IEEE PES ISGT ASIA 2012 1569526795

Reliable Power Flow Calculation with Improved Convergence Characteristics for Distribution Systems
Takamichi Ochi, Yuto Nonaka, Daiki Yamashita, Student Member, IEEE, Kaoru Koyanagi, Member, IEEE, and Ryuichi Yokoyama, Fellow, IEEE
Backward/Forward Sweep method[6] is usually employed because it is simple in algorithmic structure, and is suitable for radial systems with many P-Q specified nodes. N-R method is familiar to power system engineers for power flow calculations of transmission systems. The transmission systems consist of many loops and P-V specified buses as well as P-Q specified buses. Thus, N-R method has no restrictions in network topology and bus specifications. In this paper, at first, the convergence characteristics of conventional N-R method are evaluated using a radial distribution system with lines of high R/X ratio. Second, the authors proposed new calculation method based on N-R method, and investigated the effectiveness using several distribution systems including an ill-conditioned system with some voltages close to voltage stability limit. II. N-R METHOD FOR HIGH R/X-RATIO SYSTEMS In this chapter, comparisons of the convergence characteristics by conventional N-R method between high R/X ratio system and low R/X-ratio system were investigated. This tests were conducted using the 126-bus radial distribution system[7] which is shown in Fig.1. This distribution system has all P-Q specified nodes, no distributed generators, no transformers, and no phase modifying equipment. Conventional N-R method was tested with the distribution system to evaluate the convergence characteristics in case of high R/X ratios. Here, the N-R method is based on polar coordinates. The convergence threshold in power mismatch applied in the iteration process was set to 0.0001 [p.u.].
98 96 94 86 87 31 32 33 4344 45 68 69 77 89 97 95 90 91 88 92 83 15 93 16 17 111 112 124 18 19

Abstract-- For construction of smart grid with optimal operations, analytical studies for distribution systems are important, and advanced method which is applicable to any power systems is required. Today, Newton-Raphson method is utilized for a lot of power flow analysis. However, since R/X ratios of lines are relatively high in distribution systems, it is said that the convergence characteristics of Newton-Raphson method are poor. In addition, when some bus voltages are close to the voltage stability limit in power systems, the convergence characteristics are worse. In this paper, first the convergence characteristics of conventional Newton-Raphson method is evaluated using radial distribution systems with lines of high R/X ratios. Next, the new calculation method for ill-conditioned system is proposed which is based on Newton-like method with a little devisal. In the method, admittances of load at each P-Q specified buses are embedded in Y-matrix. The effectiveness and robustness of the proposed method for distribution systems is demonstrated. Index Terms-- Distribution systems, Newton-Raphson method, Power flow calculation, Voltage instability

I. INTRODUCTION

OR reduction of CO2 emission against global warming, large penetration of renewable energies such as solar energy and wind energy in distribution power systems is in progress. If the distributed generators are interconnected in distribution systems in large quantity, certain problems will arise. In constructing smart grids with their optimal operation, power system analytical studies for distribution systems are necessary. Newton-Raphson method(here, we call simply N-R method) which is one of major power flow calculation methods, is utilized for a lot of power flow analysis. In distribution networks, however, convergence characteristics by N-R method, especially in fast decoupled N-R method, is said to be worse than those in transmission networks due to their high R/X-ratio lines[1]. Furthermore, operating point near voltage stability limit makes the convergence characteristics worse. Such system condition is said illconditioned[2],[3]. For distribution system,

126

Takamichi. Ochi, Yuto. Nonaka, Daiki. Yamashita, Kaoru. Koyanagi and Ryuichi. Yokoyama are with Waseda University, Tokyo, Japan (e-mail: [email protected])

84 80 85 82 81 6 12 1 4 7 8 10 5 14 9 11 2 3 13 70 71 47 48 46 125 72 21 22 34 35 36 76 78 79 99 49 50 73 37 23 24 38 39 51 52 101 74 75 53 54 26 40 25 41 57 58 56 55 106 27 29 28 42 59 60 61 63 64 62 30 108 65 66 67

115 116 100 109 110 114 113 102 103 120 119 117 118 123 121 104 105 107 122

Fig.1 126-bus distribution network model[7]

was good in both cases of the original line data and the changed line data R ' + jX ' .

