Power System Operations and Control

Prof. S. N. Singh
Department of Electrical Engineering
Indian Institute of Technology, Kanpur

Module – 4
Lecture – 4

Welcome to lecture number four of module four. In previous lecture, that is lecture
number three, we discussed the some of the non-linear optimization program methods
and I discussed the contour techniques. We can go for some more non-linear
optimization technique, then I will also discuss some of the linear programming
application optimization techniques. So, those are very widely used in power system
operation and control analysis purpose.

(Refer Slide Time: 00:50)

So, in the Kuhn Tucker conditions, again here just our objective is to minimize or
optimize the function 𝑓(𝑥̇ ) bar here again shown, it is a vector of the variables that is a n
any numbers subjected to here, the various constraints. One is your ℎ𝑖 that is called
equality constraint and here 𝑔𝑖 is inequality constraints. Again, the constraints here I did
not define that in the previous lecture. So, the constraints we can say whether the
constraints are the functional constraints or parametric constraints.

For example, if you are putting some constraints on x variable here, xi itself here, then it
is you can have some limit, you can say xiupp upper limit and here I can say xilower lower
limit. So, this is basically x here is the parameter and the parametric, so this constraints is
known as parametric constraints, whereas if you are suppose writing some constraints
like your 𝑥𝑖2 + 2𝑥𝑖 − 4 ≤ 0. So, this is your function of x, and then it is called your
functional constraints, here this is inequality, so it is called functional inequality
constraint. So, we can have the two types of constraints, one is your functional
constraints and other is your parametric constraints.

In the functional constraints, again we can go for whether it is equality or inequality. So,
this ℎ𝑖 is nothing but your functional equality constraints, equality means here equal to 0
and 𝑔𝑖 is your inequality constraint. That includes both your functional as well as your
parametric inequalities. One thing also I like to remind you because this x here, if you
are writing in this form, so this is upper and lower bound of your state variable or that is
a your parameter that xi that we are going to choose. Optimum value of x in a such a
fashion that we can get the minimum value or you can say optimum value of function
𝑓(𝑥̇ ).

So, these constraints can be changed into the two inequality constraints, for example, if
𝑢𝑝𝑝
you are writing here xilower ≤ 𝑥𝑖 ≤ 𝑥𝑖 what we can do? We can go for this one, I can
𝑢𝑝𝑝
say 𝑥𝑖 − 𝑥𝑖 ≤ 0 and this just like your 𝑔(𝑥̇ ) and that function. Similarly, if you are
writing for another side here, I can say 𝑥𝑖𝑙𝑜𝑤 ≤ 𝑥𝑖 and I can simply write here, we can
send this here. So, it is I can say 𝑥𝑖𝑙𝑜𝑤 − 𝑥𝑖 ≤ 0 and this is your another function. So,
these functional constraints that is an upper and lower bounds if you are having. So, that
can be broken into the two constraints, two inequality constraints, here one is this,
another is this, and then we can solve.

So, this includes all the constraints including your functional as well as the parametric
constraints and that can be written in this function, that is your I can say 𝑔(𝑥) ≤ 0 in this
function. So, here we can have this basically shows the standard expression for any
optimal programming technique. Now, again sometimes some people say which
constraints are again I discussed binding constraints or non-binding constraints, how we
can go for this?

For example, you can see if you are having here let us suppose just I want to say, so
simple example on the binding if I will say your function here this constraints ℎ(𝑥 ), here
that is the vector of the two variables. If I can write here 𝑥1 + 𝑥2 − 3.75 = 0 and same
time I am saying this x1 here is the inequality constraints bounded by −3 ≤ 𝑥1 ≤ 3 and
your x2 here, I can say this is also −3 ≤ 𝑥2 ≤ 3. Another, we have an inequality
constraint 𝑔(𝑥 ) = 𝑥1 − 2𝑥2 − 8 ≤ 0. So, if you draw here, the region because we have
to search the minimum in that region these are the constraints.

So, in that constraints, we have to search the solution that the minimum that we are going
to get. So, for example, here for this case, if you draw here let us suppose we are having
the variable x1 this side, x2 this side, so what will happen? Now, the value of x2=3, we
will be having this zone, x2=-3,, here that to this zone, X1 positive 3 here and X2 positive
3 here, so for this what we are getting? We are getting for this and this we are getting this
inequality constraints and surrounded by this region. Sometimes, we are also having
other constraints; here I can say this is inequality constraints. That is you𝑥1 − 𝑥2 − 8 ≤
0..

So, we can again draw another function here that constraints, we can go this slope is
negative on the x2. So, we can say first here this is a constraints I can say here one here
that is linear equation here. Similarly, we will get another linear equation corresponding
to this, let us suppose your x2 is 0, then we are getting for equality first take as a 0 here.
So, 0 means x is equal x1 = 8 means if x2 = 0, we are getting some value here. If your are
getting x1 = 0, then x2 value is your minus 4, so you are getting here. So, we are getting
some sort of constraints here, so we are here getting some constraints like this, so what is
happening?

