Electrical Power and Energy Systems 33 (2011) 12431250

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Effect of load models on assessment of energy losses in distributed generation planning

Kejun Qian a,, Chengke Zhou a, Malcolm Allan a, Yue Yuan b
a b

School of Engineering and Computing, Glasgow Caledonian University, G4 0BA Glasgow, UK College of Energy and Electrical Engineering, Hohai University, Nanjing 210098, China

a r t i c l e

i n f o

a b s t r a c t
Distributed Generation (DG) is gaining in signicance due to the keen public awareness of the environmental impacts of electric power generation and signicant advances in several generation technologies which are much more environmentally friendly (wind power generation, micro-turbines, fuel cells, and photovoltaic) than conventional coal, oil and gas-red plants. Accurate assessment of energy losses when DG is connected is gaining in signicance due to the developments in the electricity market place, such as increasing competition, real time pricing and spot pricing. However, inappropriate modelling can give rise to misleading results. This paper presents an investigation into the effect of load models on the predicted energy losses in DG planning. Following a brief introduction the paper proposes a detailed voltage dependent load model, for DG planning use, which considers three categories of loads: residential, industrial and commercial. The paper proposes a methodology to study the effect of load models on the assessment of energy losses based on time series simulations to take into account both the variations of renewable generation and load demand. A comparative study of energy losses between the use of a traditional constant load model and the voltage dependent load model and at various load levels is carried out using a 38-node example power system. Simulations presented in the paper indicate that the load model to be adopted can signicantly affect the results of DG planning. 2011 Elsevier Ltd. All rights reserved.

Article history: Received 2 June 2010 Received in revised form 21 December 2010 Accepted 12 April 2011 Available online 18 May 2011 Keywords: Distributed generation Load model Energy losses Voltage prole Load forecasting

1. Introduction The load in a distribution system generally consists of three main types, i.e., residential, commercial and industrial load, with their proportion in the total load demand varying with time, e.g., hourly, daily and seasonally. The nature of these three types of loads is such that their active and reactive power components respond differently to variations in the voltage and frequency of the system [1,2]. System planners need to understand the exact nature of load sensitivity of voltage in order to precisely quantify the economic benets of installing DG. Korunovic et al. [3] studied the static load characteristics of a medium-voltage distribution network by conducting eld measurements and concluded that steady-state distribution load can be modelled as exponential voltage-dependent model in a relatively wide voltage range, from 0.96 to 1.1 of the nominal voltage with errors less than 5%. The variation in the actual power demand with voltage has become more prominent in recent years as increasing penetration of

Corresponding author. Address: School of Engineering and Computing, Glasgow Caledonian University, 70 Cowcaddens Road, G4 0BA Glasgow, UK. Tel.: +44 141 331 8919. E-mail address: Kejun.Qian@gcu.ac.uk (K. Qian).
0142-0615/$ - see front matter 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijepes.2011.04.003

renewable DGs, such as intermittent wind, has made voltage proles on the distribution feeders more dynamic. The connection of DG to distribution networks near load centre could change magnitude and direction of network power ows. This would impact on network operation and planning practices of distribution companies with both technical and economic implications. Investigations have therefore been carried out into DG planning in distribution systems in recent years [413], among which power system losses reduction due to the introduction of DG in distribution systems attracts much attention [1013]. Most of these investigations focused on assessing the power loss reductions brought about by DG and utilised a constant load model in the power ow analysis, that is, the load power was considered to be independent of variations in feeder voltage. In general, the regulator sets a loss target for each of the UK distribution network operator (DNO). DNOs are rewarded if their real losses are lower than the loss target. Otherwise, the DNOs are economically penalised. Although at present the economic incentives to reduce losses are on the DNOs, it is possible that DNOs will pass part of the reward to DG owners for assisting reducing network losses in the future [12]. However, an accurate quantication of energy losses associated with DG largely depends on the load models employed in the power ow algorithms. Therefore, the load models will have a direct consequence on DNOs prot.

1244

K. Qian et al. / Electrical Power and Energy Systems 33 (2011) 12431250

Normalised load, p.u.

