) per outage element i is the same as the size of u, with the exception of a control
variable related to an outaged network element.
Since the new network state after the outage must satisfy the transmission constraints the
model must satisfy the same equality and inequality constraints like in the ’actual network’
state (see (40) and (41)). Thus for each outage case i the following equality and inequality
constraints have to be formulated with the new outage state related variable set as defined
above:
g(i)(x(i), u(i)) = 0 ; i: all possible contingencies (44)
h(i)(x(i) , u(i)) ≤ 0 ; i: all possible contingencies (45)
In summary, by formulating contingency constraints the following points are significant:
• The number of variables is increased to (n = number of possible contingencies) n times the

number of variables of the ’actual network’, i.e. the problem is tremendously increased.
• The absolute values for the difference of control variables of the ’actual network’ and
corresponding control variables of the outage cases must be less or equal to a numerically
given value.
• A network state valid after the contingency has occured, must satisfy all operational
equality and inequality constraints. The difference in the symbolic formulation to the
actual network equations are as follows:
– A new set of variables is used for each contingency.

– The outaged network element must not be modelled in the equality constraints
– If there is an inequality constraint for the outaged element this inequality constraint
must not be formulated for the contingency network state.
22
4.4 Operational policy based constraints

4.4.1 Introduction
The constraints formulated in the preceding sections represent important restrictions on the
power system model used in the OPF calculation. Without including these constraints the
model will result in non-practical results.
One of the mathematically challenging problems is the fact that a power system operator
cannot change too many controls during a given time period. This is the case if the result
of the OPF optimization is not transferred automatically to the power system control. The
automatic control meachanism is called closed loop control. The mathematical formulation for
this problem is given in the following subsection.
4.4.2 Limited number of controller movements

In a preceding subsection the control variables u have been defined as functions of the OPF
variables x. Control variable movements are limited by the maximum physically possible change
during a given time period.
In the case of non-closed loop OPF operation further constraints must be formulated, since
the total number of control variable movements must be limited to a certain numerically given
value:
The problem can be stated in words as follows:
Given: The actual power system state is given in such a way that all equality constraints
(2) .. (12) are satisfied. The numeric values of the control variables for this state called ’Base
Case’(BC) state are represented with the vector uBC . They are computed with the equations
(31) .. (37).
Goal: Do not move more than a given number of control variables to some other network state
which is operationally more convenient. This new state and the associated control variables are
called u. A typical value for this number of maximum movable control variables is e.g. 10 % of
all control variables.
This problem of limiting the number of control variable movements is given not only from
the base case to an optimized base case network state, but also for each outage case i:
Especially when a contingency occurs the operator does not want to deal with too many
control variable movements which are needed to bring the network state back to an accep-
table new transmission constrained network state after the outage has occured. By the value
τshort , representing the time during which the power system must have been moved back to a
transmission contrained state, the operator is under heavy pressure and is interested to have a
limited number of movable control variables.
Outage related control variables movements refer the movement starting point to the state
represented by the vector u. Note that the number of movable control variables for each in-
dividual outage case can differ from the number of control variable movements for the base
case.
The control variable movements for the individual outage cases u(i) are derived from the
outage-related variables x(i) in analogy to (38) by simple transformations of the complex outage
23
case variables x(i) to real outage case related control variables u(i).
u(i) = gu(i)(x(i) ) ; i: all possible contingencies (46)
The problem is that it is not known beforehand which subset of the control variables to
move, both base case and also outage case related. This means that the status of each control
variable must be a variable and this is formulated mathematically as follows:
Each of the control variables can have a status of moved (1) or not moved (0). This status
is represented by the variable vector w which refers to the control variables of the base case.
For outage case i the corresponding status variable vectors are called w(i) .
" #
(i) 0
w, w = ; i: all possible contingencies (47)
1
(47) shows that for each element of the vectors w and w(i) the value can only be either 0
or 1.
With these additional, discrete variable type constraints the control variables u and u(i)
must satisfy the following conditions:
diag(u − uBC ) · (w − 1) = 0 (48)
and for the outage cases:

diag(u(i) − u) · w(i) − 1 = 0 ; i: all possible contingencies (49)
(47) together with (48) and (49) guarantee that the individual control variables are either
moved or not moved at all.
The number of control variable movements is limited with the following simple inequality
constraints representing the sum all individual variables of the vectors w and w(i):
For the base case:
wT · 1 ≤ CV max (50)
CV max stands for the maximum number of base case related control variable movements (a
scalar, numerically given value).
For the individual outage cases:
T max
w(i) · 1 ≤ CV (i) ; i: all possible contingencies (51)
max
CV (i) stands for the maximum number of outage case i related control variable move-
ments (i.e. per outage element i, a scalar, numerically given value).
(48) .. (51) represent the mathematical formulation for the operational problem of the
limited number of control variable movements.
24
4.5 Overview: Network type, network state and constraint set related term
definitions
Important goals of the power system control are the reliable and economic operation. These
goals are translated into OPF problem parts: Reliability is translated into the operational cons-
traints, economy into the objective function (discussed later in this paper). These operational
constraints can be grouped and important terms have been introduced in this section. They
are summarized in the following table:
Term Description
Actual network Network with given branch, shunt and generator statuses.
I.e. it is known which of these elements are switched in
and which ones are switched out.
Contingency i network This is the network where one element i is outaged as
compared to the ’actual network’. If the outage is lasting
for a long time this new network will be a new ’actual
network’.
Base case network state This is any network state (i.e. voltages, currents, powers)
computed or measured at an ’actual network’. Only the
power flow equality constraints of the ’actual network’ are
satisfied. There can be violations on all kinds of operatio-
nal constraints. It is the most general state from which
an OPF optimization can be started. One of the goals
of the operator is to shift the network from this network
state via the ’transmission constrained network state’ to
the ’n-1-secure network state’.
Transmission constrai- An ’actual network’ is in this state if all transmission
ned network state constraints are satisfied (some of the contingency cons-
traints are violated).
n-1-secure network This is a computed state of the ’actual network’ where all
state operational constraints including those for contingencies
are satisfied.
5 OPF objectives and objective functions

5.1 Introduction
An important part in any mathematical optimization problem is the objective function which
allows to make a distinct and often unique, optimal selection out of the solution region defined
by the equality and inequality constraints. Also, the objective function is needed to drive a
mathematical optimization process towards an optimal solution.
As already discussed before utilities can have different goals or power system operation
objectives. Some goals are clearly defined like minimum active power losses in the resistive parts
of the transmission system branches or minimum total cost for the active power generation of
the generators. These objectives are of economical nature and can thus easily be justified.
25
On the other side less clear goals exist which can be of the following types and often depend
on operational utility policies: If the power system is monitored and if the power system is found
to be in a state where either transmission or contingency constraints are violated, operate the
system in such a way that the violations are eliminated as quickly as possibly.
It is obvious and mathematicians can prove that objectives often are exchangeable with
equivalent inequality constraints, i.e. putting harder inequality constraints limits can also limit
the value of an objective function or the other way around: Formulating inequality constraints
as part of the objective function can give valid solutions by enforcing so-called soft limits.
This paper concentrates on the formulation and not on the solution of the OPF problem by
giving equality and inequality constraints. In this section possible objectives and the mathemi-
cal objective function formulations are discussed. However, one must always keep in mind that
the possibility to partially exchange inequality constraints and objective function formulations
is given. The solution process actually decides how to combine the constraint and objective
function formulations.
5.2 Mathematical formulation of various OPF objective functions

5.2.1 Objective: Minimum active power losses
The active power losses (called ’losses’ from now on) represent a quantity whose minimization
can easily be justified by an electric utility if the effort to actually operate in a minimum loss
mode is not too expensive. Losses are generated in the resistances of the transmission lines and
are a measure of the difference of the generated total active power to the total active load at
any time.
Two different loss-related cases exist: If the network model comprises only network parts
of the own utility controlled area the losses are computed as follows:
Loss case A:
X
N
Minimize PLoss = (Real(SGi ) − Real(S Li )) (52)
i=1
Thus in this case, the losses are a simple function of elements of the vector xE representing
the generator and load related nodal complex powers.
Loss case B:
Here the modelled network also comprises network parts of the neighbored utilities. Reasons
for this have been discussed in previous section and are often the case if the highest voltage
levels of the network must be modelled. For this case B the losses are the sum of the active
powers of each branch of the utility controlled area. For each branch the active power flow from
node i to node j must be added to the branch flow from node j to node i. This results in the
active power losses per branch. Thus the losses for a defined area ’a’ are computed as follows:
X
area ’a’ =
Minimize PLoss Real(Sel−i−j i ) + Real(S el−i−j j ) (53)
i−j∈area ’a’
Thus in this case, the losses are a simple function of elements of the vector xF representing
the complex power of branches.
26
5.2.2 Objective: Minimum total active power operating cost