These constraints inequalities constraints are showing the regions mean we are getting
this, then this, then this, then this. So, this is basically your region of the inequality
constraints. So, your solution will lie in this one if you are not having these constraints
that are equality constraints, but if you are putting the equality constraints, means
suppose you are equality constraints are written here H this is a constraints. So, your
solution will lie here in this line itself because equality constraints are always binding
that must satisfy, here I said this may be not equal to 0, it may be less than that.

So, this is the regions, so we are getting some regions for the inequalities and equality
and the solution here must be on this line. So, from the inequalities, we had this region I
can say a b, then we move to c we move to d we move to e and again a. So, this was the
region enclosed by three inequalities constraints, so this was known as the feasible
region this region. Now, putting these constraints it is the binding constraints, so the
solution will be here on this line. Again if your other constraints are binding, let us
suppose your this constraints are binding x1 = 3. So, solution optimal will be here only
because this is the binding, so the intersection of this will be over the binding and this
will be solution.

Sometimes what happens, your solutions may be not binding and it may be somewhere
else suppose this is also binding somewhere here. So, we can say this solution is
infeasible, means it is not inside, it is not on this line solution is somewhere else, and
your function is going somewhere else. So, that is called infeasible solution, so coming
to the your again this extended problem. So, this is the extended notation of any
optimization problem that is the we can optimize function f subjected to the constraints,
and again the constraints can be a equality and inequality constraints.

So, in Kuhn Tucker condition, we saw we had the equality constraints as well as
inequality constraints of objective function, here, what we are going to do? In the slack
variable, formulation, we convert first the inequality constraints to the equality
constraints by adding some slack variables, what happens? Since this function is less
than 0, we are trying to add some functions so that we can make a change, so what
happens? Here, 𝑔𝑖 (𝑥̅ )here plus Si2 we are adding Si2= 0, why S square because we want
that S to be positive in all the case. If it is not S square in some case it may be negative.

So, to avoid this confusion that is we add some Si is non-negative and this is known as a
slack variable and it is always positive, so now what happens? This is now changed to
this one, and then we can have our extended equation can be again reformulated.
(Refer Slide Time: 10:17)

Then, we can have the lagrangian multiplier for the equality here this λi for all the h’s
means x may be M in number. So, it is addition of all λ1h1+λ2h2 and so on so forth.
Similarly, here if you are having k inequality constraints, so we are having the Lagrange
multiplier λi‘. Here, 𝑔𝑖 (𝑥̅ ) + 𝑆𝑖 , so number of inequality constraints again we are going
to add that number of slack variables Si. So, the number of slack variables, now Si will be
K in number that is the K here inequality constraints. So, in that slack variable this
approach this slack variable formation again our this condition for optima first condition
that our this augmented objective function or Lagrange’s function L must be derivate
with its state variable.

We can say parameter that is xi, and then we have to put 0, so this is true for all the xi, it
will be n in number. Already, we are considered that x is a variable of n variables means
it is your x1 to your xn meansx, x2 so on so forth. So, we have the variables the function
is of n variable. Another here the condition here is similar that we can derive here this is
nothing but your equality constraints, means we have to derivate this function with
Lagrange multiplier of equality constraints. So, here we are getting nothing, but, we are
getting the same expression of equality constraints and since we are having n in number
equality constraints, so this equation will be m in number.

Another that we are going to third condition here that is we are going to derivate this
function with respective to this Lagrange multiplier corresponding to the inequality
constraints that is made to equality by adding the slack variable. So, we are just
differentiating this means we are getting this function again and it is k in number here.
So, we are getting k equation, another here we are going to derivate that is the fourth
condition. In this case is that we are going to derivate this with respect to Si, and then we
are getting what happen here the 2Siλi = 0 and we are getting this equation.

So, now we are getting these equations and based on that we can derive and we can solve
and we can get again the optimal solution, now here if Si is 0, what happens? Now, you
can see if this Si is 0 means your 𝑔𝑖 (𝑥̅ ) = 0 is equal to and once it is 0, it becomes
equality constraints without adding the slack variable means it is the binding constraints.
So, we have to say this equal inequality constraints are binding if Si=0. Now, there is a
possibility from this equation there is two condition one is your 𝜆′𝑖 = 0 or Si=0. So, if
Si=0, then we are getting this constraints and we can say our inequality constraints is
binding again the binding, which I explained in the previous one.

