Electric Power Systems Research 80 (2010) 13321340

Contents lists available at ScienceDirect

Electric Power Systems Research

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

Fast harmonic simulation method for the analysis of network losses with converter-connected distributed generation
L. Degroote , L. Vandevelde, B. Renders
Electrical Energy Laboratory (EELAB), Department of Electrical Energy, Systems and Automation (EESA), Ghent University, Sint-Pietersnieuwstraat 41, B-900 Gent, Belgium

a r t i c l e

i n f o

a b s t r a c t
The use of distributed generation (DG) in low voltage grids is becoming more common. The impact of the DG units is considered, with an emphasis on the effect of DG on the network losses. These losses consist of fundamental and harmonic losses. The converter-connected DG units have a different inuence on these two terms of the total losses. The aim of this paper is to present a fast harmonic simulation model that can be applied to study the inuence of a DG unit on the total losses and elucidate the importance of the harmonic losses in proportion to the total losses. 2010 Elsevier B.V. All rights reserved.

Article history: Received 23 September 2008 Received in revised form 24 September 2009 Accepted 16 May 2010

Keywords: Harmonics Losses Radial distribution networks Distributed generation Simulation

1. Introduction Electrical power systems are nowadays evolving from centralised systems, where generator plants are connected to the transmission networks, to a decentralised system, with smaller generation units connected directly to the distribution networks and thus near consumption. The main incentives for this evolution are the environmental concerns (use of renewable energy, use of combined heat and power) and the security of supply (diversity of energy sources, markets with large number of actors). The injections of power by the distributed generators may change the magnitude and even the direction of the power ow in the distribution network. This has several implications for the operation and planning of the network and has several technical and economic consequences [123]. In this framework, it is clear that DG units affect the network losses, which are dened as the difference between the energy sent out from the generating stations and the energy metered at the customer premises. Losses occur in all systems of electricity transmission and distribution. These are usually divided into two categories: technical (related to the characteristics of the carrier equipment, supply and demand patterns) and commercial (energy not accounted for, for example, theft and meter errors) [19,24,25].

There are different types of DG units from a constructional and technological point of view [13], which have a different inuence on the distribution network. The impacts of DG on the network are given in the next section, with emphasis on the inuence of DG on the network losses [14,18,19]. The calculation of these network losses is applied in several optimisation programs to quantify an optimal size and location of the DG unit [3,8,16,17,20]. In these programs the power quality can be taken into account. Harmonic losses are rarely considered, although it is known that several DG topologies may have a negative inuence on the total harmonic distortion and thus on the harmonic losses [9]. This paper will present a fast harmonic simulation method. The network considered in the simulations will be modelled as precisely as possible. The presented model can be used to investigate the inuence of converter-connected DG units on the harmonic losses in distribution networks, as will be shown in this paper. The presented simulation model can, for instance, be applied by the Distribution Network Operators (DNOs) to assess the inuence of the DG units connected to their networks.

2. Impact of DG on the distribution networks an overview The impact of the DG unit on the distribution network is observed in several commercial and power quality aspects of the grid. In the following paragraph an overview is given of the most important impacts.

Corresponding author. Tel.: +32 9 264 34 22; fax: +32 9 264 35 82. E-mail address: Lieven.Degroote@UGent.be (L. Degroote). 0378-7796/$ see front matter 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.epsr.2010.05.003

