SOLVING EDP WITH SOCP

ADEBAYO Rilwan L

June 11, 2024

1 Objective function

The cost function of each generating unit i, is represented by a quadratic equation, a rectified sine term

is added to model the valve-point loading effect 1.1. The EDP over the generating units considered is

formulated as:
n
X
Ci (Pi ) = ai Pi2 + bi Pi + ci + |ei sinfi (Pimin − Pi )| (1.1)
i=1

Where ai , bi , ci are cost coefficients of the generators. ei and fi are the fuel cost coefficients of the ith

unit with valve-point loading effect

Figure 1.1: EDP with Valve-point load effect

The objective function is subject to constraints

1
 1. Power Balance -(The equality constraints)

n
X
Pi − PD − PL = 0 (1.2)
i=1

2. Generation limit - (the inequality constraints)

Pimin ≤ Pi ≤ Pimax (1.3)

The objective function is a non-convex function due to the rectified sine term. The first task is to convert

the non-convex term to a convex function.

2 Decomposition for Valve-point load effect

Let the rectified sine function part of equation 1.1 be represented by a quadratic function Q(Pi ) such

that it be fitted to the valve-point load effect;

Q(Pi ) = |ei sinfi (Pimin − Pi )| (2.1)

Plotting the rectified sine term is as follow;

% Plot of the rectified sine function

% The valve point coefficient

e = 300;

f = 0.0315;

P_min = 100;

P_i = linspace ( P_min , 600 , 100) ;

F_vi = abs ( e * sin ( f *( P_min - P_i ) ) ) ;

plot ( P_i , F_vi , ' Color ' , ' red ') ;

title ( ' Rectified sine function | e_isinf_i ( P_i ^{ min } - P_i ) | ') ;

xlabel ( ' Power generated ( P_i ) ') ;

2
ylabel ( ' Cost ( F_i ( P_i ) ) ') ;

The plot of the rectified sine function is shown below in figure 2.1

Figure 2.1: Rectified sine function

Fitting the rectified sine wave to the quadratic curve, Q(Pi ) becomes

Q(Pi )ij = αij Pi2 + βij Pi + γij (2.2)

We do the poly fitting to get values for αij , βij , and γij . In order to take the curve, one at a time, We fit

over the zero points or non-differentiable points. So, the first curve is between Pimin and the next point

the cost is zero.

We find the non-differentiable points at,

sin(f ∗ (Pi − Pmin )) = 0 (2.3)

We can call that point the zero point ndth and we know from trigonometry that sine function is zero at

nπ.

so

f ∗ (Pi − Pimin ) = nπ (2.4)

nπ
ndth = Pimin + (2.5)
fi

3
Fitting the first curve which is between Pimin and the ndth equation when n is 1.

% valve point data and also

n =1;

e = 300;

f = 0.0315;

% P min

P_min = 100;

% The zero point

nd_point = ( P_min + ( n * pi / f ) ) ;

P_i = linspace ( P_min , nd_point , 100) ;

F_vi = abs ( e * sin ( f *( P_i - P_min ) ) ) ;

% Fit a quadratic polynomial to the data

p = polyfit ( P_i , F_vi , 2) ;

% Extract the coefficients

a = p (1) ;

b = p (2) ;

c = p (3) ;

% Display the coefficients

disp ([ 'a = ' , num2str ( a ) ]) ;

disp ([ 'b = ' , num2str ( b ) ]) ;

disp ([ 'c = ' , num2str ( c ) ]) ;

% Plot of the fit

F_vi_fit = polyval (p , P_i ) ;

plot ( P_i , F_vi ) ; % Original data

hold on ;

plot ( P_i , F_vi_fit ) ; % Fitted curve

4
legend ( ' Data ' , ' Quadratic Fit ') ;

title ( ' Quadratic Fit to Data ') ;

xlabel ( ' P_i ') ;

ylabel ( ' F_ { vi } ') ;

grid on ;

The plot of the quadratic fitting is shown in figure 2.2

Figure 2.2: Quadratic fitting for first curve

The αi1 , βi1 , and γi1 gotten are The coefficients gotten for the first curve are;

a = -0.12389

b = 37.1325

c = -2488.6357

>>

The quadratic fitting would be done for the j th curve, and the coefficients for the curves would be found

as well.

The objective function combining the quadratic fit of the rectified sine function is shown in equations

2.6 to 2.8;
n
X
Fi (Pi ) = ai Pi2 + bi Pi + ci + Q(Pi )ij (2.6)
i=1

n
X
Fi (Pi ) = ai Pi2 + bi Pi + ci + αij Pi2 + βij Pi + γij (2.7)
i=1

5
 n
X
Fi (Pi ) = (aij )Pi2 + (bij )Pi + (cij ) (2.8)
i=1

Where; aij = ai + αij , bij = bi + βij , , and cij = ci + γij

For j th curves, it means we’d have j th sets of quadratic functions, and have to find the max first of the

functions since the quadratic curves would have a maximum since it opens down. Our objective function

is 2.9;
n
X
Ci (Pi ) = max{(aij )Pi2 + (bij )Pi + (cij )} (2.9)
i=1

3 Handling the max function

For the cost function is,

Fi (Pi ) = (aij )Pi2 + (bij )Pi + (cij ) (3.1)

Let the auxiliary variable yi capture the maximum value across the segments such that:

yi ≥ aij Pi2 + bij Pi + cij (3.2)

yi − aij Pi2 − bij Pi ≥ cij (3.3)

  yi 
 
 ≥ cij
 2
1 −aij −bij  (3.4)
Pi 
 
Pi

4 Second Order Cone Programming

A second-order cone program to solve a quadratic objective function and SOC constraints.

min xT Qx + cT x (4.1)

s.t. : aTi x ≤ bi (4.2)

Formulating the EDP in the SOCP

aij Pi2 + bij Pi + cij

si ≥ aik Pi2

4si ≥ 4aik Pi2

(si + 1)2 − (1 − si )2 ≥ 4aik Pi2

4aik Pi2 + (1 − si )2 ≤ (si + 1)2

6
The second-order cone constraints (SOCP), can be written in the norms as follows:

 
√
2 a P
i i
≤ si + 1 (4.3)

 
1 − si
2

4.1 The EDP Problem

The EDP problem formulated in the SOCP becomes

yi ≥ aij Pi2 + bij Pi + cij (4.4)

yi ≥ si + bij Pi + cij (4.5)

min yi (4.6)

Subject to:

1. The SOC  
√
2 ai Pi 
  ≤ si + 1 (4.7)
1 − si
2

2. The max contraint

aTi x ≤ bi (4.8)
 

  yi 
 
−1 −bij   ≥ cij (4.9)
 
1  si 
 
Pi

3. Power Balance -(The equality constraints)

n
X
Pi − PD − PL = 0 (4.10)
i=1

4. Generation limit - (the inequality constraints)

Pimin ≤ Pi ≤ Pimax (4.11)