You are on page 1of 5

A New Approach for Reactive Power/Voltage Optimization Control of Regional Grid

Yuexin Dong, Honggeng Yang


School of Electrical Engineering & Information Sichuan University Chengdu, China dyxscu@126.com
AbstractThis paper puts forward a new approach for reactive power/voltage optimization control. Based on the correlation between network power loss and node voltages/injected reactive power, a dynamic adaptive adjusting method for penalty factors is presented. Particle swarm optimization algorithm (PSO) is utilized in the optimization process. Original PSO is easy to fall into a locally optimized point, to address this problem, an improved stochastic crossover PSO is proposed, which can modify the positions of the current particles using the information of the global best particle. The proposed algorithm can increase population diversity and enhance convergence speed. In addition, the stage-by-stage optimization strategy and secondary optimization strategy are introduced in order to improve the voltage quality under non-ideal operating conditions and achieve coordinated control between provincial and regional automatic voltage control (AVC). The proposed approach is applied to AVC system, the results of a practical regional grid verify the feasibility and effectiveness of the approach, and it can reduce the power loss and improve the voltage quality of power systems. Keywords- reactive power optimization; penalty factors dynamic adjusting; stochastic crossover; stage-by-stage optimization; secondary optimization

Due to the limited regulating capacity of regional grid, the optimal solution often can not meet all node restricts of voltages and injected reactive power. After the control period is determined based on load forecasting, a secondary control strategy of the on-load tap changers (OLTC) is implemented depending on the voltage quality in [5]. The infeasibility degree is introduced in [6-8], which can extend the search scope of optimization algorithms. In reference [9], in the cases of the neighborhood solutions falling into infeasible area, the voltage constraints are not satisfied, the index is switched to the voltage quality. Once neighborhood solutions return to a feasible area, the index is switched back to losses reduction during the whole day. Aiming at solving problems above, based on the correlation between network power loss and node voltages/injected reactive power, a dynamic adaptive adjusting method for penalty factors is presented. PSO is utilized in the optimization process. To address the disadvantages of PSO such as slow convergence in later evolution, easily falling into a locally optimized point, an improved stochastic crossover PSO is proposed. When power systems do not have sufficient regulating ability, this paper puts forward a stage-by-stage optimization strategy, which can spread the search space of PSO algorithm to ensure the nodes with priority satisfy constraints. Besides, to achieve coordinated control between provincial and regional AVC, a secondary optimization strategy is introduced. II. REACTIVE POWER OPTIMIZATION MODEL

I.

INTRODUCTION

Reactive power optimization of regional grid is an important way to improve voltage quality, reduce network loss and maintain the system running under normal conditions. Control variables taken for optimization problems are generator bus voltages, transformer tap positions and shunt capacitors. State variables are load node voltages and generator reactive power outputs. When the objective function is to minimize real power loss, state variable constraints being added to the function as penalty terms, the selection of penalty factors plays a key role. The penalty values are given in [1], but how to determine them is not discussed. There is a non-linear relationship between penalty factors and iterations of optimization algorithm in [2-3], the greater is the extent of constraint violation, the larger are the penalty values. In reference [4] the relation between penalty coefficients and iterations are linear, it is difficult to select the maximum and minimum, moreover, the relation between network power loss and node voltages/injected reactive power is not considered.

The goal of reactive power optimization is to minimize real power loss, while the inequality constraints on the nodal voltage and generator reactive power outputs are introduced into the object function in the form of penalty function[10,11], and the operation number of equipment are converted into economic cost[12,13], the expanded object function can be defined as

min F = Ploss + vi (U i U i lim ) 2


i =1

N PQ

+ qi (Qi Qi lim ) 2 + CTi X Ti + CCi X Ci


i =1 i =1 i =1

N PV

NT

NC

(1)

978-1-4244-4813-5/10/$25.00 2010 IEEE

Ui > Uimax Uimax Uilim = Ui Uimin Ui Uimax Uimin Ui < Uimin Qi > Qimax Qimax Qilim = Qi Qimin Qi Qimax Qimin Qi < Qimin

