Application of Multi-Objective Evolutionary Algorithm for Optimal Reactive Power Dispatch with Flexible AC Transmission System Devices

Ismail MAROUANI^*, Tawfik GUESMI, Hsan HADJ ABDALLAH, and Abdarrazak OUALI

^*Sfax National Engineering School, Electrical Department. BP W, 3038 Sfax-Tunisia

E-mail*: ismail.marouani@isetks.rnu.tn

Abstract

Because their capability to change the network parameters with a rapid response and enhanced flexibility, flexible AC transmission system (FACTS) devices have taken more attention in power systems operations as improvement of voltage profile and minimizing system losses. In this way, this paper presents a multi-objective evolutionary algorithm (MOEA) to solve optimal reactive power dispatch (ORPD) problem with FACTS devices. This nonlinear multi-objective problem (MOP) consists to minimize simultaneously real power loss in transmission lines and voltage deviation at load buses, by tuning parameters and searching the location of FACTS devices. The constraints of this MOP are divided to equality constraints represented by load flow equations and inequality constraints such as, generation reactive power sources and security limits at load buses. Two types of FACTS devices, static synchronous series compensator (SSSC) and unified power flow controller (UPFC) are considered. A comparative study regarding the effects of an SSSC and an UPFC on voltage deviation and total transmission real losses is carried out. The design problem is tested on a 6-bus system.

Keywords

Multi-objective optimization; Evolutionary algorithms; Power flow; SSSC; UPFC; Voltage profile.

Introduction

The ORPD problem is considered as a MOP. It consists to improve the voltage profile and minimize the real power loss in transmission lines under several equality and inequality constraints. Such as, load flow equations and security limits. To maintain the load buses voltage within their permissible limits many technical methods are proposed [1, 2], such as, reallocating reactive power generation in the system adjusting transformer taps, generator voltage and switchable reactive power sources. To minimize system losses, a redistribution of reactive power in the network can be used [3].

Because the recent progress of power electronics, FACTS devices have taken more attention in transmission power systems. They have the capability to change the network parameters with a rapid response and enhanced flexibility, such as, improving voltage profile and minimizing system losses.

Some types of those devices, are, static synchronous series compensator (SSSC), static synchronous compensator (STATCOM) and unified power flow controller (UPFC).

SSSC is considered as a controllable voltage source inverter that is connected in series with transmission line. This injected voltage is almost in quadrature with the line current. Consequently, it provides a variable reactance in series with the transmission line, which, can be inductive or capacitive. This reactance controls the power flow in the line where it is introduced. The STATCOM is a shunt connected FACTS devices. It consists of a voltage source converter linked to the system via a shunt transformer. Its principal function is to ameliorate the voltage profile at the point of connection. To exploit the benefits of those two devices, the UPFC can be a combination of SSSC and STATCOM.

Therefore, in first step, the objective of the present paper is to develop a power flow model for power system with FACTS devices. The modified Newton-Raphson power flow algorithm was used [4]. Then, a new ORPD problem is formulated. The solutions of this problem are the FACTS parameters and location.

In a MOP, there isn’t one solution that is best with respect to all objectives. Generally, the aim is to determine the trade-off surface, which is a set of non-dominated solution points, known as Pareto optimal solutions. Every individual of this set is an acceptable solution.

In the literature, several methods are used to solve the ORPD. In [2], a nonlinear programming technique was used. Gradient-based optimization algorithms consist to linearize the objective functions and the system constraints around an operating point was presented in [3]. Those traditional techniques consume an important computing time and they are iterative methods. Also, they can converge to a local optimum.

Recently, genetic algorithms (GA) are very much used to solve MOP. Many researchers have transformed the MOP to a single objective problem using appropriate weights. Then, GA was been applied [5]. Unfortunately, the obtained solution depends on the weight vector used in the process. Also, this method requires a number of runs equal to the size of the desired Pareto optimal solutions.

MOEAs are used to eliminate most difficulties of these methods. MOEAs are no iterative and they give the Pareto optimal solutions in one run. Thus, in this article a no conventional technique based on MOEA is employed. It is based on non-dominated Sorting Genetic Algorithm II (NSGAII) approach. Which is an elitist approach and it can maintain population diversity in the set of the non-dominated solutions.

Static models of two FACTS devices consisting of UPFC and SSSC have been used in the present work.

Material and Method

Modelling of Power Systems with Multiple FACTS

As indicated previously, SSSC and UPFC are used in this paper.

