2nd IEEE ENERGYCON Conference & Exhibition, 2012 / Future Energy Grids and Systems Symp

ACTIVE-REACTIVE OPTIMAL POWER FLOW FOR LOW-VOLTAGE NETWORKS WITH PHOTOVOLTAIC DISTRIBUTED GENERATION Aouss Gabash, Student Member, IEEE, and Pu Li Department of Simulation and Optimal Processes Institute of Automation and Systems Engineering, Ilmenau University of Technology 98684 Ilmenau, Germany Phone/Fax number: +0049 3677 69-2813/1434 E-Mail: aouss.gabash@tu-ilmenau.de, pu.li@tu-ilmenau.de
ABSTRACT Photovoltaic systems (PVSs) are more and more installed in low voltage distribution networks (LV-DNs) for absorbing solar energy. Features of such networks need to be analyzed for a reliable and optimal design as well as operation. In this study, we introduce a mathematical model derived from a combined active-reactive optimal power flow (A-R-OPF) for LV-DNs by taking into account the reactive power capability of photovoltaic distributed generation (DG). Using multiple performance criteria, it is possible to analyze the impact of controlling reactive power sources LV-DNs. A real 29-bus LV-DN is employed to demonstrate the effectiveness of the proposed method, based on which interesting results have been achieved. For instance, up to 25% of annual energy costs can be saved and there is no need in this case study to use battery storage systems (BSSs) in LV-DNs for accommodating spilled PV energy. Index TermsActive-reactive optimal power flow, low voltage, photovoltaic distributed generation, reactive power capability. 1. INTRODUCTION More and more wind and solar renewable energy sources are penetrated to distribution networks (DNs) due to environmental and economic concerns. However, they are still more expensive in comparison to conventional energy sources. Therefore, governmental regulations have been introduced to support their integration into the energy market [1]. If the portion of distributed generation (DG) units is relatively high in DNs, a large amount of DG power and energy would be curtailed and spilled out of the connected DNs. This is because of system constraints such as voltage and conductor ampacity limitations.
This work was supported by the Ministry of Higher Education of the Syrian Arab Republic.

Many studies have been made to accommodate such DG energy by means of energy storage systems (ESSs) such as battery storage systems (BSSs) [2-3]. BSSs are capable to generate both active and reactive power through power conditioning systems (PCSs) [4]. Based on this property, a method was recently proposed in [5] to simultaneously optimize the active and reactive strategies in DNs by a combined active-reactive optimal power flow (A-R-OPF). The DG reactive power capability, mainly from wind and photovoltaic (PV) DG units, were explored by studies in [68]. It is commonly recognized that PV systems (PVSs) are mostly preferred to be installed, if possible, at sites near end users, i.e., at the low voltage level. However, at this level there is a limit to install PVSs [9], mainly due to voltage rise and conductor ampacity. In [10], this problem was handled by means of an active power curtailment (APC). In contrast, it is demonstrated in [11] that DG units could provide more energy and system ancillary services (in particular, voltage regulation and partial compensation or elimination of some power quality disturbances, such as waveform distortions and voltage unbalances). Moreover, DG units in [11] were coordinated to provide the PQ compensation service and voltage regulation using a centralized control system (CCS). In [12] a decentralized nonlinear auto-adaptive controller was proposed for reducing system losses by optimal management of reactive power supplied by the inverters of PV units. Based on the above survey, features of DNs with photovoltaic distributed generation need to be analyzed for a reliable and optimal design as well as operation. In particular, the impact of energy market and governmental regulations should be taken into account. The aim of this study is to investigate a mathematical model based on the A-R-OPF by using multiple performance criteria and utilizing DG reactive power capabilities. Results of a real 29-bus LV-DN show significant potentials when reactive power dispatch is considered in operations of such distribution networks.

978-1-4673-1454-1/12/$31.00 2012 IEEE

381

Fig. 1. (a) Conceptual relation between active and reactive power of an ideal PV-PCS. (b) Active and reactive power capability of a PV-PCS.