Binding means equation will lie on; that means, that much satisfy in all the case
otherwise it will a region. Now, if your 𝜆′𝑖 = 0 means Si≠0. So, we can say𝑔𝑖 (𝑥̅ ) ≤ 0
means here this constraint is not binding or we can say this ith inequality constraints is
free to be non binding. To see this, let us see another example how we are going to solve
and how we are going to get this, so what normally we do? We normally, taking one
condition here λ = 0 and then we can solve all these. Then we can check whether it is
satisfying or not if it is violating then we have go for another one, and then we can again
solve it.
(Refer Slide Time 14:46)

To understand this, let us see here again I already explain here, let us see this one
example again that which we discussed in the previous lecture that here our objective
function is your 𝑓(𝑥̅ ) = 𝑥12 + 4𝑥22 . If you remember, it was a 𝑓(𝑥̅ ) that is to be
minimized and subjected to our constraints equality here ℎ(𝑥̅ ) = 5 − 𝑥1 − 𝑥2 = 0 this
is nothing but your ℎ(𝑥̅ ). We had the g x function is remember this was nothing but your
𝑔(𝑥̅ ) = 4𝑥1 + 𝑥2 − 14 ≤ 0. So, this is your standard formulation that we solved it
using Kuhn Tucker condition. Similarly, here I am going to solve this problem using the
slack variable formulation approach.

Using the first criteria that we have to write the complete this L matrix that is augmented
objective function here L that I can say 𝐿 = 𝑥12 + 4𝑥22 + λ(5 − 𝑥1 − 𝑥2 ) + λ′ (4𝑥1 +
𝜕𝐿 𝜕𝐿
𝑥2 − 14)here. Now, we have to apply first and first one is your = 0, and = 0. If
𝜕𝑥1 𝜕𝑥2
𝜕𝐿
we derivative this, what we are going to get here it is a 𝜕𝑥 = 2𝑥1 − 𝜆 + 4λ′ .
1

𝜕𝐿
For this we are going to here = 8𝑥2 − 𝜆 + λ′ . Now, this is your condition number
𝜕𝑥2

first, which I wrote in the just if you remember here this one is your first condition.

Now, second we it is nothing but your equality constraint and we are just going to be,
sorry here since this addition we have to make here that with your S12 because we are
going to add because we are trying to make its equality constraint. This is less than that
we are making s square added her and that differentiation with x1 x2, it will be 0 so that
equation will remain the same. Now, second here the condition here just we had to write
5 − 𝑥1 − 𝑥2 = 0.. So, we are having equation 1, equation 2, equation 3 and now we are
going to derivate that is the third condition. Here, we are going to derivate with the
lambda prime and this is nothing but we are going to get here 4𝑥1 + 𝑥2 − 14 + 𝑆12 = 0
that is your equation 4.

The fourth equation that we are going to derivate with the Si means we are going to get
λ′ 𝑆1 = 0, now how to solve this? To solve this, we have to again go for taking one value
here 0 that we have to take whether this λ = 0 or not or we can take the first take λ =
λ’=0, means here you can say either this is 0 or this is 0, then only we can get this
expression that is 5. So, let your λ′ = 0 if your λ′ = 0, now we can solve this means we
have four equations now put this value.

So, we are getting 2𝑥1 − 𝜆 = 0. we are getting 8𝑥2 − 𝜆 = 0. and for this we are getting
as it is 5 − 𝑥1 − 𝑥2 = 0.. So, from here we are having three equations and we can solve it
and that after solving this if we solve here we are going to get the value here x 1 is equal
to means I will get 𝑥1 = 4. 𝑥2 = 1. and 𝜆 = 8, if you solve this, you will get this value.
Now, put this value in your equation number 4, so putting this we are getting 4 into 4
plus 1 into 1 minus 14 plus s 1 square is equal to 0, what we are getting?

Now, we are going to get this s 1 is square is equal to here 16 plus 17 means 3 means we
are getting minus 3 and then it is not feasible because here S as I said S is always
positive, we cannot get any imaginary term. So, here this S1 this is not feasible you can
say not feasible. It means here λ′ ≠ 0 means we are going to have this Si=0 So, if you are
going to put this Si=0 S 1 here we have taken S 1 because only one constraint and then
we had to again solve here, so what we are going to get? If this is not 0 means this is 0
means lambda not is not 0.

So, we had the equation first, now we can write is 2𝑥1 − 𝜆 + 4𝜆′ = 0.. Here, again your
8𝑥2 − 𝜆 + 𝜆′ = 0., your equality constraint 5 − 𝑥1 − 𝑥2 = 0.. Now, here this S1=0 in this
case, so we are going to get 4𝑥1 + 𝑥2 − 14 = 0.. So, now we have the four variables,
variables are x1, x2, λ and λ’. We are having four equations four unknown and then we
can solve uniquely and if you will solve you will get this𝑥1 = 3. 𝑥2 = 2. and 𝜆 = 10/3
and your this slack 𝜆′ = 58/3.
So, this shows that this result is similar to your Kuhn Tucker condition means you are
getting same optimal value here and there is no change. So, now this is same as the Kuhn
Tucker condition and you can say the formulae is slightly different. So, both are equally
applicable for the solving the non-linear optimization problem, I mean both equality and
inequality constraints. Now, let us see that again sometimes if we called hard inequality
constraints and sometimes it is called soft inequality constraints. This inequality
constraints here in the power systems we are having the two type inequality constraints
like hard inequality constraints which cannot be violated and that is like transformer taps.