Fig.2. R/X ratios of distribution lines changed by phase angles

Max Mismatches [p.u.]

20 15 10 5 0

Low R/X ratios High R/X ratios

3 4 5 Iteration Counts Fig.3. Convergences process in case of high R/X ratios with fixed |Z|

B. Discussions In these tests, it was found that conventional full-Jacobian N-R method can be used even for load flow calculations for distribution system with lines of high R/X ratios and radial configurations with all P-Q specification. The authors believe that the high R/X ratio problem may be restricted to Fast Decoupled N-R Method because the links between P and , and between V and Q are weakened in case of high R/X-ratio distribution system. The other case that the conventional full-Jacobian N-R method is concerned will be some deterioration in Jacobian elements. For some reasons, magnitude of diagonal elements in Y-bus matrix may be reduced smaller than that of the offdiagonal elements matrix. This condition will arise, for example, when series capacitors are used in the system. In addition, Jacobian matrix for the conventional N-R method will cause problem of singularity near voltage stability limit. In the next chapter, the authors will present an idea to enhance the robustness of N-R method against the above problems.
III. THE PROPOSED METHOD In the previous simulations it was shown that the convergence characteristics of the conventional full-Jacobian N-R method has no problems for application to distribution systems with lines of high R/X ratios as well as to transmission systems. However, the conventional N-R method is indicated that if the operating conditions are close to voltage stability limit in power systems, the convergence characteristics are worse and sometimes make it difficult to get voltage solutions. Therefore, we investigated the new power flow calculation method based on N-R method.

A. Convergence Tests for High R/X-Ratio Distribution Lines In order to assess convergence characteristics on the 126bus radial distribution system, line impedances adopted were modified under the conditions as follows. (1) Obtain Z and from R and X of each line impedance R + jX = Z = Z ( cos + j sin ) (1a)

(2) Obtain ' = to simulate high R/X ratio lines for = 40(deg.) . (3) Obtain R ' and X ' using Z and ' for each line
impedance
R ' + jX ' = Z ' = Z cos ' + j sin '

(1b)

R/X ratios of the original line impedance[7] and R/X ratios of the modified line impedance are shown in Fig.2. The original R/X ratios[7] of each line impedance were concentrated around ratio of 1. The modified R/X ratios in case of = 40(deg.) are higher than them of original impedance were concentrated around 4 and 7.5. The both cases shown in Fig. 2 needed 4 iterations to be converged. The test results show that the convergence characteristics of conventional N-R method are satisfactory even if N-R Method is employed for distribution systems with high R/X ratios. Fig. 3 shows the largest mismatch values in iterative computation process in cases of low R/X ratios and high R/X ratios. In addition, for the test system which has P-V specified buses as well as P-Q specified buses, the convergence process

A. Basic Idea of the Proposed Method In the new method, active power and reactive power of all load buses are embedded in the Y-bus matrix as the constant diagonal admittance, and the specified P-Q values of the load buses are newly defined as the differences between the original specified P-Q values and the constant-admittance load components depending on the magnitude of bus voltage. The effectiveness of the proposed method may be recognized from two points of view, numerical solution and voltage stability. (1) Numerical solution: The proposed method makes the diagonal elements of nodal admittance matrix larger than the conventional N-R method by addition of load admittances. This can keep Jacobian matrix away from singular point in solution process. (2) Voltage stability: The proposed method changes voltage dependency of the P-Q specified load, from original constant-voltage characteristic to constantimpedance characteristic. It is said that the power system with constant-impedance loads can avoid voltage instability problem.