2. MODELING PROCEDURE 2.1. Modeling of PV systems In general, a PVS consists of two main parts: a PV panel to collect energy from sunlight to produce a direct current (DC) and a power conditioning system (PCS) to convert this current to a suitable alternating current (AC). The PCS can produce both active and reactive power which can be controlled independently [6]. Figure 1 depicts the conceptual relation between the active and reactive power of a PV-PCS. It was shown in [5] that a combined A-R-OPF can lead to better operational performances by optimizing active and reactive power strategies simultaneously. In this study, we employ this method to analyze features of LVDNs with PV distributed generation. The basic idea of the A-R-OPF is to utilize the potential available in a PCS, as shown in Fig. 1(a). This can be explored by introducing a curtailment factor curt(l,h) at each PVS connected at a specific node l during hour h. In this way a feasible solution can be ensured, i.e., to spill a part of PV energy out when system constraints will be violated. In other words, curt(l,h)=1, if no PV power will be curtailed and curt(l,h)<1 otherwise. The relations between active and reactive power can be described as follows
2 S PCS (l , h) = ( Ppv (l , h) curt (l , h)) 2 + Qdisp (l , h)

2.2. Modeling of network demands The power demand from the connected LV-DN is considered to follow the IEEE-RTS seasons days, as given in [2], it is given later in the case study. Fig. 2 describes two main operation conditions of the LV-DN with a high generation of PVSs. Here, the first demand profile, denoted by Pd(1), occurs in cloudy days or no generation of PVSs. In contrast, the second demand profile Pd(2) occurs in heavily sunny days. We assume that it is possible to transport energy produced by PVSs to the connecting network, as seen from the negative part. Typically, a radial LV-DN has a common source-and-drain bus used to import energy from and export to an upper medium voltage distribution network (MV-DN). This bus is defined here as the slack bus. 2.3. Modeling of energy prices Generally, governmental regulations stand behind the remuneration of renewable energy sources [1]. In this work, two types of prices are used for the analysis, as depicted in Fig. 2. The first is an average remuneration tariff (Cpr.art), usually called as a fixed feed-in tariff. This price is used mainly to remunerate PVSs during an investment period.

(1)

2 Qava (l , h) = SPCS.max (l ) ( Ppv (l , h) curt (l , h)) 2 (2)

2 2 ( Ppv (l , h) curt (l , h)) 2 + Qdisp (l , h) SPCS.max (l )

(3) (4)

SPCS.max (l ) Qdisp (l , h),

where SPCS (l,h) is the apparent power of a PV-PCS at bus l during hour h, Ppv(l,h) is a defined PV power profile generated from a PVS at bus l during hour h, Qdisp(l,h) is the reactive power of a PV-PCS at bus l during hour h, SPCS.max(l) is the maximum apparent power capability of a PV-PCS at bus l, Qava (l,h) is the available reactive power of a PV-PCS at bus l during hour h, respectively.

Fig. 2. Daily photovoltaic/demand power profiles and energy prices.

382

Second, on-peak/off-peak price model (Cpr) which is used for charging the demand and losses in a specific utility. It is noted that Cpr.art > Cpr since renewable energies are being supported by governmental regulations [1]. Moreover, this governmental support usually decreases from year to year with a degression rate which can vary between 1.5% and 21% per year for PV installations attached to or on top of buildings in Germany from 1 January 2012 [1]. 3. A-R-OPF UTILIZING DG REACTIVE POWER CAPABILITY In this section, a multi-period OPF is formulated as a dynamic optimization problem for LV-DN systems. The optimization framework can be described by Fig. 3 where a CCS is considered. Here we consider a LV-DN with a high penetration of solar energy from PVSs being operated by a distribution system operator (DSO) who is responsible for operating the system with a high quality. The DSO tends to maximize the benefits from the system and meanwhile minimize the total cost. The LV-DN is connected to a MVDN through a substation where the secondary bus of the transformer can be considered as a slack bus. As shown in Fig. 3, the control variables are the curtailment factor for active power and the reactive power dispatch of PVSs. Both of these controls are required as ancillary services. The mathematical model is defined in the following. 3.1. Objective function The aim of optimization is defined for the A-R-OPF [5] of the LV-DN to maximize a multi-criteria objective function. It includes the total yield from all PVSs connected to the LV-DN, meanwhile the total cost will be minimized involving the cost of power losses and the cost of demand, i.e.,
Qdisp , curt

Fig. 3. Input-output scheme.