Each utility has control over generation of different kind, e.g. hydro, hydro-thermal or thermal
power generation. Each type of generation has cost associated with it: Hydro power is, if
available, usually the cheapest power (water does not cost much in mountain areas) and thermal
power generation is more expensive (often oil, gas or nuclear material is used as primary energy
resource. These resources have to be bought at market prices).
In addition to own power resources the utility can buy or sell power from or to neighbouring
utilitities. Sometimes it can be cheaper to buy power than to produce it within the power system
control area.
It is in the interest of the utilities and also of the paying power consumers to minimize
the cost associated with active power generation. It can be assumed that the cost of each
generator can be represented as a distinct curve of relating cost to the active power which the
generator delivers. Usually this curve is given for the full range of the operating capability of
the generator. The general type of a cost curve can be written as follows:
CGi = ci (Real(SGi )) (54)
where ci is a general function of the active power of the generator i. A typical cost curve is
shown in Fig. 3.
Cost
Valve points
. Generator active power
Abbildung 3: Typical true cost curve for hydro-thermal power generation
As seen in Fig. 3 the general cost curve type is very complex: Non-convex cost curves are
possible. Also, it is important to note that the cost curves can be assumed to be separable with
respect to the active power generation of the generatorsm, i.e. the cost of each generator is
only dependent on the cost of its own active power generation and not on the cost of another
generator power.
The cost curve associated with buying or selling power is not as clear. Sometimes it is
step-wise linear, sometimes linear over the full range. Note that selling power leads to a cost
curve with negative cost values.
The problem is the shape of the cost curves. Optimization algorithm usually cannot deal
with cost curve shapes as shown in Fig. 3. Thus, the cost curves are usually modified and a
convex, smooth cost curve as shown in Fig. 4 can be assumed to be the cost curve model for
the OPF.
27
Approximated cost
Generator active power
Abbildung 4: Corrected smooth and convex power generation cost curve for OPF model
Often these cost curves are assumed to be piecewise quadratic with smooth transition at the
cost curve break points. Also, exponential cost curve representations are seen in OPF models.
Usually the approximated smooth and convex cost curves (one per active power generation) is
given as the exact reference along which the OPF optimum solution must be found. Whether
this makes sense or not is not the issue of this paper. However, it should be kept in mind that
the cost curve issue is open and needs to be resolved the more accurate the remaining power
system model is chosen. An inaccurate cost curve model can prevent an accurate optimization
result, even if the remaining model parts are model with very high accuracy or in other words,
the optimization result is not more accurate than the chosen accuracy of the model components.
The total cost of all generators and transaction interchange active power can be represented
as objective function as follows:
P
Minimize C = N capproximatedi (Real(SGi ))
P i=1
+ l∈k−l cint−approximatedk−l (Pinterchange−k−l ) (55)
k-l: all areas l connected to area k
5.2.3 Objective: Fast transition from a violated to a non-violated network state

For many reasons a power system can suddenly move into a state where important constraints
are violated. If this happens a very important objective of a utility can be, to move the power
system as quickly as possible back to a network state without violating constraints, including
transmission, contingency and utility policy constraints.
The problem to get a mathematical objective function is the term ’quick’ which is time
related. The OPF, however, does not have time as variable in the formulation. Thus another
measure for the quick operation must be found. One measure can be to determine a feasible
state (a state where no inequality constraints are violated) with a minimum number of control
variables movements. Provided a solution for the minimum number of control variables can be
found in reasonable time, this is probably one of the fastest ways for the utility to achieve a
feasible operation. Thus the objective function would be:
Minimize F = wT · 1 (56)
28
There might be other objective function formulations for this special objective. The main
problem is to translate the operating philosophies into the objective function. Once it is given
the mathematical process tries to solve the problem as accurately as possible. The optimization
result could be such that although satisfying the mathematical model, it does not conform to
the often rule-based operational policies of the utility. This will, especially in the first imple-
mentation phase of the OPF, lead to some type of iterative process for defining the objective
function and the resulting mathematical optimum. At the end of this iterative process the opti-
mization result must both satisfy the mathematical optimum criteria and also the operational,
often rule based philosophies of the utility.
6 The complete OPF formulation