(2)
Figure 2. Equivalent circuit of a branch

(3)

The equivalent circuit of a branch is shown in Fig. 2. K is the ratio of the branch; Zij is the impedance between node i and node j. The expression of active power loss can be described as follows:

Where Ploss is the active power loss; NPQ, NPV, NT, NC are the number of PQ-bus, the number of PV-bus, the number of onload transformers and the number of shunt capacitor banks respectively; Ui is the voltage magnitude at bus i; Qi is the reactive power of the generation at bus i; Uimax, Uimin, Qimax, Qimin respectively represent the upper and the lower limits of corresponding variables; vi , qi are the penalty factors of voltage overstepping restricts of node i and overstepping restricts of reactive power infused by generator i separately; X T , CT are the changes and adjustable costs of transformer tap positions separately; XC , CC are the changes and adjustable costs of shunt capacitor banks respectively.
i i
i i

Ploss = Gtij (U i2 +

U2 j K
2

2U iU j K

cos ij )

(5)

Where Gtij is the conductance between node i and node j; ij is the difference of phase angle between node i and node j. The partial derivatives of active power loss to Ui and Uj respectively are shown as 2U j Ploss = Gtij (2U i cos ij ) U i K ji
2U j 2U Ploss = Gtij ( 2 i cos ij ) K K U j i j

(6)

DYNAMIC ADAPTIVE ADJUSTING FOR PENALTY FACTORS The role of penalty function is to add the cross-border information of variables to the object function, so that the object function can change the increasing or decreasing trend at the cross-border point A. As shown in Fig. 1, the greater is the extent of constraint violation, the larger is the value of object function. This paper acquires the corresponding penalty factor of each node according to the sensitivity of object function to node voltages/injected reactive power, which is different from the unified treatment before.
A. Penalty factors definition of node voltages Consider the first part and the second part in (1), and make the partial derivative of object function to Ui equal zero as (4). Calculate the absolute value of vi , i.e., the penalty factor of

III.

(7)

B. Penalty factors definition of reactive power infused by generators As shown in Fig. 2, the injected reactive power of node j will affect Uj, thus affect the network loss. Consider the first part and the third part in (1), and make the partial derivative of object function to Uj equal zero as (8). Calculate the absolute value of q j .
Q j P F = loss + 2qj (Q j Q j lim ) =0 U j U j U j
Q j U j

(8)

node i.
F Ploss = + 2vi (U i U i lim ) = 0 U i U i

is calculated as follows:

(4)

The branch current flowing Zij Uj K (9)

I ij =

Ui

Z ij

The injected power of node j Uj K

Pji + jQ 'ji =
Figure 1. Action of the penalty function

* I ij

(10)

Considering the equivalent capacitance to earth of transmission lines, the charging power should be added to the injected reactive power of node j: Y Q j = Q + 0 Uj 2 2
' ji

Where Itermax is the maximum iterations; max and min are the maximum and minimum values of inertia weight. B. Improved stochastic crossover PSO Original PSO has disadvantages such as slow convergence in later evolution, losing population diversity and easily getting trapped in local minima. This paper presents a stochastic crossover strategy. The current particles and the global best particle make a stochastic crossover. Randomly select a part of control variables from the global best particle and exchange with the corresponding variables of particle i, then form a new particle i. If the fitness which corresponds to the objective function of particle i is better than the fitness of particle i, the particle i is substituted by particle i. Since power flow is needed to calculate fitness values, in this paper each particle does only one stochastic crossover in order to meet real-time requirement. The position of particle i is generated as follows:

(11)

Where Y0 is the admittance of the equivalent capacitance to earth. If it is a transformer branch, the value of Y0 is zero. Introduce (9) and (10) to (11) and express these variables in the form of polar coordinate, then Qj is shown as Qj = [
i j