Tfinal

F3 = Cpr (h) Pd (i, h).

h =1 i =1

(8)

Here, G is the real component of the complex admittance matrix elements, Ve, Vf are the real and imaginary components of the complex voltage, respectively. Pd(i,h) is the demand at bus i during hour h. N is the total number of buses, i, j are indexes for buses, Tfinal is the final time point in the optimization horizon (1 year, i.e., Tfinal =8760 h). Note that the optimization problem is solved daily and repeated for one year. This is typically used for a short-term analysis in power systems. Briefly, the market strategy considered in this work can be summarized as follows. The PV based DG units and reverse active energy to the MV-DN are paid for the same price model, i.e., the average remuneration tariff. The active energy losses and demands are charged by the same price model, i.e., the two-tariff price model. 3.2. Equality constraints The equality constraints are the active power balance at each bus in the network as follows

max F1 F2 F3

(5)

Ve (i, h) (G (i, j )Ve ( j , h) B (i, j )V f ( j , h))

j =1 ji

where the three terms include the total yield of the PVSs in the LV-DN

F1 = Cpr.art (h) Ppv (i, h) curt (i, h)

h =1 i =1 il

Tfinal

+ V f (i, h) (G (i, j )V f ( j , h)
j =1 ji

(9)

(6)

+ B (i, j )Ve ( j , h)) + P(i, h) = 0 , i N

where B is the imaginary component of the complex admittance matrix elements and P is the active power injection which is given by (7)

the loss cost

F2 =

N N 1 Cpr (h) G (i, j )(Ve2 (i, h) 2 h =1 i =1 j =1

Tfinal

+ V f2 (i, h) + Ve2 ( j , h) + V f2 ( j , h) 2(Ve (i, h)Ve ( j , h) + V f (i, h)V f ( j , h)))

and the demand cost

P (i, h) = Pd (i, h) Ppv (i, h) curt (i, h) PS (h), i N.

(10)

Here Ps is the active power at the slack bus. The reactive power balance at each bus is given as follows

383

V f (i, h) (G (i, j )Ve ( j, h) B (i, j )V f ( j , h))

j =1 ji

Ve (i, h) (G (i, j )V f ( j, h)
j =1 ji

(11)

+ B (i, j )Ve ( j , h)) + Q(i, h) = 0,

iN

where Q is the reactive power injection which is given by

Q (i, h) = Qd (i, h) Qdisp (i, h) QS ( h), i N . (12)

where Qs is the reactive power at the slack bus. 3.3. Inequality constraints The inequality constraints include the restrictions (3) and (4) and voltage bounds at each PQ-bus
Fig. 4. A low voltage distribution system for the case study. Table 1. LV-Demand Daily Peak and Power Factors

Bus
3 4 5 6 7 13

Ppeak(kW)
1 3 16 3 1 3

PF
0.9 0.9 0.9 0.9 0.9 0.9

Bus
17 20 25 26 27 29

Ppeak(kW)
5 3 3 3 3 3

PF
0.9 0.9 0.9 0.9 0.9 0.9

Vmin (i ) V (i, h) Vmax (i ), i N (i S )

the active and reactive bounds at the slack bus

(13)

PS.rev.max PS (h) PS.fw.max QS.max QS (h) QS.max

the feeder apparent power capacity bounds

(14) (15)
Table 2. Load Data for the System under Study As a Percentage of the Annual Peak Load Hour Winter Spring Summer Fall 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 0.4757 0.4473 0.426 0.4189 0.4189 0.426 0.5254 0.6106 0.6745 0.6816 0.6816 0.6745 0.6745 0.6745 0.6603 0.6674 0.7029 0.71 0.71 0.6816 0.6461 0.5893 0.5183 0.4473 0.3969 0.3906 0.378 0.3654 0.3717 0.4095 0.4536 0.5355 0.5985 0.6237 0.63 0.6237 0.5859 0.5796 0.567 0.5544 0.567 0.5796 0.6048 0.6174 0.6048 0.567 0.504 0.441 0.64 0.6 0.58 0.56 0.56 0.58 0.64 0.76 0.87 0.95 0.99 1 0.99 1 1 0.97 0.96 0.96 0.93 0.92 0.92 0.93 0.87 0.72 0.3717 0.3658 0.354 0.3422 0.3481 0.3835 0.4248 0.5015 0.5605 0.5841 0.59 0.5841 0.5487 0.5428 0.531 0.5192 0.531 0.5428 0.5664 0.5782 0.5664 0.531 0.472 0.413

S (i, j , h) Sl.max (i, j ),

i, j N (i j )

(16)

and bounds of the curtailment factors

0 curt (l , h) 1.
4. A CASE STUDY

(17)