We can change, we can exceed otherwise transform will burn, and we also having some
soft inequalities, if there is slight variations in the inequality that is not harmful and small
violation is allowed. For example, if I am putting the tapping of the transformer one
should be less than this tapping value. So, it is called hard inequality must satisfy for
example, voltage if I can say voltage at any, it is less than 1.05 per unit if it is voltage is
1.055 hardly matters and that is allowable. So, this is called your soft inequality and this
is called your hard inequality. So, the first handling the soft inequality, normally we use
the penalty function methods means, and again what is that penalty function methods.

The penalty function methods are slightly here we in objective function, here in 𝑓 (𝑥̅ ) we
add that a penalty function we can say W, and then we just for this equation what we do
we write this we can go for (Vi - 1.05) term here. So, what happens this is the penalty
factor based on that is slack variation then objective will be slightly different. So, this is
handled, this inequality in this fashion. So, instead of going foe the 𝑔(𝑥 ) in this fashion
here or we can say 𝑔(𝑥 ) we used here, we can go for the penalty function for this one for
the soft inequality.

For the hard inequality, we have to consider the 𝑔(𝑥 ) and we have to be meet we have to
meet that one very strictly and that is why it is called hard inequality. So, in power
system both are allowed, so if you are going for the soft, then you can adding the
objective function with the some penalty function W. Again, this weight factor is very
important if you are using very high value of weight factor what will happen? You are
going to minimize this, then your optimization problem will shift from here to here, it
will try to match this one first, and then it is this one. So, if you are using very low value
of this one, then what will happen? If your low value of this one causes violation of these
constraints.
So, normally some optimal some value between the higher and lower side value this
weight factors are decided and these are added because again you have to see what is the
value of this, what is the unit of this? Suppose, you are adding something thousands here
in five, so always it will try to minimize thousand rather than five because it has no
meaning, it will ignore this optimization this completely. So, this weight factor or
penalty factor is also very important and this factor value is basically chosen very
judiciously.

(Refer Slide Time 25:44)

Now, come to other programming methods that is your linear programming approach for
solving some optimization problem.

As I said for the optimization program, this method is very fast compared to your non-
linear programming approach because non-linear programming approach every time just
you are solving some set of differential equations. You are sometimes solving the inverse
of that because solving the whole set of equations every time you are inverting in each
iteration. So, that requires very huge time, so the linear programming approach are very
fast, but, sometimes converges problem you get problem some problematic give some
diverged solution. Also, in the actual sense in our power system, we do not have the
problems of linear in nature.

Our power system is a non-linear system, then how you can assume that we can have the
linear system. If you are having the linear, your means linear programming methods can
be only applied if your objective and your constraints both are here linear. So, you have
to have your objective linear and your constraints must also be linear if any one of this in
it is non-linear, and then you have to use your non-linear approach. Now, since we are
not having the linear constraints very few, mostly we are having functional constraints
their non-linear highly non-linear you can say objective functions are also almost non-
linear, then, what we do? We normally linearise the problem at the operating points.

Then, we move and solve the LP programming, so it is called LP methods they are very
popular. So, the LP we have to linearize first then you apply this linear programming
methods to optimize your problem and then you can move ahead. Again to understand
this, let us slightly go into the linear programming formulations, and then we will see
some methods to solve the optimization problem. Here, in this case here your objective
function as I said, it will be linear, means it is a linear combination of your state variable
that is your x1 to x2 here again this x. So, we are having some coefficients some cross
functions associated with all this your state variables that is x1 to xn and that we are
adding here.

So, this your objective function subjected to your sets of equality and non your inequality
constraints, what here? I am doing here if it is equal to 0 or b1 b1 to bn value may be 0, so
then it is your equality constraints. So, here it can be equal or it can be less than this, we
can write in the general formulation in this fashion. Your problem may be combined of
this, so above inequalities m in number and in upper form here this is upper form.
(Refer Slide Time 28:49)

So, what do? We can write this minimization function here that is c t the transpose of the
cross function with the state variable subjected to AX. Again, we have formulated this
here this complete matrix, we can write here a x less than or equal to B and a is a matrix
that is a coefficient a11 a12 means we are getting this value is your a11 a12 a13 to your a1n
and here again an1 to here ann. So, we are having here m we have taken, so m n we are
taking this and this we can we have that you can see here. So, A is your m cross n matrix,
in this case also all the x here, they are greater than 0 and at the same time the b we also
assuming B is greater than 0.