U iU j K

(Gtij sin ij + Btij cos ij )

Btij K2

U2 + j

Y0 2 Uj ] 2

(12)

Where Btij is the susceptance between node i and node j. The partial derivatives of Qj to Uj is shown as Q j U j = [
i j

xid ' xid = pgd

rand (0,1) Cr rand (0,1) < Cr

(17)

Ui (Gtij sin ij + Btij cos ij ) K 2 Btij K2 U j + Y0U j ] (13)

Where Cr is the crossover rate. V. THE STAGE-BY-STAGE OPTIMIZATION STRATEGY AND SECONDARY OPTIMIZATION STRATEGY

IV.

IMPROVED STOCHASTIC CROSSOVER PARTICLE SWARM OPTIMIZATION

A. Original particle swarm optimization In 1995, Kennedy and Eberhart first introduced the PSO method [14]. PSO is initialized with a group of random particles and each single particle is a potential solution in the search space. The position and velocity of each particle is updated at discrete intervals according to the following equation:
k k k k k k vid+1 = vid + c1 r1 ( pid xid ) + c2 r2 ( pgd xid )

A. The stage-by-stage optimization strategy The usual methods search for the optimal solution only in feasible region. However, when the constraints are strict and the search scope of control variables is limited, the space of infeasible region is very large .It is difficult to find the optimal solution in feasible region. Some infeasible solutions, especially these infeasible solutions around the border of feasible region are usually closer to the optimal solution. As shown in Fig. 3, the infeasible solutions of b1 and b2 are both closer to the best solution than the feasible solution c. Therefore, the dotted region in Fig. 3 can be formed by relaxing the constraints appropriately. And it is helpful for the optimum solution to meet restricts of the state variables. On the basis of the principle above, this paper proposes a stage-by-stage optimization strategy. In the optimization control of regional grid, a part of nodes with priority can be set to have precedence to meet voltage limits.

(14) (15)

k k k xid+1 = xid + vid+1

Where i=1,2, ,m, m is the population size; d=1,2, ,n, n represents n-dimensional search space; k is the current iteration; is inertia weight, which can be reduced gradually in successive iterations as (16); c1 and c2 are acceleration constants; r1 and r2 are two random functions with a rage [0,1]; vid is the dth dimensional velocity of particle i; xid is the dth dimensional position of particle i; pid is the dth dimensional best position of particle i; pgd is the dth dimensional position of the global best particle.

k = max

max min
I ter max

(16)
Figure 3. Feasible region and infeasible region

When encountering the non-ideal operating condition that the grid does not have sufficient regulating ability and the optimal solution can not meet the constraints of all nodes, then the constraints of connected nodes without priority can be relaxed based on the priority of control objectives and the load levels of substations. The improved PSO algorithm first does ideal stage optimization, i.e., the control solution can satisfy power flow equations, the constraints of node voltages/injected reactive power and the times of equipment oscillation. If the limits of all nodes are satisfied, PSO stops and the result is output; or else, if the limits of nodes with priority are not met, go to the next stage optimization. For each substation, the second stage optimization gives precedence to privileged nodes to satisfy constraints. The constraints of nodes without priority in the corresponding substation can be relaxed. If the limits of privileged nodes are still not met, go to the next stage optimization. Relax the constraints of nodes without priority in the connected substation with a higher voltage grade. The load levels of substations should be considered when relaxing the limits of nodes. The load factor can be described as follows:

B. The secondary optimization strategy To achieve combination control between provincial and regional AVC systems, the provincial AVC transmits limits of coordinated variables to regional AVC at regular intervals, i.e., the power factor and voltage limits of 220kV substations. This paper introduces a secondary optimization strategy to reach this optimization target above. The sketch of coordinated control is shown as Fig. 4. The regional AVC finishes reactive power/voltage optimization in this district first. When receiving control commands from provincial AVC, the regional AVC will treat the control objectives of coordinated variables as its own constraints and then finish secondary optimization. In normal situation, secondary optimization can not break the limits of variables in this region, or else secondary optimization is stopped and the results will be transmitted back to provincial AVC so as to amend the limits of coordinated variables. In emergency provincial AVC does not have regulating ability and only uses regional AVC as reactive power supports, now the regional AVC is subject to its actual control. VI. PRACTICAL CASE

i =

Si Si max

(18)