6.1 Mathematical OPF formulation - summary
The OPF formulation can be split in three main parts: The equality constraints, the inequality
constraints and the objective function. Each of these three parts has its own properties, as
discussed in the preceding sections. The following three tables summarize the OPF problem:
Equality constraints:
Description Mathematical formulation Equations
Actual network state g(x, u) = 0 (40)
Contingency i network g(i)(x(i), u(i)) = 0 (44) ∀ i
state
h iT
Control variable move- w, w(i) = 0T 1T (47) ∀ i
ment status
Actual network control diag(u − uBC ) · (w − 1) = 0 (48)
variable movement
Post-contingency case i diag(u(i) − u) · w(i) − 1 = 0 (49) ∀ i
variable movement
Inequality constraints:
Description Mathematical formulation Equations
Actual network state h(x, u) ≤ 0 (41)
Contingency i network h(i) (x(i), u(i)) ≤ 0 (45) ∀ i
state
max max
Post-contingency i con- u − ∆u(i) ≤ u(i) ≤ u + ∆u(i) (43) ∀ i
trol movement
Actual network control wT · 1 ≤ CV max (50)
movement status
T max
Contingency case i con- w(i) · 1 ≤ CV (i) (51) ∀ i
trol movement status
29
Typical OPF objective functions:

Description Equations
Active power losses (52), (53)
Total active power operating cost (55)
Fast transition from a violated to a non-violated network state (56)
Note that only one of the possible objective functions together with all equality and ine-
quality constraints can be chosen and can be solved for one optimal solution variable set.
Here another problem with the present OPF formulations and solutions becomes obvious:
Although a utility would like to satisfy more than one objective at the same time, the classical
mathematical optimization algorithms allow only one objective function at a time. The pro-
blem of getting a solution which considers more than one objective function (a multi-objective
function problem) at a time exists and should be addressed in the future to make the OPF
result more practically acceptable and applicable.
6.2 Sensitivity of the OPF solution with respect to parameter changes

When using the OPF with any of the above mentioned objectives and the corresponding ma-
thematical objective functions, the OPF formulation is today such that parameters like the
network data, the inequality constraint limit values and the measured or predicted load values
are numerically precisely given. The operator, however, has a different view of these parameters,
especially with respect to the limit values of the inequality constraints: They do not represent
precisely given values but are often derived based on experience and on heuristic assumptions.
The operator would like to know how much the OPF result varies when the parameters are
slightly varied. The idea is to give the operator a feeling for example how much the violation of
a maximum branch MVA flow reduces the chosen objective function. From such a sensitivity
analysis the operator could derive important operating policies and it would provide deeper
insight into the OPF optimum.
Mathematically the problem is often formulated under the assumption that the binding
inequality constraint set of the OPF solution remains identical even with a slight change of the
parameters, mentioned before. Thus the problem is as follows: Assume that the OPF result
(solution process to obtain an OPF result see paper by H. Glavitsch) has been obtained by
some solution method. The resulting, numerically given values for this optimum are as follows
(denoted with index opt ):
uopt , xopt (optimum state for base case variables)

opt opt
u(i) , x(i) (optimum state for outage case i related variables)
(57)
wopt (optimum base case control variables statuses)
opt
w(i) (optimum control variable movement status for contingency i)
The problem is now a parametric equality constrained optimization problem and can be
formulated as follows, assuming that all variables of the OPF problem are summarized into X,
all equality constraints into G(X), all binding inequality constraints into Hbinding (X) and the
30
objective function is F (X):
Minimize F (Xopt + δx)

subject to G(Xopt + δx) = δg (58)
and Hbinding (Xopt + δx) = δh
In (58) the values of the vectors δg and δh are slight variations to the right hand sides of
the equality and binding inequality constraints. The problem is to get a solution of the type
δx = function of (Xopt, δg, δh) (59)
where the term ’function of’ can in the simplest case be a numeric matrix.
6.3 Discussion and future outlook