B. Mathematical Procedures The proposed method is a revised version of conventional N-R Method in treatment of P-Q specified buses. It can be described as follows. The injected power equation at each P-Q bus i can be written as

I * Pi spec + jQi spec = V i i = Vi

Y
j =1

ij

* V j

(2)

The injected power at bus i, Pi spec + jQi spec , are converted to load admittance in Y-bus matrix, * , Pi spec + jQispec Pi spec jQispec Li = (3) y = 2 2 V V

(7) at all P-Q nodes( Pi spec reviced + jQi spec reviced ); Step 3: Calculate P and Q , mismatch powers between specified values, and calculated values at each bus. If all P and Q are smaller than the pre-determined threshold value, , the solution process is terminated. If not, go to Step 4; Step 4: Update Jacobian matrix, and obtain V and by using the updated Jacobian matrix; Step 5: Add V and to previous V and , then obtain new V and ; Step 6: Back to Step 2; The proposed method has the feature that load impedance is connected at each P-Q specified node and specified injected powers are updated each iteration. Fig. 4 shows the image of connected load impedance at the P-Q specified node.

Equation (2) is modified using the admittance

Li as follows, y

Li ) (Yii + y

+V i

Y
j i

ij

* V j
*

n Pi spec jQispec *V * = Vi Yii + V Y i ij j 2 Vi j i 0 Li Here, we define constant load admittance, y ,

(4)

0 Li y = Pi spec + jQispec

(5)

In the proposed method, depending on bus voltage.

0 Li y

Li that is is used instead of y

C. Demonstration using Simple Power System Features of the proposed method are demonstrated here using a simple power system with 2 buses. Fig.5 shows the power system with single generator bus and single load bus. The power factor of the load is 0.9(lagging). As shown in Fig.5, specified powers at load bus were selected as 1.00, 1.045, 1.050(p.u.). The case P=1.0(p.u.) is normal state. However, the case P=1.045 is critical condition and the case P=1.050(p.u.) is no solution condition with voltage instability. Table II shows the power flow calculation
1.00 j0.3 V Load P+jQ P.f.=0.9(lag)

(P

spec

2 + jQispec 1 V i = Vi

(Y

ii

0 Li +y

+V i

Y
j i

ij

* (6) V j

Then, the revised specified-injected powers can be modified using the original specified-injected powers as follows. 2 Pi spec revised + jQispec revised = Pi spec + jQispec 1 V i (7) The proposed method needs to update specified injected powers at each iteration cycle by using previous values of voltage.. The outline of computation process in the proposed method is described as following six steps. Step 1: Make nodal admittance matrix (Ybus), and add load 0 Li admittance y to the diagonal elements of nodal admittance matrix; Step 2: Calculate specified values of P and Q by equation

Fig.5 Simple power system

Bus Voltage

Specified Power
1.00 1.045 1.050

Nose Curve

Power
Fig.6 Nose curve and specified load power TABLE I POWER FLOW CALCULATION RESULTS

Method
Conventional N-R Method Our Proposed Method

P=1.00
YES YES

P=1.045
NO (Hunting) YES

P=1.050
NO (Hunting) NO (Stable Solution)

Fig.4 Treatment of P-Q Specified bus in the proposed method

4
Convergence Process by Conventional N-R Method for Specified P=1.05 Bus Voltage Phase Angle

0.2

Correction per Iteration

0.15 0.1 0.05 0 -0.05 0 -0.1 -0.15 -0.2 Iteration 1 2 3 4 5 6

N-R method has singularity after iteration 5. On the other hand, the proposed method has no problem. This indicates stability in numerical solution.
9 10

IV. POWER FLOW CALCULATIONS The proposed method was tested on the several power systems to examine the convergence characteristics in the solution process. The convergence characteristics were compared with that of conventional N-R method. The test systems used were 126-bus distribution system [7], the 13-bus system [8], and the 11-bus system [2]. The 11-bus system had been regarded as ill-conditioned, since some bus voltages are quite close to voltage stability limit. On this account, the 11bus ill-conditioned system makes convergence difficult for conventional N-R method. In the case of this power system, if calculation is started from flat voltages as initial values, the
100 10
Max Mismatches [p.u.]