Where i is the load factor of substation i; Si, Simax are the current load and the peak load of substation i respectively. The load factor should satisfy the constraint:

The proposed approach is applied to a practical 53-bus system, including 19 on-load transformers and 23 capacitor banks. The base capacity is 100MVA. The voltage magnitude limits of 220kV, 110kV and 10kV are required between 1.02 and 1.06(p.u.). And the voltage limits of 35kV are 1 and 1.07. The algorithm programs are written by C# language and all calculations are completed in the computer with 1.6G dual CPU and 1G memory. The population scale and the maximum iterations of PSO are set to be 40 and 100 respectively. The adjustable costs of transformers and capacitor banks are 8kW per time and 5kW per time respectively. The initial power loss of the system is 0.0144847(p.u.). Stationary penalty factors in [1], linear penalty factors in [4] and the dynamic adaptive adjusting method for penalty factors proposed in this paper are adapted respectively to optimize the above system for 30 times. The outcome is shown in Table I. It can be seen from Table I that the dynamic method has better optimization results. The minimum loss of is 0.0132699 and the reduction ratio is 8.3868%. As a result, the dynamic adjusting method for penalty factors in this paper has better global optimal solutions. In order to reflect the difference between the original PSO and the improved stochastic crossover PSO, Fig. 5 shows the performances of two algorithms. We can see that the improved PSO has better outcome. It converges at about 30 iterations and has higher search efficiency.
TABLE I. Method stationary linear dynamic COMPARISON OF POWER LOSS BY DIFFERENT METHODS Best Ploss/pu 0.0134543 0.0133032 0.0132699 Worst Ploss/pu 0.0148107 0.0145205 0.0138792 Average Ploss/pu 0.0140992 0.0139479 0.0136275 Standard deviation 0.0003631 0.0003471 0.0001748

i min i i max

(19)

Where imin, imax are the light load factor and peak load factor respectively. In this paper, their values are 0.3 and 0.8 respectively. Considering the requirement for inverse voltage regulation, if i exceeds imax and the substation i is operating at low voltage, then the lower limits can not be relaxed; conversely, if i exceeds imin and the substation i is operating at high voltage, then the upper limits can not be relaxed.

Figure 4. Sketch of coordinated control of provincial and regional AVC

0.1 0.09 0.08 0.07 Fitness 0.06 0.05 0.04 0.03 0.02 0.01 0 10 20 30 40 50 60 Iterations 70 80 90 100 Original PSO Improved PSO

VII. CONCLUSION This paper puts forward a dynamic adaptive adjusting method for penalty factors, which acquires the corresponding penalty factor of each node according to the sensitivity of object function to node voltages/injected reactive power. An improved stochastic crossover PSO is proposed to overcome the disadvantages of slow convergence in later evolution, easily getting trapped in local minima and so on. And the stage-bystage optimization strategy and secondary optimization strategy are introduced to improve the performance of regional grid. The approach proposed is applied to AVC system, the results of a practical 53-bus system show a better global searching ability and faster convergence, and it can improve voltage quality effectively. REFERENCES

Figure 5. Convergence curves of two algorithms

TABLE II.