SSSC Mathematical Modelling

Figure 1 show's the circuit model of an SSSC connected at position d between two buses n and m. So, another two buses i and j are added to the total number of buses of the system (where d is the portion of the impedance of line n-m, ).

Figure 1. SSSC location between buses n and m

In its equivalent circuit shown in Figure 2, the SSSC is represented by a voltage source V_se in series with the transformer impedance X_s [6, 7]. In practice, V_se can be regulated to control the power flow of line m-n and the voltage at buses i and j.

Figure 2. Voltage-source model of SSSC

            This last model can be developed by replacing voltage source V_se by a current source I_se parallel with the transmission line as shown in Figure 3.

Figure 3. Replacement of series voltage source by a current source

Ise = -jbseVse

(1)

Vse=rVie^jγ

(2)

r and γ are respectively the p.u. magnitude and phase angle of series voltage source.

0 ≤ r ≤ r_max and 0 ≤ γ ≤ 2π

            The power injection model of the SSSC can be seen as two dependent power injections at auxiliary buses i and j as shown in Figure 4 [8].

Figure 4. SSSC mathematical model

            The apparent power supplied by the model is calculated as:

Si = Vi(-Ise)^*

(3)

Sj = Vj(Ise)^*

(4)

            Then, active and reactive power supplied by the SSSC can be deduced from Equations (5) to (8).

Pi,SSSC = -rbseVi^2sin(γ)

(5)

Qi,SSSC = -rbseVi^2cos(γ)

(6)

Pi,SSSC = rbseViVjsin(αi-αj+γ)

(7)

Qi,SSSC = rbseViVjcos(αi-αj+γ)

(8)

UPFC Mathematical Modelling

An UPFC can be represented as shown in Figure 5. It consists of, two voltage sources V_se and V_sh, and two transformer impedances X_se and X_sh. Voltage sources V_se and V_sh are controllable in both their magnitude and phase angles [9, 10].

Figure 5. Two voltage-source model of UPFC

V_se is defined as in equation (2).

The voltage source V_se in the last model can be replaced by a current source I_se parallel with the transmission line as shown in Figure 3.

The shunt branch in UPFC is employed mainly to provide all the losses in the UPFC and the active power, P_series, which is injected with the system by the series branch.

If the total losses of the two converters are estimated to be about 2% of P_series, then, the provided power by shunt branch P_shunt will be expressed by [4]:

Pshunt = -1,02Pserie

(9)

The function of the reactive power delivered or absorbed by the shunt branch is to maintain the level of tension to bus i within acceptable limits. Finally, UPFC mathematical model can be constructed by combining the series and shunt power injections at both bus i and j as shown in Figure 6 [4].

 The elements of equivalent power injection in Figure 6 are:

Pi,ufc = 0,02rbseVi^2sinγ-1,02rbseViVjsin(αi-αj+γ)

(10)

Pj,upfc = rbseViVjsin(αi-αj+γ)

(11)

Qi,upfc = -rbseViVjcos(γ)

(12)

Qj,upfc = rbseViVjcos(αi-αj+γ)

(13)

Figure 6. UPFC mathematical model

Problem Formulation

The ORPD problem is to optimise the steady performance of a power system in terms of one or more objective functions while satisfying several equality and inequality constraints.

            In this section, we suppose that the extremities FACTS devices are referred by bus i and j.

Objective Functions

            In this paper two objective functions are used.

i. Real power loss

This objective consists to minimise the real power loss P_L in transmission lines that can be expressed as [1, 11]:

(14)

(15)

where:

N_b : number of buses;

V_k<α_k and V_h<a_h : respectively voltages at bus k and h;

Y_kh and q_kh: respectively modulus and argument of the kh-th element of the nodal admittance matrix Y.

ii. Voltage deviation

This objective is to minimize the deviation in voltage magnitude at load buses that can be expressed as:

(16)

where:

N_L: number of load buses;

: prespecified reference value of the voltage magnitude at the i-th load bus.

is usually set to be 1.0 p.u.

Problem Constraints

The problem constraints are divided to equality and inequality constraints.

i. Equality constraints

These constraints represent typical load flow equations as follows:

(17)

(18)

where:

P_Gi and Q_Gi: generator real and reactive power at i-th bus, respectively;

P_Di and Q_Gi: load real and reactive power at i-th bus, respectively;

G_ij and B_ij: transfer conductance and susceptance between buses i and j, respectively.

ii. Inequality constraints

These constraints can be summarized by:

(a) Security constraints

These include the constraints of voltage at the i-th load buses V_Li as follows:

(19)

(b) Parameters FACTS constraints

rmin ≤ r ≤ rmax

(20)

γmin ≤ γ ≤ γmax	(21)
dmin ≤ d ≤ dmax	(22)

            We should note that the vector of decision variables is U = [r,g,d].