The OPF formulation as shown in section 6.1 has several characteristics which are summarized
in the following points.
• The OPF formulation comprises a huge number of variables and equality / inequality
constraints. Typical OPF problem sizes are summarized in the following table:
Number of Number of actual network Number of contingency network
nodes related constraints related constraints
equality inequality equality inequality
80 160 640 32000 120000
500 1000 4000 1.4 · 106 5 · 106
1000 2000 8000 2.8 · 106 20 · 106
2000 4000 16000 5.6 · 106 40 · 106
Note that in this table the equality constraints are assumed to be formulated in the most
compact form (this includes elimination of variables).
• The variable sets of the actual and contingency case related networks are almost perfectly
decoupled. Each variable set is related to a different network with distinct operational
constraints. The coupling between the variable sets is ’simple’ in that only maximum
difference values between a subset of the variable sets (the control variables) for each
different contingency case to the base case must be satisfied.
• The variables are either real, complex or discrete. Either the polar or rectangular coor-
dinate system must be used to represent the complex variables in real variable form.
The polar form introduces sin and cos - type functions into the OPF equations. The
rectangular form is responsible for square or square root - type and also arctan - type
equations.
• Today the complete OPF formulation as given before cannot be solved in closed form.
Typically, steps are undertaken to make the OPF problem solvable with classical optimi-
zation methods:
31
– The discrete variables of the vector w, w(i) are assumed to be known, i.e. the set of
moveable control variables is known. Doing this leads to an optimization problem
with continuous variables only.
– The maximum possible moves of the control variables within the time τshort after the
contingency occurs is often assumed to be zero, meaning that the control variable
values do not change from the n-1-secure state to any post contingency state. This
leads to an optimization problem with much less control variables.
– The contingency based equations g(i) and h(i) are often linearized around the n-1
secure state variables. The equations resulting from this linearization are seen as the
exact equation set which the OPF optimum has to satisfy.
Today (1992), the OPF problem as stated in this paper is not solvable in real-time with
computers presently available in Energy Management Centers. Faster computers, new optimi-
zation algorithms and the feedback obtained from practical OPF usage will change the OPF
formulation and its real-time usage in the near future. Modelling the OPF as realistically as
possible together with a robust, fast OPF execution will lead to high acceptance of this powerful
power system control tool.
For the author of this paper, it is clear that the OPF will be more and more important for
the electric power industry due to its potential in increasing both power system reliability and
also power system economy without much investment in power system hardware.
32
A APPENDIX
A.1 Symbols
The following notations are used througout this text:
• Symbols representing complex variables are underlined.
• Matrices are shown in capital boldface letters.
• Vectors are shown in small boldface letters.

∗ Conjugate complex
opt Associated variable is optimum variable
T
Transposed
low
Low limit
high
Upper (high) limit
∆ Change
|(.)| Absolute value of a variable (.) (variable can be complex)

6 (.) Angle value of a complex variable (.)
Real(.) Real part of a complex variable (.)
Imag(.) Imaginary part of a complex variable (.)
(i) Related to contingency case i
diag(.) Diagonal matrix
33
N Total number of electrical nodes
n Total number of network elements
I el−i−j i Complex current of element from node i to node j, computed at node

i
y el−i−j
ii
Primitive Y-matrix element (i,i) of branch element between nodes i
and j
y el−i−j
ij
Primitive Y-matrix element (i,j) of branch element between nodes i
and j
y el−i−j
ji
Primitive Y-matrix element (j,i) of branch element between nodes i
and j
y el−i−j
jj
Primitive Y-matrix element (j,j) of branch element between nodes i
and j
Vi Complex voltage at node i
Y el−i−j
ii Two-port admittance element (i,i) of branch between nodes i and j
el−i−j
Y ij Two-port admittance element (i,j) of branch between nodes i and j
Y el−i−j
ji Two-port admittance element (j,i) of branch between nodes i and j
Y el−i−j
jj Two-port admittance element (j,j) of branch between nodes i and j
tel−i−j Complex tap of transformer branch between nodes i and j
I el−i−o Complex current of shunt element at node i
y el−i−o
ii
Admittance element of shunt at node i
sel−i−o Shunt tap value (corresponding to the number of switched in shunts)