L. Degroote et al. / Electric Power Systems Research 80 (2010) 13321340

1333

2.1. Voltage prole It is commonly known that DG units support the voltage prole along the feeder [26]. However, some drawbacks need to be considered. For instance, voltage regulators can be disturbed when the DG unit is connected closely to the regulator, because there is less power through the regulator. A rule of thumb is that DG units improve the voltage prole if the current injected from the DG unit is less than 5% of the feeder load [2]. The authors of [5] state that the voltage rise caused by a single unit is a function of DG power and the short circuit power of the grid at the point of connection. Larger single units may violate the constraints of maximum voltage rise even if the aggregated generated power in the grid is low. Simulations presented in [11] reveal that the voltage prole and the power ow in the branches are notably impacted by DG in the case of the near-urban or rural networks. Rural networks are more sensitive than urban networks due to the grid impedance at the point of connection. DG units connected close to the MV/LV transformer have less inuence on the voltage prole than those further away. Due to the resistive character of low-voltage networks the active power produced by the DG has a greater inuence on the voltage prole than the reactive power. The research reported in [10] presents an algorithm for Volt/VAr control in distribution networks with DG. Simulations show that, with proper placement of DG and using appropriate controllers, it is possible to obtain a much better Volt/VAr control, along with decreasing the system losses in the network. 2.2. Harmonics Harmonics can be dened as a sinusoidal component of a periodic wave having a frequency that is an integral multiple of the fundamental frequency. Non-characteristic harmonics are those that are not integer multiples of the fundamental power frequency and are usually called inter-harmonics. A special subset of interharmonics is called sub-harmonics. They have a frequency that is less than the fundamental frequency. Lighting icker is one indication of the presence of sub-harmonics [27]. The increasing use of power-electronic equipment has given rise to problems concerning these harmonic components. Most DG units are connected to the grid by means of converters and so they inuence the harmonic distortion in the grid. Several publications have been written regarding the inuence of DG units on the harmonics present in the grid [1,2,4,9,11,12,2123]. In [21], a framework has been developed to evaluate the power quality issues for the interconnection of DG installations to the grid. The power quality issues focused on are the slow and fast voltage variations, icker and harmonic emissions. For the harmonics, an approach in three steps has been presented based on the IEC set of standards. Firstly, the acceptable voltage distortion limits are dened, secondly, the global harmonic voltage limits are allocated to individual users and thirdly, the corresponding harmonic current distortion limits are determined for a specic installation. Measurements and simulations have been performed in [1,9,11,12]. They show the dependency of the harmonic distortion to the grid properties and other loads connected to the grid. The research in [1] shows that the risk of icker increases if the X/R ratio of the grid is low and if wind turbines which have a tendency to produce large periodic power uctuations are used. In [9] an analytical model has been erected to predict the allowable distribution generation resources on a radial distribution feeder before voltage harmonic limits are exceeded. Simulations show that the penetration level is dependent on the type of feeder and the loads connected to the network. Possible interaction between large number of photovoltaic inverters and the distribution net-

work has been studied in [12]. Resonance phenomena have been observed and studied in a low voltage network with a large number of PV-inverters, while the emission level of an individual inverter satises the PQ standards. Other research shows the negative effect of DG units on the harmonics [2], while in [11] the positive effect of DG units on the total harmonic voltage distortion is revealed. Considering the above mentioned research, it is made clear that the inuence of DG units on the harmonic distortion level is dependent on a great deal of aspects. In order to nd solutions for the harmonic emissions due to the connection of DG units to the grid, research has been carried out to control the converter in such a way that it reduces the harmonics in the network [4,22,23]. In [4] the inductor of a current-controlled voltage source inverter is split to connect a damping second order lter at its midpoint. This system minimizes the harmonic problems when variable speed wind turbines are connected to the grid. The research presented in [22] presents a stability analysis of a converter extended with an active power lter devised for utility current harmonic compensation. In [23], the inuence of converterconnected DG units with integrated harmonic voltage damping and harmonic current compensation function on the distortion level of the voltage has been analysed and experimentally veried. 2.3. The network losses The research reported in [14] indicates that both active power losses and reactive power consumption are reduced with small amounts of DG connected to the power system. As penetration increases, the power losses may increase again. In [18] an approach to compute annual loss variations is presented when different penetration and concentration levels of DG are connected to a distribution network. The impact on losses of different primary drivers for DG units, such as combined heat and power, wind power, photovoltaics, and fuel-cells, is analysed. Different types of load ow algorithms have been implemented to calculate the losses. In conclusion, the radial load ow algorithm is preferred in the case of larger networks. The rst result claiming attention is the shape of all traces. Network losses decrease with increasing DG penetration. At a certain DG penetration grade, this trend reserves, and the losses start to increase again. They can even become larger than in the case without DG. Regarding the impact of each type of primary driver, it can be observed that wind turbines have the least positive impact on the annual energy losses (for the same amount of penetration), because the injected energy is intermittent, presenting high time variability, and does not match well with the feeder load pattern. The impact on losses of the rest of the technologies is also explained by the expected matching between hourly energy generation and hourly load patterns. The reactive power supplied or consumed by DG also impacts the energy losses. Furthermore, it is stated that only large DG generators should control voltages in real time. Medium and small generators can keep a constant power factor with time discretisation. The above-mentioned conclusions can also be deduced from [19], but the authors mark that especially in rural networks the losses can increase for higher penetration levels. In contradiction to urban and mixed networks the overall losses are reduced with DG presence. 2.4. Transient stability In [6] the impact of DG on the power system transient stability has been investigated. The authors conclude that the impact of DG on the power system transient stability depends both on the penetration level and the technology of the distributed generators. Unlike asynchronous generators, synchronous generators have more inuence on transient stability. They decrease the over-