Multi-Objective Optimization

In a MOP, there may not exist one solution that is best with respect to all objectives. Usually, the aim is to determine the trade-off surface, which is a set of nondominated solution points, known as Pareto optimal solutions. Every individual in this set is an acceptable solution.

For any two X₁ and X₂, we can have one of two possibilities: one dominates the other or none dominates the other. In a minimization problem, we say that the solution X₁ dominates X₂, if the following two conditions are satisfied [12]:

(23)

where:

N_obj : Number of objective functions;

f_i : i-th objective function.

The goal of a multi-objective optimization algorithm is not only to guide the search towards the Pareto optimal front, but, also to maintain population diversity in the set of the nondominated solutions.

In the rest of this section, we will present the elitist MOEA NSGAII. So, we must be start with a presentation of the NSGA approach.

NSGA Approach

The basic idea behind NSGA is the ranking process executed before the selection operation. The ranking procedure consists to find the nondominated solutions in the current population P. These solutions represent the first front F₁. Afterwards, this first front is eliminated from the population and the rest is processed in the same way to identify nondominated solutions for the second front F₂. This process continues until the population is properly ranked. So, can write [13]:

(24)

where, r is the number of fronts.

The same fitness value f_k is assigned to all of individuals of the same front F_k. This fitness value decreases while passing from the front F_k to the F_k+1. To maintain diversity in the population, a sharing method is used. Let consider d_ij the variable distance (Euclidean norm) between two solutions and .

(25)

where S is the number of variables in the MOP. The parameters  and  respectively are the upper and lower bounds of variable X_k.

(26)

The sharing procedure is as follows:

Step 1: Fix the niche radius s_share and a small positive number e.

Step 2: Initiate f_min = N_pop+e and the counter of front j =1.

Step 3: From the r nondominated fronts F_j which constitute P.

(27)

Step 4: For each individual  :

· associate the dummy fitness f_j^(q) = f_min - ε;

· calculate the niche count n_cq as given in [13];

· calculate the shared fitness .

Step 5:  and j = j + 1.

Step 6: If j ≤ r, then, return to step 4. Else, the process is finished.

The MOEAs using nondominated sorting and sharing have been criticized mainly for their O(MN³) computational complexity (M is the number of objectives and N is the population size). Also, these algorithms are not elitist approaches and they need to specify the sharing parameter. To avoid these difficulties, we present in the following an elitist MOEA which is called Nondominated Sorting Genetic Algorithm II (NSGAII).

NSGAII Approach

In this approach, the sharing function approach is replaced with a crowded comparison.

Initially, an offspring population Q_t is created from the parent population P_t at the tth generation. After, a combined population R_t is formed [13].

R_t is sorted into different no domination levels F_j as shown in the NSGA approach. So, we can write:

, where, r is number fronts.

Finally, one iteration of the NSGAII procedure is as follows:

Step 1: Create the offspring population Q_t from the current population P_t.

Step 2: Combine the two populations Q_t and P_t to form R_t.

Step 3: Find the all nondominated fronts F_i and R_t.

Step 4: Initiate the new population P_t+1 = f and the counter of front for inclusion i = 1.

Step 5: While |P_t+1| + |F_i| ≤ N_pop, do:

Step 6: Sort the last front F_i using the crowding distance in descending order and choose the first (N_pop - |P_t+1|) elements of F_i.

Step 7: Use selection, crossover and mutation operators to create the new offspring population Q_t+1 of size N_obj.

To estimate the density of solution surrounding a particular solution  in a nondominated set F, we calculate the crowding distance as follows:

Step 1: Let’s suppose q = |F|. For each solution  in F, set d_i = 0.

Initiate m = 1.

Step 2: Sort F in the descending order according to the objective function of rank m.

Let’s consider  the vector of indices, i.e. I_i^m is the index of the solution  in the sorted list according to the objective function of rank m.

Step 3: For each solution  which verifies 2 ≤ I_i^m ≤ (q-1), update the value of d_i as follows:

(28)

Then, the boundary solutions in the sorted list (solutions with smallest and largest function) are assigned an infinite distance value, i.e. if, I_i^m = 1 or I_mⁱ = q, d_i = ∞.

Step 4: If m = M, the procedure is finished. Else, m = (m+1), and return to step 2.