Frantz et al. [14] highlighted that the load model can signicantly affect the predicted system performance. A detailed voltage-dependent load model was adopted in an optimal capacitor installation/switching study by Rizy et al. [15] in a distribution system where the authors demonstrated that the detailed load model accurately predicted that the active power injection increased with the voltage increase whilst the constant load model failed to do so. Singh and Misra [16] performed a comparative study of real and reactive power loss, real and reactive power intake at the main substation and MVA support provided by installing DG resources for different type of load models. The authors in [16] highlighted that load models can signicantly affect the optimal location and sizing of DG resources in distribution system and they also integrated the detailed load models in DG planning using multiobjective optimisation [17]. Qian et al. [18] carried out a study of the effect of load models on the assessment of power losses when a high penetration level of DGs is integrated in LV distribution systems. These studies successfully accounted for load models in DG planning, however, only catered for a single load level and failed to take into account the variations in renewable DG power output and making them impossible to determine the actual impact of variable forms of renewable DGs. As renewable DG is highly variable power source. At any instant in time its export will alter the existing power ows and bus voltages in the direct path from the connected bus to the substation. Subsequently, the voltage and voltage-dependent load at all buses will change at each time instant. Therefore, with a high penetration level of renewable energy generation, a single power ow solution can no longer describe the possible system states in a representative way. The focus of the present paper is to evaluate the signicance of load models on assessment of energy losses in a distribution system with high penetration level of DG, using a 38-bus example power system as a vehicle. In order to capture the effects that the variability of both demand and renewable DGs has on the performance of distribution systems, this paper proposes a methodology based on time series simulations to assess the energy losses of distribution systems with high penetration level of renewable DGs. 2. System modelling and simulation 2.1. The example system The example system shown in Fig. 1 [19] will be used for the study. The voltage rating of the radial system is 11 kV. DGs are to be connected to the system as embedded generations, instead of being connected directly to the Grid Supply Point (GSP, dened as the point of supply from the national transmission system to the local system of the distribution network operator or nonembedded customers [20]). Details of the system can be found in Table A1 in Appendix A. Simulations are carried out for a calendar year to analyse the impact of load models on distribution losses under different scenarios. The computation of energy losses is on a half-hourly basis

1.05

Load demand, p.u.

1.00 0.95 0.90 0.85 0.80 0.75 0.70 1 2 3 4 5 6 7 8 9 10 11 12

Month
Fig. 2. Monthly load level of the studied system.

1.2

Load demand, p.u.

1.0 0.8 0.6 0.4 2

Typical winter load Typical summer load

10 12 14 16 18 20 22 24

Time of day, h
Fig. 3. Normalised load proles in winter and summer of the studied system.

1.0 0.8 0.6 0.4 0.2

Residential Industrial Commercial

2 4 6 8 10 12 14 16 18 20 22 24

Time of day, h
Fig. 4. Normalised load patterns for the three load classes [22].

which requires half-hourly demand data; however, the 38-node sample system does not provide half hourly load data. As a result, the data given in Table A1 is assumed to be the peak load and loads for the rest of time are obtained by assuming the same load variation as real historical data in Figs. 2 and 3. The power demand at each of the 38 nodes arising from each of the residential, industrial and commercial sectors are assumed to follow the pattern shown in Fig. 4, with detailed gures in Table A2. Power system simulation was carried out using the Interactive Power Systems Analysis (IPSA) software [21]. 2.2. Components of load power demand Fig. 2 shows the typical annual load prole in the UK. The daily demand prole varies over the seasons of a year as given in Fig. 3 where the peak demand occurred on 23rd January 2008, and the minimum demand occurred on 22nd June 2008 [22]. Fig. 4

Fig. 1. 38-node radial distribution system [19].

K. Qian et al. / Electrical Power and Energy Systems 33 (2011) 12431250 Table 1 Typical load types and exponent values [24]. Load type Residential npr Summer Winter Constant impedance Constant current Constant power Day Night Day Night 0.72 0.92 1.04 1.30 2.0 1.0 0.0 nqr 2.96 4.04 4.19 4.38 2.0 1.0 0.0 Commercial npc 1.25 0.99 1.50 1.51 2.0 1.0 0.0 nqc 3.50 3.95 3.15 3.40 2.0 1.0 0.0 Industrial npi 0.18 0.18 0.18 0.18 2.0 1.0 0.0 nqi 6.00 6.00 6.00 6.00 2.0 1.0 0.0

1245

1.0

Wind power output, p.u.

0.8 0.6 0.4 0.2 0

Note: subscripts r, c, i represent residential, commercial and industrial load, respectively.