(a) Conventional N-R method

Correction

0.05 0 -0.05 0 -0.1 -0.15 -0.2 Iteration 1 2 3 4 5 6 7 8 9 10

(b) The proposed method Fig.7 Convergence process for the case P=1.050(p.u.)
Values of Determinants of Jacobian Matrix
Determinants of Jacobian Matrix
16 14 12 10 8 6 4 2 0 -2 0 1 2 3 4 5 Iteration 6 7 8 9 10 Conventional N-R Method The Proposed Method

1 0.1 0.01 0.001 0.0001

5
N-R method Proposed method Tolerance

Fig.8 Values of determinants of Jacobian matrix

0.00001

results for 3 cases by conventional N-R method and the proposed method. As shown in Table II, the proposed method can obtain the solutions for P=1.045. In the case of P=1.050, that is no solution case, convergence process curves for both methods are shown in Fig,7(a) and (b). Compared to the conventional N-R method, the proposed method shows stable convergence process, in other word, approaches asymptotically to the target. This indicates no solution case and the near point to zero level will give the approximated values of solution. Fig.8 shows the values of determinants of Jacobian matrix in convergence process for both methods. The conventional
TABLE II INITIAL VALUES FOR 11-BUS ILL-CONDITIONED SYSTEM Node Number V [p.u.] [rad] 1 2 3 4 5 6 7 8 9 10 11 1.024 1.0552 1.0444 1.0306 1.0329 1.0501 0.831 0.9131 1.1264 0.8266 1.0186 0 -0.02121 -0.08042 -0.04939 -0.0843 -0.05089 -0.21059 -0.2593 -0.2777 -0.3671 -0.41881

Fig.9 Convergence characteristics of 126-bus system by the conventional N-R method and by the proposed method
10 1
Max Mismatches [p.u.]

1 0.1 0.01 0.001 0.0001 0.00001

5
N-R method Proposed method Tolerance

0.000001

Iteration Counts

Max Mismatches [p.u.]

Fig.10 Convergence characteristics of 13-bus system by the conventional N-R method and by the proposed method 1 0 10 20 30 N-R Method 0.1 Proposed Method

0.01

0.001

Iteration Counts Fig. 11 Convergence characteristics of 11-bus ill-conditioned system by the conventional N-R method and by the proposed method

0.0001

computation could not be converged [9]. Thus, the authors selected initial values as the values referenced in [2]. Initial values used for the 11-bus ill-conditioned system were listed in Table II. Fig. 9, Fig. 10 and Fig. 11 show the largest mismatch values in each iterative process in case of the 126bus distribution system, 13-bus system and 11-bus illcondition system, respectively. As shown in Fig. 9 and Fig. 10, in cases of 126-bus and 13bus systems, the proposed method and the conventional N-R method could get convergences with no problems. However, computation process by the proposed method required 1 more iteration than by conventional N-R method, and the rate of decrease of mismatches was relaxed after iteration counts, three. This relaxation in convergence process is due to the nonfixed(variable) specified injection powers that are depending on the magnitude of bus voltages. However, the proposed method can be improved by taking account of voltage or power prediction. The improvements will be described in the next chapter. For the case of the 11-bus ill-conditioned system, Fig. 11 shows that the computation of the conventional N-R method was diverged, or hunting in solution process. However, the maximum mismatches after certain iteration of the proposed method were kept about 0.0005[p.u.]. The result represents that the proposed method could approach voltage solutions close to voltage stability limit compared with the conventional N-R method. The fact that maximum mismatches never reduce to zero means that there is no power flow solutions for the specified P-Q conditions. Thus, it is recognized that the solution process in the proposed method is stable compared to conventional N-R method, and we can determine whether voltage solutions exist or not by the results. In summary, from these results mentioned above, it was confirmed by the simulations that the proposed method could get convergences in case of 126-bus distribution system and 13-bus system. In addition, in case of 11-bus ill-condition system, this method can present approximate solutions though the exact solutions do not exist. Therefore, the proposed method was the effective and robust method compared with conventional N-R method for the systems which have lines of high R/Z ratios, or voltage solutions close to the voltage stability limits. V. IMPROVEMENTS IN THE PROPOSED METHOD Fig. 9 and Fig. 10 show that the rate of decrease of mismatches by the proposed method was relaxed compared with the conventional N-R method. The authors recognized that it is due to the change of specified P-Q values that is depended on the bus voltage. In conventional N-R method, Pi spec + jQispec is constant during iterations. However, in the

Fig.12 Convergence characteristics of 126-bus system considering acceleration factor

Active Power [p.u.]