We solve the above formulated optimization problem for a real three phase balanced 29-bus LV-DN, as shown in Fig. 4. This network was studied in [11]. It contains 29 LV buses at 0.4 kV connected to a medium voltage bus at 20 kV. The daily demand peak and power factors (PFs) are given in Table 1. The load data as a percentage of the annual peak load is given in Table 2. The rated active load is 47 kW. We assume that there is a PVS being connected at each load bus, i.e., at buses 3,4,5,6,7,13,17,20,25,26,27, and 29, respectively. Each PVS has a capacity equal to 9 kVA. These assumptions are made based on the study in [9]. The main feeder capacity is restricted by its maximum capacity of 75 kVA. Values in per unit system are given on 100-kVA base, unless otherwise specified. The bus number 1 is selected as the slack bus (1.05 fixed amplitude voltage and 0 phase angle), whereas the rest are considered as PQ-buses (the upper and lower limits of the amplitude voltage are 1.06 and 0.94, respectively).

Real data of a PVS from a city in Germany is used as PV power penetrations. The Cpr.art is chosen as 0.40$/kWh [1], whereas the on- and off-peak prices Cpr are assumed to be 0.10$/kWh and 0.05$/kWh [5], respectively. The problem is solved by using the general algebraic modeling system (GAMS), whereas other calculations are done in MATLAB.

384

5. OPTIMIZATION RESULTS AND DISCUSSIONS The optimization problem defined in Fig. 3 is solved for two modes. In the switch-off mode, the reactive power capability are deactivated, i.e., Qdisp = 0. In the switch-on mode, Qdisp is defined as an optimization variable. For more details about the physical meaning of the reactive power capability, see [6]. The results in Table 3 show that a significant amount of saving can be gained in the total objective by the switching on mode. This gain is obtained mainly from the first term F1, where a large amount of PV power and energy will be lost when the first mode, i.e., without-Q dispatch, is being used. The second term F2 has an unexpected value in the second mode, i.e., with-Q dispatch. It is commonly recognized that controlling the reactive power dispatch would lead to minimized energy losses, but here our result shows it leads to an increase of energy losses. This interesting point comes from the fact that the increase in the first term F1 is much greater than the decrease in the second term F2. The total effect leads to the result that more gain can be achieved. The value of the third term F3 is not changed, because the demand is the same in the first and second mode. Our aim of calculating F3 is for a short-term (one year) analysis. Figure 5(a) and 5(b) show the voltage amplitude and voltage angle profiles at bus 17 during typical seasonal days in a year. It can be clearly seen that the voltage amplitude rises extremely during midday hours. Figure 5(c) and 5(d) show import and export of the active and reactive power of the network. The negative part during midday hours means that the energy is being exported through the slack bus. In addition, it can be seen that more export can be gained with the with-Q dispatch mode. It is interesting to see in Fig. 5(d), that a huge amount of reactive power needs to be imported. This amount is needed to cover the reactive power absorption from the PV-PCSs inside the network. In other words, these PV-PCSs work during midday hours of heavily sunny days as an inductive demand and absorb a huge amount of reactive energy in order to hold the voltages at the buses and keep them within their predefined ranges. Figure 6 shows the active power losses during the same selected days. It is clearly seen that a large amount of losses occurs during days in spring. This means that the solution provides a strategy to convert a large amount of PV active energy to energy losses instead of curtailing and spilling it out from the network. In other words, this converted energy is being remunerated by governmental regulations, because it passes through the meter and is regarded as PV energy generation, as seen in Fig. 4. Another important conclusion from the optimization is that a BSS is not needed to accommodate spilled energy, because there are no curtailments when using the with-Q dispatch mode. But if reactive power dispatch is not considered, there will be power curtailments and spilled energy, as depicted in Fig. 7(a) and 7(b).
Criterion Total objective ($/year) F1 ($/year) F2 ($/year) F3 ($/year) SB (kVA) EB (kWh)

Table 3. Results of the Analysis First-Option Second-Option Without Q With Q 13100 34699 779 20820 37 207 16839 38858 1199 20820 0.0 0.0

Diff. 3739 (+28.54%) 4159 (+11.98%) -420 (-53.91%) 0.0 (0.0%) 37 (100%) 207 (100%)

Fig. 5. Typical four days: (a) Voltage amplitude and (b) voltage angle of bus No. 17. (C) Active and (d) reactive power import/export at slack bus. Note: from (a) to (d) the thin-blue lines stand for the option without-Q and the bold-red lines stand for the option with-Q.