0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000

Number of hours
Fig. 5. Wind power output [22].

illustrates the three components of the typical load pattern, i.e. residential, industrial and commercial. The magnitudes are normalised by the daily peak power demand [23]. 2.3. Voltage-dependent load model In conventional load ow analysis, it was assumed that active and reactive power loads are constant values, regardless of the magnitude of voltages in the same bus. Due to the voltagedependence of distribution load, in this paper, practical voltagedependent load models, i.e., residential, commercial and industrial given in [24] has been adopted. A voltage dependent load model is a static load model that represents the power relationship to voltage as an exponential equation, which can be mathematically expressed as,

put during a whole year [23]. A wind farm is simulated as a single wind power unit with installed capacity equal to the total wind farm capacity. It is now common that wind turbines use doubly-fed, variable speed induction generators with an IGBT converter in the rotor circuit, which enables the DG to operate at any desired power output [25]. Therefore DGs with capability of controlling reactive power (leading/lagging power factors) and those operating at unity power factor are both considered in the paper. 2.5. Scenarios In this study, it is assumed that a single DG (aggregated DGs) is connected to one of the load nodes on the distribution feeder. The study will take into account various DG penetration levels (penetration level is dened as the ratio of the amount of DG power injected into the network to the total feeder load), DG location, and reactive power control strategies of DG. DG penetration level is gradually increased from 0% (base case, i.e. without DG) to 40%, which is the long term target set in Scotland, UK for 2020. According to the current Grid Code requirements in the UK, any medium or large wind farms must provide a reactive power capability of 0.95 lead/lag at its Grid Entry Point [26] (GEP, dened as a point at which a Generating Unit directly connected to the National Electricity Transmission System, or connects to distribution systems [20]); therefore there are three particular situations to be studied for energy loss analysis of the studied system where the DG at the GEP is represented by a virtual power plant which can provide a reactive power capability of 0.95 leading, unity and 0.95 lagging, while the voltage at the GEP should be within the operating limits.

PL PL0 V np Q L Q L0 V nq

1 2

where np and nq are active and reactive power exponents, respectively. PL and QL are the values of real and reactive powers, while PL0 and QL0 are the values of active and reactive powers at nominal voltages, respectively. V is the voltage magnitude at a load node. Eqs. (1) and (2) neglect the frequency-dependence of distribution load, due to the fact that the range of frequency variation is relatively narrow. Common values for exponents of static loads are given in Table 1 [25] in order to evaluate the effects of various load models on DG planning. For practical application, the evaluation of coefcients np and nq requires eld measurement and parameter estimation techniques. It is assumed in the studied system that each load node consists of three components of load consumption. The proportion of each load type in the total load is varying across the time scale, as shown in Table A2. Let a, b, and c are the percentages of residential, commercial and industrial load at each load node respectively (a, b, and c is determined by the values shown in Table A2), the voltage dependent load model can therefore be expressed as follows,

3. Methodology In order to estimate the effect of load models on the assessment of energy losses in a distribution system with high penetration of DGs, a methodology is developed to compute annual energy losses based on time series simulation. The single power ow problem in conventional load ow analysis is expanded to multiple simulations with individual values for each time step. In order to account for the variability of daily load prole and wind power output, the method calculates system losses for every half hourly period during a calendar year. This requires running the load ow for each time interval (half an hour), allowing for consideration of the detailed load model and DG power output during that time interval. The impact of load models on energy losses are measured as the difference between losses with traditional constant load model and the proposed detailed load model in the considered scenarios of DG penetration.

PL aPL0 V npr bPL0 V npc cPL0 V npi Q L aQ L0 V nqr bQ L0 V nqc cQ L0 V nqi

2.4. Distributed generation model

3 4

In this paper, wind generators are considered in the simulation as the DG sources. The power produced by wind turbines was estimated using the characteristics provided by the manufacturers, taking into account the half-hourly wind speed data. Fig. 5 shows the wind power out-