at node i
Y el−i−o
iio Admittance element of total shunt at node i
34
I Gi Complex generator current at node i
I Li Complex load current at node i
S Gi Complex power of generator at node i
S Li Complex power of load at node i
S el−i−j i Complex power of branch from node i to node j computed at node i
S el−i−j j Complex power of branch from node i to node j computed at node j
S el−i−o Complex power of shunt element at node i
{xA1 } = {Y el−i−j
ii , Y el−i−j
ij , Y el−i−j
ji , Y el−i−j
jj , Y el−i−o
iio } (numerically given
parameters of all passive network elements)
{xA2 } = {tel−i−j , sel−i−o } (transformer and shunt related variables)
{xB } = {I el−i−j i , I el−i−j j , I el−i−o } (branch and shunt related current

variables)
{xC } = {I Gi , I Li } (generator and load related current variables)
{xD } = {V i } (nodal voltage variables)
{xE } = {S Gi , S Li } (generator and load related complex power variables)
{xF } = {S el−i−j i , S el−i−j j , Sel−i−o } (branch and shunt related complex

power variables)
gA Equation set mainly related to variables xA2
gB Equation set mainly related to variables xB
gE Equation set mainly related to variables xE
gF Equation set mainly related to variables xF
x Variable set which include xA2 , xB , xE and xF
g(x) Equation set which includes gA , gB , gE , gF
35
S oLi Given complex power value of load at node i

max
Iel−i−j Maximum current of branch between nodes i and j
max
Sel−i−j Maximum MVA-power of branch between nodes i and j
Vimin Minimum voltage magnitude at node i
Vimax Maximum voltage magnitude at node i

min
PGi Minimum active power of generator at node i
max
PGi Maximum active power of generator at node i
Qmin
Gi Minimum reactive power of generator at node i
Qmax
Gi Maximum reactive power of generator at node i
tmin
el−i−j Minimum tap position of in-phase transformer between nodes i and j
tmax
el−i−j Maximum tap position of in-phase transformer between nodes i and
j
δtmin
el−i−j Minimum tap position of phase shifter tap transformer between nodes
i and j
δtmax
el−i−j Maximum tap position of phase shifter tap transformer between no-
des i and j
smin
el−i−o Minimum tap for shunt bank at node i
smax
el−i−o Maximum tap for shunt bank at node i
min
δel−i−j Minimum angle value between nodes i and j
max
δel−i−j Maximum angle value between nodes i and j
min
Preserve Minimum active generator power spinning reserve
min
PBS Minimum active power value for transfer at branch set k
k
max
PBS Maximum active power value for transfer at branch set k
k
PGi Active power of generator at node i (a control variable)
QGi Reactive power of generator at node i (a control variable)
VGi Voltage magnitude of generator at node i (a control variable)
δtel−i−j Phase shift transformer tap magnitude of transformer between nodes

i and j (a control variable)
tel−i−j In-phase transformer tap magnitude of transformer between nodes i
and j (a control variable)
sel−i−o Shunt tap (a control variable)
Pinterchange−k−l Active power transaction interchange between area k and area l (a

control variable)
36
u Control variable vector of the actual network
u(i) Control variable vector of contingency case i
x OPF state variable vector of the actual network
x(i) OPF state variable vector of contingency case i
umin Control variable minimum values
umax Control variable maximum values
g(x, u) OPF equality constraint set represented in OPF variables and control
variables (actual network)
h(x, u) OPF inequality constraint set represented in OPF variables and con-
trol variables (actual network)
∆umax Maximum move of the control variables from base case state to some
other (optimized) state (actual network)
max
∆u(i) Maximum move of the control variables from (optimized) base case
state to some state after contingency i occurs (contingency case i)
τshort Short time (in seconds) during which the operator has to move the
control variables to some optimal state
∆u/sec.max Maximum move of the control variables per second (actual network)
max
∆u/sec.(i) Maximum move of the control variables per second (contingency i
network)
g(i) (x(i) , u(i)) OPF equality constraint set represented in OPF variables and control
variables (contingency case i network)
h(i) (x(i), u(i)) OPF inequality constraint set represented in OPF variables and con-
trol variables (contingency case i network)
(i)
gu (x(i) ) Functions relating the OPF variables x(i) to the control variable u(i)
of the contingency case i
w OPF control variable movement status variables (actual network)
w(i) OPF control variable movement status variables (contingency i

network)
uBC Given base case (BC) control variable values at any time t describing
a network state with possible violated operational constraints
CV max Maximum number of moveable control variables (actual network)
max
CV (i) Maximum number of moveable control variables (contingency i
network)
37
PLoss Loss function of the whole network