So, this is your standard formulation of this, now all these inequalities here they are
nothing but they are the giving some regions, and based on that we can have the solution
should lie in this feasible region that is called your feasible region. So, this is nothing but
this is a contour of objective function that is equal to one in this region. This is your this
parallel line showing your contour of objective functions that is for k and for different
value, what will be this value, when it will be for given value k? We are having always
lines parallel, now if your objective function is parallel to any of the constraints line, then
optimal will be not unique means for example, here your this constraint is parallel to this
what happens then you are you will not have the unique solution.

However, in the previous case here you can see if it is not parallel, so what we are
getting? We are getting for any value here, this value is solution of the optimal solution
or this is your optimal solution, again depending upon whether you are minimizing or
maximizing the function.

(Refer Slide Time 31:00)

So, this here if it is a any of your objective function here contour these are the objective
function if it is parallel to the any of constraints here this is was constraints are here. So,
it was parallel, so it is called it will not have the unique optimal. So, in standard normally
the LP programming techniques I approach we write here 𝑓 (𝑥 ) is the cost function
multiplied by state variable subjected to x is less than or equal to b and x will be always
positive and B is also positive will represent in terms of this. No doubt here let us
suppose you are having x is your B and if it is a b is negative, what we can do? We can
multiply by this negative, we can multiply by this and we can change in this fashion.

So, this value will become the positive, and then we can write again here means we can
convert this inequality from this side to that side and this is always true. So, it is very
well true that any here the function that is for example, I will show you. Let us suppose
your x1 + x2 here that is your less than I can say greater than x1 + x2 >5, what I can do?
This is greater than 5, so this can be converted here, this is your x1 + x2, again I am going
to for minus here and then it is this minus let us this minus I, so this minus I want to
convert into positive.

So, I have multiplied by the minus, and then it will be positive, but this sign will also
reversed if you are multiplying by minus. So, this is your function and again you can say
you are having this term, some case it is not possible, then you have to add some slack
variables and other things just we will discuss later on.

(Refer Slide Time 32:54)

So, if you are this any problem should be connected means converted to any standard
form now we are having the different possibility. You see your A matrix was your m
cross n matrix means you are having the x vector is your x1 to xn you are this a x here
that is less than b here this equation was your m in number. Now, if m=n then we will
have one unique solution, if you are having m>n, then you will also have one solution,
but if you are having m<n, in this case we normally go for optimization and some will be
of course will be in this case.

If m>n, some will be dependent on another equation and then we can have the rank of n.
So, if m<n then optimum is needed because we are going to have many solutions. In this
case, if m<n, so there is a possibility that we have a infinite number of solutions and then
we can have only further optimization. If we are having equal to this, then your function
here that we are going to minimize here, it will be 𝑓(𝑥 ) and we have to only solve here
AX that is a AX=B, and then you can solve it. So, the problem arise only when m is less
than n, so, we can again let us see the various methods.
(Refer Slide Time 34:29)

Let in the case one the inequality in the upper form as I said here this we are having the
inequality and the objective function of this. So, we can add a non negative slack
variable to convert the inequality constraints to equality. We want to convert in terms of
here equality constraint we want to solve. So, the standard form which I showed here you
that here normally we try to go for equality constraint here. In that case we see what this
number of equations here is and number of variable. I will come to that point again how
it will be converted then it will be clearer. Now, this is then we can add here negative
slack variable here, and we can make this y i is added.

This value is positive always, and then we can have this equality constraint, and this yi is
called slack variable, and now we are having total number of variables are now increase
to (n + m). It means we have all this n equation here n variables we are adding. So, and
m here variables we are adding this yi is m in number, so the total variables that is now it
is going to be x and y it is your (n + m) variables.
(Refer Slide Time: 35:51)

If it is in lower form as I said if it is in the fashion, what will you do? Because we are
taking bi is positive you must remember, so here what we can do? We can subtract this
because this is a surplus and then it is yi is called surplus variable. In the case three, what
we can do? If this is your xk not bounded to be non negative or free variable means if it is
not bounded. It means xk may can be a negative or positive as I said x should be greater
than or equal to 0, but I can say it is a non binding x can be any value there is no binding
here. Here, we are putting the binding some constraints we are putting if it is a free
variable, then this xk can be changed as another variable for any kth variable (u – v) and
these 2 will be your non negative value.

If it is more than this it will be negative if this is u is more than v, then it will be positive.
So, it can be free variable and then what we are going to do we are just creating this free
variable into this two variable. Now, here I can say your u k must be k > 0 and your vk
also is greater than 0. So, one free variable now we are just going to have the two
variables means again we are increasing the number of variables accordingly.
(Refer Slide Time 37:18)

So, in standard form here as I said this is your equality constraint, now we can change in
this fashion because we can this equation here, we can this xk – xmin we can go this side
and then we can convert and you can get it. So, elimination of the free variables there is
several variables extra, so we can eliminate and then we can use some optimization
program. So, now let us see some examples for this few variables, let us suppose we
have a function that is a minimization of 𝑥1 + 3𝑥2 and subjected to the constraints that is
here 𝑥1 + 𝑥2 ≤ 3 and your 2𝑥1 + 3𝑥2 < 8.