1246

K. Qian et al. / Electrical Power and Energy Systems 33 (2011) 12431250

The complete procedure of the developed approach is shown in Fig. 6. In order to quantify the impacts of load models on the results of energy loss reductions due to the presence of DG during a whole year, two indices are developed. 3.1. Voltage prole Network losses largely depend on the voltage prole when the system load is represented by a voltage dependent model. In order to quantify the changes in voltage prole, due to the introduction of DGs, across the network during a period of time, it is necessary to develop an index of voltage prole. The proposed voltage index quanties the improvement in the voltage magnitude in a simple manner with the inclusion of DG. It is dened as the difference between actual voltage prole and the nominal voltage (1.0 p.u.). Since the simulation will be carried out on a half-hourly basis, the proposed voltage prole index is developed allowing for consideration of time-varying voltage magnitude and load demand at each time interval, shown as below,

level due to the reverse power ow along the lines. The lower the energy loss index, the better the benets in terms of power loss reduction due to DG. 4. Results and analysis In this section, results are presented with the proposed approach to evaluate the effect the load models have on the results of the energy losses calculation. As assumed in this study, a single DG will be installed alternately on various nodes along the feeder. The DG penetration level will gradually increase from 0% to 40%. DGs are modelled with three power factors: 0.95 leading, unity and 0.95 lagging, respectively. 4.1. Impact of load models on energy losses with various DG penetration levels Table 2 summaries the results obtained for various DG penetration levels, using an example in which a single DG (operating at 0.95 lagging power factor) is located at node 14 of the studied feeder. The load models Const. and Volt. are abbreviations of the constant and voltage dependent load model, respectively. It can be observed from Table 2 that when DG is connected to the feeder with a low penetration level (630%), both the voltage prole and the energy loss indices decrease with the increase in DG penetration level until they reach a minimum level. Once the minimum level is reached, the voltage index begins to increase due to the voltage improvement by DG, while energy loss index continues to decrease. It should be noted voltage at each bus should be maintained within 6% of the nominal voltage (1.0 p.u.). Therefore, the voltage index improvement listed in Table 2 due to the DGs indicates a signicant impact on voltage prole. Fig. 7 shows the difference in energy losses (extracted from Table 2) when the two load models are employed. It can be observed from Fig. 7 that when a low penetration level of DG connected to the studied system, simulation with constant load model produces conservative results, i.e., the energy loss index KLoss with constant load model is greater than that with voltage dependent load model. However, as DG penetration level increases to a higher level, simulation with constant load model starts to show optimistic results. The reason is that although DG can improve system voltage prole with a low penetration level, the voltage magnitudes at all feeder nodes except the root node are still less than 1.0 p.u., leading to lower power consumption when voltage-dependent load is adopted. As DG penetration increases to a critical level, the load demand resulting from voltage improvement due to the DG exceeds the constant load. The increase in load would in turn lower partly the energy loss reduction brought by

Pnb Pnc K volt

i1

j1 jV ij V 0 j 100% nb nc

where Kvolt is the voltage index, nb is the number of load nodes in the studied system, nc is the number of time intervals (half-hourly), Vij and V0 are voltage magnitudes for actual and nominal voltage (1.0 p.u.), respectively, at bus i at time instant j. This index should be used only after making sure that voltage magnitude at each load node is within its allowable minimum and maximum limits, typically between 0.94 p.u. and 1.06 p.u., reecting the UK practice. Therefore, the less the voltage index, the better the voltage prole. 3.2. Energy losses Annual energy losses index is dened as the ratio of the total energy losses to the total energy demand on the feeder on an annual energy basis, which can be expressed as below,
j1 P Lossj K Loss Pnb Pnc 100% j1 P Lij i1

Pn c

where KLoss is the energy losses index, PLossj is the system energy losses at time instant j. While DG may reduce energy losses by reducing power ows, it can increase losses at a high penetration

Table 2 Simulation results for various DG penetration levels. DG penetration (%) 0 10 20 30 40 Fig. 6. Flowchart of the developed methodology. Load model Const. Volt. Const. Volt. Const. Volt. Const. Volt. Const. Volt. Voltage index Kvolt (%) 4.425 3.942 3.152 2.780 2.426 2.196 2.125 2.019 2.143 2.196 Loss index KLoss (%) 4.972 4.523 3.941 3.595 2.968 2.727 2.043 1.878 1.616 1.638

K. Qian et al. / Electrical Power and Energy Systems 33 (2011) 12431250

1247

3.5

With constant load model With voltage-dependent load model

Difference in energy losses between the two load models K loss, %

4.0

20
Simulation results Curve fitting

Energy loss index, %

3.0 2.5 2.0 1.5 1.0 0.5 0

10% 20% 30% 40%

15 10 5 0

1.5

2.5

3.5

4.5

Voltage index K volt, %

Fig. 8. Correlation between the voltage index and the difference in energy losses index by using the two load models in power ow analysis.