area ’a’
PLoss Loss function of a part of the network
CGi Cost function of the active power of the generator at node i
capproximatedi Approximated, smooth active power cost curve for generator i
cint−approximatedk−lApproximated, smooth active transaction interchange power cost cur-

ve for between area k and area l
C Total cost of all generators
F Special objective function related to the quick move of control

variables
38
Literatur
[1] D.J. Gaushell, H.T. Darlington; Supervisory Control and Data Acquisition, Proceedings
of the IEEE, Special issue on computers in power system operations, (1987), Vol. 75, No.
12, pp. 1645 - 1658
[2] T.M. Athay; Generation Scheduling and Control, Proceedings of the IEEE, Special issue
on computers in power system operations, (1987), Vol. 75, No. 12, pp. 1592-1606
[3] B. Stott, O. Alsac, A.J. Monticelli; Security Analysis and Optimization, Proceedings of
the IEEE, Special issue on computers in power system operations, (1987), Vol. 75, No. 12,
pp. 1623 - 1644
[4] N. Balu, et. al; On-Line Power System Security Analysis, Proceedings of the IEEE, Vol.
80, No. 2, February 1992, pp. 262 - 280
[5] H. Glavitsch, R. Bacher; Optimal Power Flow Algorithms, Control and Dynamic Systems,
Vol. 41, Academic Press, Inc. (1991), pp. 135 - 205
[6] H.Happ; Optimal Power Dispatch. A Comprehensive Survey, IEEE Transactions on Power
Apparatus and Systems, Vol. 96. (1977), pp. 841-853
[7] J. Carpentier; Optimal Power Flows; Electrical Power & Energy Systems, Butterworths,
Vol 1, No.1, April 1979
39
Inhaltsverzeichnis
1 Introduction 2
2 The role of the optimal power flow (OPF) computation within the overall
power system control 3
3 The power flow model as equality constraint set of the OPF 6

3.1 Basic model assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1.1 Power system loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1.2 Power system model: Steady state and symmetric power system operation 6
3.1.3 Power system model: Generation, load and the transmission system . . 7
3.2 Power flow: Mathematical model of the power system for steady state simulation 8
3.2.1 Passive power system elements . . . . . . . . . . . . . . . . . . . . . . . 8
3.2.2 Kirchhoff-law: All currents at a node must add up to zero . . . . . . . . 10
3.2.3 Power - voltage - current - relationship . . . . . . . . . . . . . . . . . . . 11
3.3 Mathematical formulation of the various equality constraint sets . . . . . . . . 12
3.3.1 General characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3.2 Classification of the variables and equality constraints . . . . . . . . . . 12
4 Mathematical formulation of operational constraints 13

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2 Transmission constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2.1 Given complex loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2.2 Branch current magnitude - maximum limit . . . . . . . . . . . . . . . . 15
4.2.3 Branch MVA-power - maximum limit . . . . . . . . . . . . . . . . . . . 15
4.2.4 Lower and upper nodal voltage magnitude limits . . . . . . . . . . . . . 15
4.2.5 Lower and upper generator active power limits . . . . . . . . . . . . . . 16
4.2.6 Lower and upper generator reactive power limits . . . . . . . . . . . . . 16
4.2.7 Upper and lower transformer tap magnitude limit . . . . . . . . . . . . 16
4.2.8 Upper and lower transformer tap angle limit . . . . . . . . . . . . . . . 16
4.2.9 Upper and lower shunt tap limit . . . . . . . . . . . . . . . . . . . . . . 17
4.2.10 Upper and lower limits on branch voltage angle . . . . . . . . . . . . . . 17
4.2.11 Minimum generator active power spinning reserve . . . . . . . . . . . . 17
4.2.12 Upper and lower limits on total active power of a given set of branches . 17
4.2.13 Control variable time-related movement limits . . . . . . . . . . . . . . 18
4.2.14 Summary: Transmission constraints . . . . . . . . . . . . . . . . . . . . 19
4.3 Contingency constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.4 Operational policy based constraints . . . . . . . . . . . . . . . . . . . . . . . . 23
4.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.4.2 Limited number of controller movements . . . . . . . . . . . . . . . . . . 23
4.5 Overview: Network type, network state and constraint set related term definitions 25
40
5 OPF objectives and objective functions 25