Now, it is said here x1 is free variable is x1 is free variable free variable, so what we have
to do? We have to convert the whole problem into that we can have the total number of
variables we can exceed. So, in this normally we add some more constraints means I can
say this here we have to go for in that fashion means we have to add first slack variables,
what I am going to do? Here, this 𝑥1 + 𝑥2 + 𝑥3 = 3 means we have to convert this
inequality to equality constraints by adding some extra variables. Now, similarly, here
your 2𝑥1 + 3𝑥2 + 𝑥4 = 8.

So, now here we have added two more variables as I said that can be converted, so here
x3 and x4 are the slack variables and now the total number of variables now increased to
4. To see that we are having some free variables free variables that I am going to say that
your let us suppose x1 is your free variable. So, x1 here I can say your (u1 - v1), so in that
we can replace this one here x1 x2 and then we can solve, so now what we can do? You
can say again in terms of different fashion we can replace. So, here we can put the x one
here we can put x1 here and now our whole problem can be formulated that your 𝑓(𝑥) is
your CTX’. Then, here AX=B and here this x variables are greater than 0, so this your
standard form.

Then, we can solve without any problem, now, some times when we can eliminate that
variable that we should be free there is no constraints here we are putting x. So, here we
have to eliminate x1 we have to put these values and then we can solve it. So, this is your
standard problem and then to this here just see how we can go to solve it because this is
the region this is the equality constraints some equality constraints here different are
coming. So, we are going to solve, and we are going to get the values, and after getting
that value, we will see which value will be the optimum why I am saying this? Here, you
know this your x will be now several variables in this case, how much it is now 4 your
equation is only 2.

It means I can say here this is a condition, when m<n, it shows that we are having the
infinite number of solutions. Now, in that we are searching which one is your minimum
or maximum value as per this one? So, we have to take the state vectors those will the
cost function minimum. So, as I said if you are having m<n then only it is possible that
we can have the optimum solution and then we can say. So, here we are going to see this,
so we have to solve first this means we have to get the various solution, and then at we
will see which one is minimum. Some methods are there like simplex method, it will
take care along with the solution of this so that you can say which are the variables are
basic variables which are not basic variable.
(Refer Slide Time 41:50)

So, the solution of the simultaneous equations here now you are having here equations
and then let us sees this how much. So, if number of equations is equal to number of
variable solution will be unique here. So, number of here n equations n variables, so
unique solution will get means your x1 to your xn will get one solution here.

(Refer Slide Time 42:14)

We can use the various methods and normally the methods are called the Gaussian
elimination method. That basically use the pivotal method normally what we do we put
in the canonical form means we can have this here unity and then we can go for the
simplifying the equation. We can get the one that x1 here and other coefficients will be 0,
then we will go for x2 another will be 0. It means we have to do some sort of
multiplication subtraction addition so that we can get this type of means we have to see
only this side automatically this value will be coming accordingly.

So, we can replace that standard form here this equation here in terms of here, and then
you can see from here what we are getting your x1 will be your b1, x2 will be your b2‘ and
xn will be bn and this is your solution. So, this is called use of pivot operation, and then
we can above form of simultaneous equation is known as the canonical form, because
here you can see what you are getting unity, others are 0 0. Here, your x1 to your xn, and
then you are getting here b1’ to bn’, so this is your canonical form means I matrix.

(Refer Slide Time 43:31)

If total number of equations is less than number of variables m in this case, then how will
you solve, to solve this again what normally do we have to select the m variables because
your number of equations are m. So, we are trying that since your number of equations
are m, so we are trying to form this whole matrix this.

Now, you’re A matrix is now can be divided into here m cross m remaining here m cross
m minus n is the remaining is coming here. So, we can divide this matrix means here you
can say we have the matrix that is a square matrix a11 a12 a1m here am1 am1 two amm and
then we can write this. So, the first column is linearly independent, we can apply the
pivot operation to the first m column. For example, here this is 0, then we cannot apply
the pivot because here if you are dividing by 0, it is not defined. For example, if you are
having your x2 + x3 = 0 you have x1 + x3 = 5.

You have to choose this equation first because here this x1 is 0 here, so you cannot apply
the pivot because the division of 0 is undefined. So, you have to just change here and
there you have to take this x one here and then so on s forth or you can change x1
variable here and there. So, we can choose, so that is why it is said if a first column is
linearly independent, then we can apply the pivot to the first m column because we want
that this matrix should have the rank m, otherwise we cannot go for here the canonical
form. It means it must have the rank m of this portion, then we can take those coefficient
and variables, and then we can go for the pivot operation.