DG penetration level
Fig. 7. Differences in energy losses index between employing constant and voltage dependent load models.

Energy loss index, %

DG, hence, an optimistic result due to the constant load model. In such a case, if the DNO chooses to reward the DG owners to help reducing power losses, DNO will pay more than actual benets brought by DG if constant load model is employed in the energy loss analysis. This would not be revealed if proper load model in energy loss assessment was not taken into account. The critical penetration level between conservative and optimistic results due to constant load model for wind power generation is 32.5% for the studied system. However, it may vary among various power systems depending on the system topology and load model parameters. Another phenomenon has also been observed in Fig. 7. The difference in energy loss index in absolute value monotonically decreases when the DG penetration level is less than 30%. This is due to the fact that, at the base case (without DG), the voltage prole of the studied system is less than 1.0 p.u., making the actual load appear to be less than the nominal load demand due to the voltage dependence of load. As DG penetration level increases, the voltage prole increases and brings a rise in load demand (within the studied penetration level from 0% to 30% in this case, the greater the DG penetration, the closer the load to the nominal load), consequently lower value of difference in simulation results between voltage-dependent load and constant load models is observed (simulation with constant load model always uses nominal load, which is based on nominal voltage). Since the difference in energy losses between using the two load models is caused by the deviation in voltage from its nominal value due to DG, it is necessary to understand to what extent the variation in voltage can affect energy losses from a mathematical point of view. It is dened in this paper that the difference in energy losses (DKLoss) between using constant load and voltage-dependent load models, is determined by,

the voltage index (absolute value) and the difference in energy loss index (absolute value) by using the two load models can be best t by a linear function. Fig. 8 reveals that: (1) the difference in energy losses between the constant load model and the voltage-dependent load models increase monotonically with the increase in voltage index; (2) a 1% variation in voltage index due to the connection of DG (operating at 0.95 lagging power factor) to the studied system will result in 3.58% energy losses difference, between adopting constant load and voltage-dependent load models in power ow analysis. 4.2. Impact of load models on energy losses with various reactive power control strategies Although utilities favour and even require DGs to operate at lagging power factor (DG supplies reactive power to utility grid), various DGs are likely to operate at unity or leading power factors. Energy losses are calculated for all scenarios in which DGs operate with power factors in the range of 0.95 lagging to 0.95 leading (DG draws reactive power from utility grid) as shown in Fig. 9. In order to study the difference in energy losses between the constant load model and the voltage-dependent load model when various reactive power control strategies of DG are employed, the penetration level of DG is xed at 20%. Table 3 shows the simulation results associated with various DG operating power factors. It can be observed from Table 3 that both voltage and energy loss indices decrease as the DG reactive control strategy moves from leading power factor through unity to lagging power factor.

4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0 0.95 leading unity 0.95 lagging
With constant load model With voltage-dependent l d model lt d d t load d l

DK Loss

K Loss-con K Loss-vol 100% K Loss-con

where DKLoss is the difference in energy losses between using constant load and voltage-dependent load models. KLoss-con and KLoss-vol are energy losses calculated with constant load and voltage-dependent load model, respectively. A positive symbol of DKLoss indicates that simulation results calculated with constant load model is greater than that with voltage dependent load model. The correlation between the voltage index and the difference in energy losses by using the two load models is quantied in Fig. 8. Regression analysis is performed in order to obtain a mathematical correlation for energy losses. It is found that correlation between

DG operating power factor

Fig. 9. Differences in energy losses index between employing constant and voltage dependent load models with various DG operating power factors.

1248

K. Qian et al. / Electrical Power and Energy Systems 33 (2011) 12431250

Table 3 Simulation results for various DG operating power factors. DG power factor 0.95 Leading Unity 0.95 Lagging Load model Const. Volt. Const. Volt. Const. Volt. Voltage index Kvolt (%) 3.965 3.421 3.354 3.057 2.426 2.196 Loss index KLoss (%) 4.078 3.507 3.474 3.089 2.968 2.727