Implementation of the NSGAII

The proposed NSGAII has been implemented using real-coded genetic algorithm (RCGA) [14]. So, a chromosome X corresponding to a decision variable is represented as a string of real values x_i, i.e. X=x₁x₂…x_lchrom. lchrom is the chromosome size and x_i is a real number within its lower limit a_i and upper limit b_i. i.e. . Thus, for two individuals having as chromosomes respectively X and Y and after generating a random number, the crossover operator can provide two chromosomes X’ and Y’ with a probability P_c as follows [13]:

(29)

In this study, the non-uniform mutation operator has been employed. So, at the tth generation, a parameter x_i of the chromosome X will be transformed to other parameter  with a probability P_m as follows:

(30)

(31)

Where t is a random binary number, r is a random number  and g_max is the maximum number of generations. b is a positive constant chosen arbitrarily.

Results and Discussion

The proposed procedure for solving dispatch VAR including FACTS devices is tested on the 6-bus system [16]. The one-line diagram of this system is shown in Figure 7. The system consists of three generators at buses 4, 5 and 6. Bus 6 is considered at the slack bus. Buses 1, 2 and 3 are the load buses.

Figure 7. Test system

Presentation of the Test System

The characteristics of lines and buses are marked in the Tables 1 and 2 respectively. These values are given in p.u. considering a base power of 100 MVA for the overall system and base voltages of 100 KV. The lower voltage magnitude limits at all buses are 0.9 p.u and the upper limits are 1.1 p.u. Three cases of power analysis are considered. Case 1 assumes the study without any compensation. Case 2 assumes an SSSC between two load buses. In case 3, an UPFC is assumed too between two load buses.

Table 1. Data lines

Line		Impedance
Bus i	Bus j	R	X
1	2	0.02	0.4
1	6	0.01	0.15
2	3	0.05	0.5
2	4	0.015	0.5
3	5	0.05	0.5
3	6	0.05	0.5
4	5	0.08	0.8

Table 2. Data buses

Bus	V	PG	QG	PD	QD
1	-	0	0	0.840	0.500
2	-	0	0	0.400	0.400
3	-	0	0	0.96	0.300
4	1.025	0.200	-	0.100	0.100
5	1.084	0.100	-	0.100	0.100
6	1.000	-	-	0.100	0.100

Figure 8 shows the convergence characteristics of the load flow program after 6 iterations with a tolerance of 10^-5.

Figure 8. Convergence of the load flow

For the system without FACTS controllers, the voltage magnitude of load buses given by the load flow program, are not maintained within their permissible limits (Table 3).

Table 3. Voltage magnitudes and phase angles for the system without FACTS controllers

Bus	1	2	3	4	5	6
V	0.828	0.868	0.947	1.025	1.084	1.00
α [rad]	-0.299	-0.367	-0.195	-0.268	-0.141	0

The corresponding values of voltage deviation and real power loss are respectively:

V_D = 0.441 pu and P_L = 0.133 pu.

Resolution of the ORPD

After execution of the optimization program, the values of the voltage deviation and the real power losses for the various positions of elements FACTS, are given in Figure 9.

(a) UPFC

(b) SSSC

Figure 9. The best position for system FACTS

Figure 9 shows that the best position of UPFC and SSSC for minimum voltage deviation and minimum real power loss, is in line 1-2.

Therefore, we will give only the results corresponding to this optimal position.

Optimisation Mono Objective

To get convergence of power loss and voltage deviation functions which are shown in Figures 10 and 11, these two objective functions are optimized individually.

Figure 10 shows the convergence of power losses function versus number of generations.

Figure 10. Convergence of power losses with generations

From Figure 11, we can see that voltage deviation corresponding to the two cases converges to 0.418 pu and 0.375 pu, respectively, for the system with SSSC and with UPFC .

Figure 11. Convergence of voltage deviation with generations

Optimization Bi-Objective

In this section, the two objective functions are optimized simultaneously.

Figure 12 gives the Pareto-optimal front corresponding to the two cases. And it shows that the UPFC corresponds to the best results.

After optimization by the NSGAII approach, we have obtained the Tables 4 and 5 giving the best solutions for minimum voltage deviation and minimum real power loss respectively.

The values of r, d, V_D and P_L are given in p.u. g is in radian.