(Refer Slide Time 45:33)

So, if we are going for the pivotal, and then these are known as the pivotal variable you
can say we have done the same pivoting operations means x1 x2 x3 and xm. Remaining,
just we have change in this and we can say it is a non pivotal variables.

So, it is in your canonical form, now what we can do if you put all these values to 0
means xm to xn 0, then we are getting x1 is equal to this x2 equal to this. So, this is one
solution means x1 to xm the values just regards from here, remaining we put 0 because
this we are put in 0 because our x is also greater than or equal to 0 means it is a 0 is one
variable. So, this is one solution as I said if your number of equations are less than
number of variables, you will have infinite solutions number of solution are infinite. So,
this is one solution and this solution is known as the basic solution basic solution means
we have short listed that the out here n variables.

We have just chosen these m variables so that we can have the pivotal operation we can
form in the canonical form as I have shown here. Then, we can remain we can go and
another side and then we put the non pivotal variables 0, pivotal variables we can get
here and that solution is called as basic solution this one solution. So, if bi > 0, xi will be
non negative and that because our condition is x is greater than this vector is greater than
0. Then solution is called feasible solution, but there may be possibility that this is not
true then it is called non feasible solution basic solution.

So, the hence solution is basic, then it is called basic solution feasible because we are
getting this value this positive means xi is your non negative or positive that is satisfying
this. So, this is your feasible solution as I said s many solutions will be there someone
will be non feasible and other may be your feasible solution.

(Refer Slide Time 47:45)

So, what we can do? We can go for this is your identity matrix remaining is c converted
matrix and we just converted here in this form let be prime. So, we put this xb that is a
non basic variables means x is divided means your x here it is nothing but x basic
variables, and then others are your x non basic variables. So, non basic variables we put
0 basic variables, we get the values from here. So, this xnb here put 0, and we will get the
force to 0 and called non-basic variables and then xb remaining that we get the solution
from here they are called basic solution.

Even though without doing any iteration, without doing any manipulation in the first
equation can be converted into this form. Here it is your xb and xnb, so we can simply this
matrix without doing any pivot we can write the B and here another is your C matrix this
B is your m cross m and this your matrix here m cross (m – n) matrix. This B must be
non singular, otherwise you will get not you will not get the basic solution you cannot
invert it here you cannot go for the pivotal operation. So, this matrix B is known as the
basic and C `is called the non basic matrix.

So, the matrix which you are going to choose where you are going to operate the pivotal
operation you want to make the canonical or identity matrix here this 1, 1, 1, 1, this is
called identity matrix similarly, elements are 0, then it is called your basic. For the basic
solution, the necessary condition is the rank of A should be m, means the total here the
rank of A should be m in complete. So, here the B may not be m, but we have to search
we have to change here and there we have to choose the variable, here accordingly in
such a fashion that the B matrix should also have the rank m. So, if the rank of A is less
than m, so of course, you will not have the basic solution here.

So, the resulting solution is the basic solution if the solution satisfies all the constraints is
called the basic feasible solution. So, all the constraints we have to satisfy then that
solution is called your basic solution.
(Refer Slide Time 50:16)

Sometimes it is a degenerative solutions, what is the people called the degenerated

solution is the some of the variables. Assume 0 value, the basic solution of the basic
variable is 0 means some of the basic variable even the suppose it is coming 0 means
here you are just xb is 0, it is valid because your basic solution as I said xb is greater than
equal to zero it is satisfying this. So, if it is 0, then it is called degenerated basic solution
because if you see here as I said means here we are getting the solution here, this is a x m
means it is your basic solution. Here, some of this value may be 0 here we are forced to 0
here we are getting 0.

So, in that case it is called some degenerated solution, and it is called degenerated basic
solution. Now, the basic solution is optimal solution in that case this basic solution is the
optimal solution means it will give your basic optimal solution. If your solution which
you are getting is optimal one, then it is called your basic optimal solution.

If it is satisfying all the constraints in addition to the optimal is known as the basic
optimal feasible solution. So, the optimal as well as feasible solution we want for your
complete problem. So, for that we have to solve and we have to see some algorithms are
there, there are some fundamental theorems are there those are used to see whether the
solution is feasible and optimal feasible. It means all combinations we can say feasible
non feasible optimal non optimal feasible this non optimal feasible or non feasible, of
course there will in optimality there.
(Refer Slide Time 52:13)

So, let us see some fundamental theorem in this LP problem, so first fundamental
theorem is any feasible solution of LP that is a linear programming problem correspond
to the basic feasible solution. So, any feasible solution that is you are getting that will be
your basic feasible solution. Second is your any optimal feasible solution will be your
basic optimal feasible solution here just we have add the basic. So, the basic solution just
we are getting it will be the optimal basic feasible solution. So, the number of solution as
I said the number of solution that is a possible here the nCm means you are going to get
this much solution. If your number of here n equals to 5 m equal to 2 the maximum
number of basic solution will be your 10.