-0.0024 -0.0026 -0.0028 -0.003 -0.0032

Iteration Counts

Fig.13 Convergence characteristics of 126-bus system without acceleration factor (=0)

5
Injection P Calculated P Specified injection P

-0.0024 -0.0026 -0.0028 -0.003 -0.0032

Iteration Counts

Fig.14 Convergence characteristics of 126-bus system with the acceleration factor =0.09

injection powers and calculated powers. One of the improvements in convergence process will be extrapolation of specified injection powers described as,
Pi spec rev. ( k ) = Pi spec rev. ( k )
'

+ Pi spec rev. ( k ) Pi spec rev. ( k 1)

(8)

proposed method, specified P-Q values are changed according to equation (7). In other words, the targets are moving. In order to overcome this deterioration, the authors considered ingenuity in mismatch computations between specified

Here, is acceleration factor ( > 0 ). The same improvement method is applied also to Q. This improvement by (8) was utilized in case of 126-bus radial distribution system. The results are shown in Fig.12. When the factor was 0.09, the number of iteration needed was reduced from 5 to 4. This reduction means effectiveness of the improvement of convergence characteristics. The other ideas may be possible for improvements in the proposed method. In summary, the proposed method is said as a slightly modified version of the conventional N-R method. The convergence process of the proposed method is stable even

when the operating condition is close to, and a little over, the voltage limit compared to the conventional N-R method. However, number of iterations to reach convergence increases a little by the proposed method and this fact may need more study to get more improvements in future. VI. CONCLUSIONS In this paper, the convergence characteristics of fullJacobian N-R method on distribution system with lines of high R/X ratios were investigated. Also, the authors proposed a new calculation method based on N-R method which is able to calculate voltages close to voltage stability limit. In addition, the effectiveness and robustness of the proposed method was tested using several examples of distribution systems. The results will be summarized as follows; (1) The convergence characteristics in conventional fullJacobian N-R method are not affected with high R/X ratio of lines. This indicates that even for distribution system N-R method (Newton-like method) is applicable with no problems as well as transmission systems. (2) The authors proposed an improved version of N-R method. In the new method, active power and reactive power of all load buses are embedded in the Y-bus matrix as the constant diagonal elements, and the new specified P-Q values of the load buses are newly defined. It can get voltage solutions even in the case that the operating condition is close to voltage stability limit, where conventional N-R method has difficulty in solution because of singularity of Jacobian matrix. (3) The convergence characteristics of the proposed method were compared with conventional N-R method. The test systems used were 126-bus distribution system, 13-bus system, and 11-bus system that are reported as sample power systems in the literature. The proposed method was able to get solutions for 126-bus distribution system and 13-bus system with no problems, and also to indicate no exact solution for 11-bus system that is a little over voltage stability limit. The proposed method is thus effective and robust compared with the conventional N-R method for the critical system where voltage solutions are close to, or a little outside, the voltage stability limits. (4) The authors presented an idea to get improvement in convergence process for the proposed method by extrapolation of the updated values of specified injection powers. VII. REFERENCES
[1] [2] M.S. Srinivas, "Distribution load flows: A brief review", IEEE Power Engineering Society Winter Meeting, vol. 2, pp. 942-945, January. 2000. S.C. Tripathy, G. Durga Prasad, O.P. Malik, and G.S. Hope, "Load-flow solutions for ill-conditioned power systems by a Newton-like method", IEEE Trans. Power Apparatus and Systems, vol. PAS-101, No. 10, pp. 3648-3657, October, 1982 Y. Wang, da Silva, L.C.P, W. Xu, and Y. Zhang, "Analysis of illconditioned power-flow problems using voltage stability methodology",