Energy loss index, %

This interprets that among the above three reactive power control strategies, DG can most benet distribution system when it operates at lagging power factor, i.e., supplying reactive power to power systems. Fig. 9 shows the difference in energy losses between using constant load and voltage-dependent load models with various DG operating power factors. As shown in Fig. 9 the study reveals that: (1) the DG operating power factor can play a vital role in line-loss reduction, (2) energy losses index exhibits lower value under lagging power factor conditions and decreases as the power factor moves from leading through unity to lagging for both the constant load and voltagedependent load models. Fig. 10 shows the variation of the difference (DKLoss) in energy losses between the two load models with various power factors as well as DG penetration levels. Fig. 10 reveals that (1) the difference in energy loss results (DKLoss) between using constant load and voltage-dependent load models monotonically increases when the power factor moves from lagging through unity to leading. The reason is that the consumption of reactive power by DG (at leading power factor) degrades the networks power factor and results in greater voltage drop, and therefore a greater difference in energy losses between employing constant load model and voltagedependent load model in power ow analysis, (2) DKLoss decreases as DG penetration level increases for the three operating power factors. The rate of decrease in DKLoss shows a greatest value when DG is operating at lagging power factor. 4.3. Impact of load models on energy losses with variable DG operating power factor Since both the wind power output and system load prole vary with time, in this paper, another reactive power control strategy is proposed to vary the power factors of DG units in different time

periods. This is a strategy to tune renewable DG units to supply reactive power during peak hours and to consume reactive power during off-peak hours. Fig. 11 shows the difference in energy losses between the constant and voltage dependent load models associated with the aforementioned variable reactive power control strategy versus keeping a constant power factor across the studied time scale. The rst three cases assume that DG is working at each time interval (half hourly) with a constant power factor equals to 0.95 leading, unity and 0.95 lagging, respectively. In these three cases, DG is modelled as PQ node. In the fourth case it is assumed that the DG supplies reactive power during peak hours and consumes it during off-peak hours. In this case, DG is modelled as a PV node maintaining the voltage at the reference value set by the network operator. It can be seen from Fig. 11 that the variable power factor has a benecial effect on energy losses. 4.4. Impact of load models on energy losses with various DG locations Consider now the case when DG, operating at 0.95 lagging power factor with a penetration level of 20%, is moved to downstream (e.g., node No. 18) and upstream (e.g., node No. 5) along the feeder. The difference in energy losses between employing constant load and voltage-dependent load models at various load nodes has been studied. Fig. 12 shows the difference in energy losses index between the two load models by installing a single DG alternately at all load nodes along the main feeder.

4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0 0.95 leading unity 0.95 lagging variable
With constant load model With voltage-dependent load model

DG operating power factor

Fig. 11. Differences in energy losses index between employing constant and voltage dependent load models with various DG operating power factors.

Difference in energy loss index between two load models Kloss , %

20 15 10 5 0 -5 10% 20% 30% 40%

Energy loss index KLoss , %

0.95 lagging p.f. Unity p.f. 0.95 leading p.f.

5 4 3 2 1 0 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
With constant load model With voltage-dependent load model

DG penetration level
Fig. 10. Differences in energy losses index between employing constant and voltage dependent load models with various DG operating power factors.

Node no. where DG is located

Fig. 12. Energy loss index by installing a single DG alternately at various locations along the feeder.

K. Qian et al. / Electrical Power and Energy Systems 33 (2011) 12431250

1249

It can be seen from Fig. 12 that (1) losses start to decrease when the distance between DG location and Grid Supply Point (GSP, node No. 1) increases until it reaches a minimum value (at node No. 14). Once the minimum value is reached, energy losses begin to increase slowly as DG location moves to the direction of feeder end. This is because when DG location is moving towards the GSP, the transmission of DG power to distant load centre increases the system losses. When DG is moving towards the load centre (feeder end), losses may also increase due to reversed power ow along the feeder (i.e., from load centre where DG is located to the distribution substation). However, there is a location where DG could benet most to the system in terms of power loss reduction. In this case, locating DG at node 14 results in the least network losses. (2) The difference in energy losses between employing constant load and voltage-dependent load models has a maximum value at node No. 2 (the node next to the GSP). This is because when DG is located at node No. 2, the system has the lowest voltage prole compared to the case that DG is located at other potential locations. As a result, simulation with voltage-dependent load model for DG at node No. 2 resulted in the largest difference in energy loss index between the two load models. 5. Conclusions In this paper, the effect of load models on the assessment of energy losses in DG planning is investigated. The paper comes to the following conclusions: (1) The real and reactive power injection by installation of DG could signicantly affect the system voltage prole. In a future distribution system with high penetration level of distributed generation, where the quantication of DNOs and DG owners benets largely depends on the models of power ow analysis, the assumption of a constant power load model is no longer appropriate. (2) It has been established in this paper that DG planning based on constant power load models could lead to either conservative or optimistic results, depending on the DG penetration level and load model parameters. Simulation results show that for the scenario with 1% variation in voltage prole, there is 3.58% of total losses difference between employing constant load and voltage dependent load models. (3) There are many factors that can affect the difference in simulation results between constant and voltage-dependent load models. In this paper, the impact of DG penetration level, DG location and DG reactive power control strategies on the difference of simulation results between the two load models have been studied. The constant load model would result in conservative results when DG penetration is at low level, while optimistic results at higher penetration level. Meanwhile, the less the distance between DG location to the Grid Supply Point, the greater the difference in energy losses between using constant and voltage-dependent load