So, from this one will be optimal, so the basic solution that we are going to get it is your
ten in number although the solution is infinite. So, basic we will be only your ten
solutions, so in that ten that one will be optimal. We have to find that optimal value so
that we can say our linear programming our standard here that is f(x) is minimum
subjected to constraint that is our constraints is equal to b and here x is greater than B
and of course here B is also greater than equal to 0.
(Refer Slide Time 53:35)

So, this we have to search and the various methods are there first one is very popular that
is called simplex method. This method is start from the basic feasible solution means we
have to get the basic feasible solution first and then we have to see the optimality. For
example, that in phase one it will computing the basic feasible solution, then in phase
two it is searching for the optimal basic solution. So, it is said that it is start from the
basic feasible solution and wants to maximize the reduction in the objective function at
the same time. So, this is your standard problem that is a function that to be minimized
that includes here x’ all the including slack and surplus, means we are making that here
equality constraints.

Then, we are having the constraints here, so we are having m in m rows and now n
variables and we have to solve this. So, we have to solve the basic solution here and then
we have to see the optimality at the same time.
(Refer Slide Time 54:33)

What we do in this changing from the basic to non basic always we extend the variable
the basic variable to non basic variable. Again, getting back to another basic feasible
solution means we have to keep on same time we have to see which whether it is optimal
or not. So, it is not possible that we have to solve one basic solution, then search you
solve one then see the optimal. So, you are suppose for in a simple problem as I said in
the variables here it is only this your number of variables five equation is two you are
getting ten basic solutions. If your number of variables are hundred and here this two you
will get tremendous variable

So, it is not possible to every time to solve this and then you can search for optimal. So,
there are some techniques that even though while changing your basic solution here, and
there from basic to non basic variables we can check whether it will be suitable or not for
our optimal value.
(Refer Slide Time 55:27)

So, that is why here the question arise that which non basic variable should be the basic
and which basic variable should be non basic. So, we have to choose in that fashion to
this simplex method gives some idea, so at least one element must be positive in a vector
subjected to this will be the minimum. So, we have to choose in that vector in that here
the various vector that equal to here b s upon a and this will be giving your minimum.
For example, here you are having x1 - x2 + x3 >5, let us go to your two, so which one will
be the basic? I am telling you have to see here this is a positive this is positive, so you
have to see that here two other positive.

Now, which one will be the basic variable you have to divide by 2 here how much
getting this is less it is one, so this will be your basic variable. So, there will be more
number of one column which will be at least one element positive as I said several
columns, now which will be chosen and we will choose which require the minimum cost.
So, we have to choose according to the value which will give the minimum cost, and
then we can go for the simplex table and then this simplex programming give idea that
we have to go for the minimization of your objective function.

Now, I have to now recap is this module because we saw in the previous two lectures we
saw the HVDC we also saw the curves and unit commitment problem, even though I just
briefly introduced the optimization that is economically problem. Then, I discussed the
various optimization techniques that is non-linear, and also I get some idea about this
non-linear programming prose.

(Refer Slide Time 57:18)

In this optimization algorithm, just I said if you are function here is f(x) is a single
𝜕𝑓
variable then the necessary condition for your this optimal here this 𝜕𝑥 = 0 is equal to be

0. You have to obtain solve this you have to get the x, and then you have to see double
𝜕2𝑓
differentiation of your > 0. So, this is the necessary condition for the sufficient
𝜕𝑥 2
𝜕𝑓
condition, it should be minimum global minimum here we have to go for the 𝜕𝑥 = 0 and
𝜕2𝑓
your > 0. So, this is for single variable x only one variable here there is a no need to
𝜕𝑥 2

go for partial because its one variable. So, I can say it is a direct derivative, we can use
no partial because other variables are not there.

However, if you are going for the multi variable system, in that case we have to again
𝜕𝑓
this necessary condition will be your 𝜕𝑥̅ = 0 will be your 0. So, this will be the 0 at x =x*

that is your optimal value, and we can solve it and that it also called the ∇f =0. Another
here that is your del f here upon del x square this should be positive semi definite matrix.
So, here this matrix basically it is Hessian matrix called H matrix is called and it is also
denoted by here ∇f, and then we have to see.
 𝜕2𝑓 𝜕 2𝑓
Basically, this matrix x is nothing, but, here 𝜕𝑥 2, 𝜕𝑥 and so on so forth. So, this is the
1 1𝑥2

h square matrix and then it must be positive definite. Now, for the sufficient conditions
this must be your not semi definite it must be your positive definite variable, and then we
can solve and we can say our global optimality. So, here in this module we saw the
various power flow techniques and various optimization techniques are used to control
the power flow to achieve certain objective, may be minimum cost minimum emission
and other variables are well.

Thank you.