Figure 12. Pareto-optimal front of the proposed approach

Table 4. The best solution for voltage deviation

	R	g	d	VD	Corresp. PL
SSSC	0.1525	1.253	0.442	0.418	0.133
UPFC	0.1931	1.481	0.485	0.375	0.138

Table 5. The best solution for real power loss

	R	g	d	PL	Corresp. VD
SSSC	0.1537	1.128	0.461	0.132	0.46
UPFC	0.1952	1.369	0.527	0.132	0.45

Table 6. For minimum voltage deviation

	V1 [pu]	V2 [pu]	V3 [pu]
SSSC	0.9821	0.9274	0.9365
UPFC	0.9915	0.9691	0.9467

Table 7. For minimum real power loss

	V1 [pu]	V1 [pu]	V1 [pu]
SSSC	0.9778	0.9146	0.9217
UPFC	0.9870	0.9541	0.9406

The profile voltage in load buses corresponding to the best solutions given in Tables 4 and 5 are shown respectively in Tables 6 and 7. V_i is in pu.

Conclusion

In this paper, a procedure based on MOEA to solve ORPD problem with FACTS was presented. This problem consists to minimize simultaneously real power loss in transmission lines and voltage deviation at load buses by using FACTS devices. The decision variables are parameters and location of FACTS devices. The NSGAII approach is opted to solve this nonlinear MOP. UPFC and SSSC are used in this work. The resolution of the MOP shows that UPFC and SSSC have a positive effect on ORPD. Knowing that, UPFC gives the best results. The presented procedure was been tested on a 6-bus system.

            Acknowledgments

            This paper extends and completes a previous communicated research [Marouani I. Guesmi T.; Abdallah H. H.; Ouali, A. Application of a multiobjective evolutionary algorithm for optimal location and parameters of FACTS devices considering the real power loss in transmission lines and voltage deviation buses. 6th International Multi-Conference on Systems, Signals and Devices, 2009. SSD '09. DOI: 10.1109/SSD.2009.4956757 (6pages)].

References

1. Abido M.A., Bakhashwain J.M., Optimal VAR dispatching using a multiobjective evolutionary algorithm, Electrical Power and Energy System, 2005, 27(1), p. 13-20.

2. Ben Aribia H., Hadj Abdallah H., Multi Objectives Reactive Dispatch Optimisation, Leonardo Journal of Sciences (LJS), 2007, 6(10), p. 101-114.

3. Mamandur K.R.C., Chnoweth R.D., Optimal control of reactive power flow for improvement in voltage profiles and for real power loss minimization, IEEE Trans Power Apparat Syst. 1981; PAS-100(7), p. 3185-3194.

4. Mete Vural A., Tumay Mehmet, Mathematical modelling and analysis of unfied power flow controller: A comparison of two approaches in power flow studies and effects of UPFC location, Electrical Power and Energy Systems, 2007, 29(8), p. 617-629.

5. Mishra S., Dash P.K., Hota P.K., Tripathy M., Genetically optimized neuro-fuzzy PFC for damping modal oscillations of power system, IEEE Trans. On Power Systs., 2002, 17(4), p. 1140-1147.

6. Ghadir Radman, Reshma S. Raje, Power flow model/calculation for power systems with multiple FACTS controllers, Electric Power Systems Research, 2007, 77(12), p. 1521-1531.

7. Sen K.K., SSSC-static synchronous series compensator: Theory, modelling, and application, IEEE Trans. Power Delievery, 1998, 13(1), p. 241-246.

8. Zhang X.P., Advanced modelling of multicontrol functional static synchronous series compensator (SSSC) in Newton-Raphson power flow, IEEE Trans.Power Syst., 2003, 18(4), p. 1410-1416.

9. Fuerte-Esquivel C.R., Acha E., Unified power flow controller: a critical comparaison of Newton-Raphson UPFC algorithms in power flow studies, IEE Proc. Generation Transmission Distribution, 1997, 144(5), p. 437-444.

10. Noroozian M., Angquist L., Ghandhari M., Andersson G., Use of UPFC for optimal power flow control, IEEE Trans. Power Delivery, 1997, 12(4), p. 1629-1634.

11. Abido M. A., A novel multiobjective evolutionary algorithm for environmental/economic power dispatch, Electric Power System Research, 2003, 65(1), p. 71-81.

12. Dias A. H. F., de Vasconcelos J. A., Multi-objective genetic algorithms applied to solve optimization problems, IEEE Trans. On Magnetics, 2002, 38(2), p. 1133-1136.

13. Herrera F., Lozano M., Verdegay J. L., Tackling real-coded genetic algorithm:operators and tools for behavioural analysis, Artif. Intell. Rev, 1998, 12(4), p. 265-319.