COMPARISON OF OPTIMAL RESULTS BY DIFFEREBT METHODS the number of cross-border nodes with priority
|| U i ||1 = | U i U i lim |
i =1 N PQ

[1] [2]

Method usual strategy this strategy


1.08 1.07 1.06 1.05 1.04 1.03 1.02 1.01 1

optimization time/s 3.298 5.477 [4] [3]

4 0

0.03051 0.02845

usual strategy this strategy

[5]

[6]

[7]

220kV

110kV

10kV

[8]

Figure 6. Comparison of privileged node voltages

[9]

For verifying the effectiveness of the stage-by-stage optimization and secondary optimization strategy, compare this strategy with usual strategy. Before optimization, 6 nodes with priority cross limits. The optimal results are shown in Table II and Fig. 6. It can be seen that after optimization, for the usual strategy 4 nodes with priority still cross limits, but the proposed strategy in this paper can make the voltages of all privileged nodes meet constraints. Comparing 1-norm of the deviation of voltage overstepping restricts at PQ nodes, Table II shows that, the 1-norm values of the two strategies are similar, however, the voltages of all nodes with priority are within limits using the approach in this paper. According to the grid performance, this approach maybe gets into stage-by-stage optimization and costs more calculating time, but it can improve voltage quality effectively.

[10]

[11]

[12]

[13]

[14]

Kenji Iba, Reactive power optimization by genetic algorithm, IEEE Transactions on Power Systems, vol.9, no.2, pp.685-692, May 1994. Bo Zhao, Chuangxin Guo, Yijia Cao, Optimal power flow using particle swarm optimization and non-stationary multi-stage assignment penalty function, Transactions of China electrotechnical Society, vol.19, no.5, pp.47-54, 2004. Junxia Yu, Bo Zhao, Improved particle swam optimization algorithm for optimal power flow problems, Proceedings of the CSU-EPSA, vol.17, no.4, pp.83-88, 2005. Gonggui Chen, Zhihuan Li, Yongfa Sun, Jianwei Zhong, LRS-PSO algorithm for optimal reactive power flow, Proceedings of the CSUEPSA, vol.20, no.4, pp.92-97, 2008. Xiuli Wang, Xiawen Li, Zechun Hu, Time-interval based comprehensive control strategy for daily voltage/VAR optimization in distribution systems, Automation of Electric Power System, vol.30, no.7, pp.5-9, 2006. Shengjing Mu, Hongye Su, Weijie Mao, Zhenyi Chen, Jian Chu, A new genetic algorithm to handle the constrained optimization problem, Proceedings of the 41st IEEE Conference on Decision and Control, Las Vegas, USA, 2002, pp.739-740. Yuexuan Wang, Lianchen Liu, Shengjing Mu, Cheng Wu, Constrained multi-objective optimization evolutionary algorithm, Journal of TsingHua University (Science and Technology) , vol.45, no.1, pp.103106, 2005. Jian Wang, Shan Zhang, Voltage regulation research of an improved genetic algorithm considering infeasibility degree, Journal of ShanDong University (Engineering Science), vol.38, no.6, pp.21-24, 2008. Wei She, Jian Liu, Yan Zhou, Power systems reactive power regulation considering schedule optimization, Automation of Electric Power Systems, vol.33, no.4, pp.31-35, 2009. Zhen Ren, Hongmei Zhong, Yongjun Zhang, Zhuoyao Tang, Improved genetic algorithm for reactive power optimization of electric network, Electric Power Automation Equipment, vol.22. no.8, pp.16-19, 2002. Dan Li, Liqun Gao, Shun Lu, Jia Ma, Yang Li, Adaptive particle swarm optimization algorithm for power system reactive power optimization, Proceedings of the 2007 American Control Conference, New York City, 2007, pp.4733-473. Yongjun Zhang, Zhen Ren, Distributed cooperative optimization for reactive power/voltage dynamic control, Proceedings of the CSEE, vol.24, no.4, pp.34-38, 2004. Jia He, Yaowu Wu, Suhua Lou, Xinyin Xiong, Dynamic reactive power optimization based on particle swarm optimization algorithm, Power System Technology, vol.31, no.2, pp.47-51, 2007. Kennedy J, Eberhart R, Particle swarm optimization, IEEE International Conference on Neural Networks, Perth, Australia, 1995, pp.1942-1948.

You might also like