models. With respect to the reactive power control strategies, DG operating at leading power factor (DG draws reactive power from the utility grid) is more sensitive to load models than that at lagging power factors. A variable power factor control strategy is proposed to reduce the sensitivity of energy losses to the load model.

Appendix A See Tables A1 and A2.

Table A1 System and load parameters for 38-node system. Branch parameters F 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 2 19 20 21 3 23 24 6 26 27 28 29 30 31 32 8 9 12 18 25 T 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 R/p.u. 0.000574 0.00307 0.002279 0.002373 0.0051 0.001166 0.00443 0.006413 0.006501 0.001224 0.002331 0.009141 0.003372 0.00368 0.004647 0.008026 0.004558 0.001021 0.009366 0.00255 0.004414 0.002809 0.005592 0.005579 0.001264 0.00177 0.006594 0.005007 0.00316 0.006067 0.001933 0.002123 0.012453 0.012453 0.012453 0.003113 0.003113 X/p.u. 0.000293 0.001564 0.001161 0.001209 0.004402 0.003853 0.001464 0.004608 0.004608 0.000405 0.000771 0.007192 0.004439 0.003275 0.003394 0.010716 0.003574 0.000974 0.00844 0.002979 0.005836 0.00192 0.004415 0.004366 0.000644 0.000901 0.005814 0.004362 0.00161 0.005996 0.002253 0.003301 0.012453 0.012453 0.012453 0.003113 0.003113 Loads on to-node PL0/p.u. 0.1 0.09 0.12 0.06 0.06 0.2 0.2 0.06 0.06 0.045 0.06 0.06 0.12 0.06 0.06 0.06 0.09 0.09 0.09 0.09 0.09 0.09 0.42 0.42 0.06 0.06 0.06 0.12 0.2 0.15 0.21 0.06 0.0 0.0 0.0 0.0 0.0 QL0/p.u. 0.06 0.04 0.08 0.03 0.02 0.1 0.1 0.02 0.02 0.03 0.035 0.035 0.08 0.01 0.02 0.02 0.04 0.04 0.04 0.04 0.04 0.05 0.2 0.2 0.025 0.025 0.02 0.07 0.6 0.07 0.1 0.04 0.0 0.0 0.0 0.0 0.0

F: from node, T: to node, R: resistance, X: reactance, PL0, QL0: real and reactive power load at nominal voltage.

Table A2 Load compositions of the case study during a day. Hour Residential Commercial Industrial 1 0.66 0.17 0.17 13 Residential Commercial Industrial 0.14 0.37 0.49 2 0.63 0.17 0.20 14 0.14 0.39 0.47 3 0.60 0.18 0.22 15 0.15 0.46 0.39 4 0.58 0.20 0.22 16 0.18 0.41 0.41 5 0.60 0.23 0.17 17 0.20 0.44 0.36 6 0.55 0.15 0.30 18 0.33 0.47 0.20 7 0.30 0.14 0.56 19 0.60 0.30 0.10 8 0.11 0.32 0.57 20 0.7 0.23 0.06 9 0.10 0.34 0.56 21 0.74 0.19 0.07 10 0.11 0.33 0.56 22 0.76 0.15 0.09 11 0.12 0.37 0.51 23 0.75 0.15 0.10 12 0.17 0.46 0.37 24 0.71 0.16 0.13