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.2 Mathematical formulation of various OPF objective functions . . . . . . . . . . 26
5.2.1 Objective: Minimum active power losses . . . . . . . . . . . . . . . . . . 26
5.2.2 Objective: Minimum total active power operating cost . . . . . . . . . . 27
5.2.3 Objective: Fast transition from a violated to a non-violated network state 28
6 The complete OPF formulation 29

6.1 Mathematical OPF formulation - summary . . . . . . . . . . . . . . . . . . . . 29
6.2 Sensitivity of the OPF solution with respect to parameter changes . . . . . . . 30
6.3 Discussion and future outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
A APPENDIX 33
A.1 Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
41

Operational Constraints

Uploaded by

Copyright:

Available Formats

Operational Constraints

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Operational Constraints

Uploaded by

Copyright:

Available Formats

Optimization in Planning and Operation of Electric Power Systems. Physica-Verlag (Springer), Heidelberg, May 93, pp.

Power System Models, Objectives and Constraints in Optimal

Swiss Federal Institute of Technology (ETH)

P ower(t) = V oltage(t) ∗ Current(t) (1)

In (1) the variable (t) refers to time.

2 The role of the optimal power flow (OPF) computation wi-

• that no power transmission based power system limits are violated.

• is precise in mathematical terms;

3 The power flow model as equality constraint set of the OPF

• an additional mechanism with the incorporation of measured or accurately predicted end

• Overhead transmission lines and

(5) is the equation for a shunt in the model for category 2.

3.2.2 Kirchhoff-law: All currents at a node must add up to zero

Abbildung 1: Currents at a node i: Conventions

• N represents the total number of electrical nodes

• All generators at a node i are summarized in one generator (I Gi ).

• All loads at a node i are summarized in one load (I Li ).

• All shunts at a node i are summarized in one shunt (I el−i−o ).

• If there is no generator at a node i then I Gi = 0.

• If there is no load at a node i then I Li = 0.

• If there is no shunt at a node i then I el−i−o = 0.

• If there is no branch-type connection between nodes i and j then I el−i−j i = 0 and

3.2.3 Power - voltage - current - relationship

S Gi = V i I ∗Gi ; i = 1 ... N (8)

S Li = V i I ∗Li ; i = 1 ... N (9)

S el−i−j i = V i I ∗el−i−j i ; i-j = all branch elements (10)

S el−i−j j = V j I ∗el−i−j j ; i-j = all branch elements (11)

S el−i−o = V i I ∗el−i−o ; i = 1 ... N (12)

3.3 Mathematical formulation of the various equality constraint sets

3.3.2 Classification of the variables and equality constraints

• {xA2 } = {tel−i−j , sel−i−o } (transformer and shunt related variables)

• {xC } = {I Gi , I Li } (generator and load related current variables)

• {xD } = {V i } (nodal voltage variables)

• {xE } = {S Gi , S Li } (generator and load related complex power variables)

• {xA1 }: Numerically given network parameters of the passive elements.

(7) is characterized as follows:

(8) and (9) are characterized as follows:

(10), (11) and (12) are characterized as follows:

• The sets of equality constraints gE and gF could be eliminated from a mathematical

with xT = (xTA2 , xTB , xTC , xTD , xTE , xTF ) and g = (gA , gB , gE , gF ).

4 Mathematical formulation of operational constraints

4.2 Transmission constraints

4.2.1 Given complex loads

As already discussed in a preceding section individual loads cannot be influenced by operating

4.2.2 Branch current magnitude - maximum limit

4.2.3 Branch MVA-power - maximum limit

This inequality constraint set deals with the variable set xF .

4.2.4 Lower and upper nodal voltage magnitude limits

Vimin ≤ |V i | ≤ Vimax ; i = 1 .. N (21)

This inequality constraint set deals with the variable set xD .

4.2.5 Lower and upper generator active power limits

This inequality constraint set deals with the variable set xE .

4.2.6 Lower and upper generator reactive power limits

This inequality constraint set deals with the variable set xE .

4.2.7 Upper and lower transformer tap magnitude limit

el−i−j ≤ |tel−i−j | ≤ tel−i−j ; i-j: all transformer branches

4.2.8 Upper and lower transformer tap angle limit

4.2.9 Upper and lower shunt tap limit

4.2.10 Upper and lower limits on branch voltage angle

4.2.11 Minimum generator active power spinning reserve