IEEE Proceedings Generation, Transmission and Distribution, vol. 148, No. 5, pp. 384-390, September, 2001. [4] B. Stott and O. Alsac, "Fast decoupled load flow", IEEE Trans. Power Apparatus and Systems, vol. PAS-93, No. 3, pp. 859-869, May, 1974. [5] S. Iwamoto and Y. Tamura, "A fast load flow method retaining nonlinearity", IEEE Trans. Power Apparatus and Systems, pp.15861599, September, 1978. [6] D. Shirmohammadi, H.W. Hong and A. Semlyen, "A compensationbased power flow method for weakly meshed distribution and transmission networks", IEEE Trans. Power Systems, vol. 3, No. 2, pp. 753-762, May,1988. [7] H. Yang, F. Wen, L. Wang and S.N. Singh, "Newton-downhill algorithm for distribution power flow analysis", IEEE 2nd International Power and Energy Conference, pp. 1628-1632, December, 2008. [8] K. Zollenkopf, "Load-flow calculation using loss-minimisation techniques", Proceedings of the Institution of Electrical Engineers, vol. 115, No. 1, pp. 121-127, January, 1968. [9] S. Iwamoto, and Y. Tamura, "A load flow calculation method for ill conditioned power systems", IEEE Trans. Power Apparatus and Systems, vol. PAS-100, No. 4, pp. 1736-1743, April, 1981. [10] Y.H. Moon, B.K Choi, B.H. Cho, S.H. Kim, B.N. Ha, and J.H. Lee, "Fast and reliable distribution system load flow algorithm based on the YBUS formulation", IEEE Power Engineering Society Summer Meeting, vol. 1, pp. 238-242, July, 1999.

VIII. BIOGRAPHIES
Takamichi Ochi was born in Tokyo, Japan, in November, 1988. He received B.S. degree in Department of Electrical Engineering and Bioscience from Waseda University, Tokyo, Japan, in 2011. He is now a graduate student at the Graduate School of Environment and Energy Engineering in Waseda University in Japan.

Yuto Nonaka was born in Saitama, Japan, in September, 1987. He received B.S. degree in Department of Electrical Engineering and Bioscience from Waseda University, Tokyo, Japan, in 2010. He is now a graduate student at the Graduate School of Environment and Energy Engineering in Waseda University in Japan. Daiki Yamashita (StM07) is a Ph.D student at Waseda University, Tokyo Japan. He has received his B.S degree from the Department of Electrical Engineering and Bioscience, Waseda University in 2008, and an M.S. degree from the Graduate School of Environment and Energy Engineering, Waseda University, in 2010. His current interests are optimal scheduling problem concerning CO2 emission trading or emission limitations and model of Electric Vehicle Markets and Charging Infrastructure. He is a student member of IEE of Japan and IEEE. Kaoru Koyanagi (M96) received his B.S degree in applied physics from Tokyo University of Education in 1971. He obtained Ph.D. degree in electrical engineering from Tokyo Metropolitan University in 2000. From 1971 to 1996, he was with Toshiba Corp. working on power system analytical engineering. He has been involved in various aspects of power system stability analysis and design of control systems. He is currently with Environment and Energy Research Center, Waseda University, Japan. He is chartered engineer in UK, and is a member of IEE of Japan, IET and IEEE. Ryuichi Yokoyama received the degrees of B.S., M.S., and Ph.D. in electrical engineering from Waseda University, Tokyo, Japan, in 1968, 1970 and 1974 respectively. After working in Mitsubishi Research Institute, from 1978 through 2007, he was a professor in the Faculty of Technology of Tokyo Metropolitan University. Since 2007, he has been a professor of the Graduate School of Environment and Energy Engineering in Waseda University. His fields of interests include planning, operation, control and optimization of large-scale environment and energy systems, and economic analysis and risk management of deregulated power markets. He is a fellow of IEEE and a senior member of IEE of Japan, and a member of CIGRE.

[3]