1250

K. Qian et al. / Electrical Power and Energy Systems 33 (2011) 12431250 [14] Frantz T, Gentile TIhara S, Simons N, Waldron M. Load behavior observed in LILCO and RG&E systems. IEEE Trans Power Appar Syst 1984;PAS103(4):81931. [15] Rizy DT, Lawler JS, Patten JB, Nelson WR. Measuring and analyzing the impact of voltage and capacitor control with high speed data acquisition. IEEE Trans Power Delivery 1989;4(1):70414. [16] Singh D, Misra RK. Effect of load models in distributed generation planning. IEEE Trans Power Syst 2007;22(4):220412. [17] Singh D, Misra RK. Multiobjective optimization for DG planning with load models. IEEE Trans Power Syst 2009;24(1):42736. [18] Qian K, Zhou C, Allan M, Yuan Y. Load modelling in distributed generation planning. In: International conference on sustainable power generation and supply, 2009 (SUPERGEN 09); April, 2009. p. 16. [19] Baran ME, Wu FF. Network reconguration in distribution systems for loss reduction. IEEE Trans Power Delivery 1989;4(2):14017. [20] National Grid Electricity Transmission plc (National Grid). National Electricity Transmission System: Seven Year Statement; May, 2010. [21] IPSA (Interactive Power Systems Analysis) software. IPSA Power Group, Manchester. [22] National Grid plc. Operational Data. <http://www.nationalgrid.com/> accessed June, 2009. [23] Distributed Generation and Sustainable Electrical Energy Centre. (October 2007). United Kingdom Generic Distribution System. <http://www.sedg.ac.uk/ >. [24] IEEE Task Force on Load Representation for Dynamic Performance, System Dynamic Performance Subcommittee, Power System Engineering Committee. Bibliography on load model for power ow and dynamic performance simulation. IEEE Trans Power Syst 1995;10(1);52338. [25] Ackermann T, Knyazkin V. Interaction between distributed generation and the distribution network: operation aspects. In: Proc second int symp distributed generation: power system market aspects, Stockholm, Sweden; 2002. [26] OFGEM Grid Code Subgroup. Recommendations for the application of Grid Code technical requirements to offshore electricity transmission networks, OFGEM, London, UK; November, 2007.

References
[1] Eminoglu U, Hocaoglu MH. A new power ow method for radial distribution systems including voltage dependent load models. Electr Power Syst Res 2005;76(13):10614. [2] Haque MH. Load ow solution of distribution systems with voltage dependent load models. Electr Power Syst Res 1996;36(3):1516. [3] Korunovic LM, Stojanovic DP, Milanovic JV. Identication of static load characteristics based on measurements in medium-voltage distribution network. IET Gener Transm Distrib 2008;2(2):22734. [4] Carmen LTB, Djalma MF. Optimal distributed generation allocation for reliability, losses, and voltage improvement. Int J Electr Power Energy Syst 2006;28(6):41320. [5] Sudipta G, Ghoshal SP, Saradindu G. Optimal sizing and placement of distributed generation in a network system. Int J Electr Power Energy Syst 2010;32(8):84956. [6] Singh RK, Goswami SK. Optimum allocation of distributed generations based on nodal pricing for prot, loss reduction, and voltage improvement including voltage rise issue. Int J Electr Power Energy Syst 2010;32(6):63744. [7] Fras P, Gmez T, Cossent R, Rivier J. Improvements in current European network regulation to facilitate the integration of distributed generation. Int J Electr Power Energy Syst 2009;31(9):44551. [8] Wang C, Nehrir MH. Analytical approaches for optimal placement of DG sources in power systems. IEEE Trans Power Syst 2004;19(4):206876. [9] Chiradeja P, Ramakumar R. An approach to quantify the technical benets of distributed generation. IEEE Trans Energy Convers 2004;19(4):76473. [10] Sder L. Estimation of reduced electrical distribution losses depending on dispersed small scale energy production. In: Proc of 12th power systems computation conference, Zurich, Switzerland; 1996. p. 122934. [11] Quezada VHM, Abbad JR, Roman TGS. Assessment of energy distribution losses for increasing penetration of distributed generation. IEEE Trans Power Syst 2006;21(2):53340. [12] Payyala SL, Green TC. An estimation of economic benet values of DG, 2007 IEEE power engineering society general meeting, Tampa, USA; June, 2007. p. 18. [13] Costa PM, Matos MA. Avoided losses on LV networks as a result of microgeneration. Electr Power Syst Res 2009;79(4):62934.