385

utilized. Moreover, there is no need in this case study to install a battery storage system (BSS) to accommodate spilled PV active energy, because no further spilled energy is present. The aspects and results presented in this paper can be used for planning and operating future energy systems. 7. REFERENCES
[1] [Online]. Available: http://www.erneuerbare- energien.de/ files/english/pdf/application/pdf/eeg_2009_verguetungsdegres sion_en_bf.pdf/ (2012) [2] Y. M. Atwa, and E. F. El-Saadany, Optimal allocation of ESS in distribution systems with a high penetration of wind energy, IEEE Transactions on Power Systems, Nov. 2010, Vol. 25, No. 4, pp. 1815-1822. [3] C. Changsong, D. Shanxu, C. Tao, L. Bangyin, H. Guozhen, Optimal Allocation and Economic Analysis of Energy Storage System in Microgrids, IEEE Transactions on Power Electronics, vol.26, no.10, pp.2762-2773, Oct. 2011 [4] A. Gabash, P. Li, Evaluation of reactive power capability by optimal control of wind-vanadium redox battery stations in electricity market, Renewable Energy & Power Quality J., no. 9, pp. 1-6, May 2011. [5] A. Gabash, P. Li, Active-Reactive Optimal Power Flow in Distribution Networks with Embedded Generation and Battery Storage, IEEE Transaction on Power Systems, in press, 2012, (DOI: 10.1109/TPWRS.2012.2187315). [6] L.J. Borle, M.S. Dymond, C.V. Nayar, Development and testing of a 20-kW grid interactive photovoltaic power conditioning system in Western Australia, IEEE Transactions on Industry Applications, Mar/Apr. 1997, Vol. 23, No. 2, pp. 502-508. [7] R.J. Konopinski, P. Vijayan, V. Ajjarapu, Extended Reactive Capability of DFIG Wind Parks for Enhanced System Performance, IEEE Transactions on Power Electronics, Aug. 2009, Vol. 24, No. 3, pp. 1346-1355. [8] K. Zou, A. P. Agalgaonkar, K. M. Muttaqi, S. Perera, Distribution System Planning With Incorporating DG Reactive Capability and System Uncertainties, IEEE Transactions on Sustainable Energy, Jan. 2012, Vol. 3, No. 1, pp. 112-123. [9] R.A. Shayani, M.A.G. de Oliveira, Photovoltaic Generation Penetration Limits in Radial Distribution Systems, IEEE Transactions on Power Systems, Aug. 2011, Vol. 26, No. 3, pp. 1625-1631. [10] R. Tonkoski, L.A.C. Lopes, and T.H.M. El-Fouly, Coordinated Active Power Curtailment of Grid Connected PV Inverters for Overvoltage Prevention, IEEE Transactions on Sustainable Energy, April. 2011, Vol. 2, No. 2, pp. 139147. [11] A. Bracale, R. Angelino, G. Carpinelli, M. Mangoni, D. Proto, Dispersed generation units providing system ancillary services in distribution networks by a centralised control, IET Renewable Power Generation, July. 2011, Vol. 5, No. 4, pp. 311-321. [12] A. Cagnano, E. De Tuglie, M. Liserre, R.A. Mastromauro, Online Optimal Reactive Power Control Strategy of PV Inverters, IEEE Transactions on Industrial Electronics, vol.58, no.10, pp.4549-4558, Oct. 2011. [13] [Online]. Available: http://www.entsoe.eu/ (2012)

Fig. 6. Power losses in typical four days: the thin-blue line stands for the option without-Q dispatch and the bold-red line for with-Q dispatch.

Fig. 7. (a) Total PV power curtailments. (b) Total PV spilled energy. Note: Both of (a) and (b) are in the mode of without-Q dispatch.

SB and EB in Table 3 stand for the maximum power curtailed at an hour and maximum energy spilled at a day, respectively. SB and EB are used as criteria, for instance in [2], to indicate a BSS size. Now, it is clearly seen that no BSSs are needed to accommodate such spilled energy. It is worth mentioning here that reactive energy to be imported is much cheaper than the PV energy [13]. 6. CONCLUSIONS In this paper, we formulated a mathematical model for active-reactive optimal power flow to analyze low voltage distribution networks with a high penetration of photovoltaic (PV) based DG units. Our aim is to reveal the impact of controlling and utilizing DG reactive power capabilities on the operations of distribution networks. Some interesting points have been found and we discussed the results through a case study. It is shown that a huge increase in the total yield (more than 25%) can be gained if the reactive power capability of PV-based DG units is optimally

386