1334

L. Degroote et al. / Electric Power Systems Research 80 (2010) 13321340

speeding of the large scale generators, but they also decrease the transient stability by increasing the oscillation duration. Powerelectronic based DG units decrease the overspeeding of generators, because they disconnect during faults. 2.5. Economical aspects In [5] the emphasis lies on the potential cost savings of DG and the inuence of grid support (reduction in grid losses, typically savings of 1015%, voltage support, and power factor correction) from an economical point of view. The main conclusions from simulations are that the effect of DG on the distribution network is largely dependent on the power ow in the network, that losses are reduced by the implementation of DG and that the network reliability can be affected by the increasing presence of DG. Research presented in [7] tackles the problems introduced by Domestic Combined Heat and Power (DCHP). The authors state that the main problems will be commercial, such as loss of distribution use of system revenue, and regulatory. 3. Distribution network loss modelling In this section, a simulation method will be presented in order to perform a fast harmonic calculation of a distribution network. This method will be used in order to investigate the inuence of DG units on the losses in distribution networks and to examine the relative contribution of the harmonic losses to the total losses. This calculation method could be used in algorithms determining the optimal size and placement of a DG unit, with the harmonic losses taken into account. This simulation method can also be used by the DNOs to investigate the inuence of DG units on the distribution network, in order to establish a grid with distributed generation, while the power quality problems are reduced to a minimum. 3.1. Load ow algorithm Next to the commercially available software packages numerous power ow methods have been proposed in literature [2836]. These models all have their own innovative contribution to the development of fast, accurate and robust power ow algorithms. The admittance-matrix-based harmonic analysis method is the most widely used method for the harmonic distribution system analysis. However, this method is time consuming as it does not take fully advantage of the topology characteristics of distribution systems. Furthermore, an LU-decomposition is necessary for each harmonic order. In [33,37,38] the CPU time of the forward/backward method has been compared to the simulation time with the Newton method. Due to the topology and properties of the distribution grid, it was made clear that the forward/backward method is the most adequate to use [33,37,38]. It also allows the introduction of meshes into the grid without large computational burden [28,30]. The presence of the neutral and ground wire has been introduced in [34] and the introduction of harmonics in the power ow algorithm has been presented in [35]. At last, a lot of equivalent circuits of network elements have been presented [29,32] and the inherently present unbalance in distribution networks has been addressed in [35,36]. The contribution of the presented power ow algorithm lies in the accuracy of the equivalent circuits and the use of symmetrical components. The latter leads to a faster algorithm, especially in the cases with a low degree of unbalance. The difference with the research presented in [36] is in the removal of the multi- to singlephase converting algorithm. As compared to [34] this method has the advantage that a 3 3 impedance matrix is used, without the loss of the explicit information about the neutral currents and voltages. Furthermore, the line impedance matrix in symmetrical

Fig. 1. Schematic overview of the forward/backward method.

components is diagonal. Consequently, where in [31] approximations were needed to obtain such an impedance matrix, this is not needed in the current approach. The possibility of the load ow algorithm to consider harmonic components is another advantage. The additional problems introduced by the harmonic components regarding to convergence criteria will be discussed and solved (Section 3.4). The forward/backward method is used to analyse the radial distribution system. A schematic overview of this iterative method is given in Fig. 1. The method can be divided in two parts. The rst part is the backward sweep: based on a known harmonic voltage prole (step A), the currents in every node are calculated (step B). Then, the currents that ow in all network branches can be calculated, starting from the receiving end of the feeder and going towards the sending end (step C). The second part is the forward sweep: starting from the source node and going towards the terminal nodes the voltage at every node is calculated (step D). When these two sweeps are performed, the voltage spectrum in every node is compared with that of the preceding iteration. If the difference in every node lies beneath a given tolerance, the iteration process stops, otherwise a new iteration is performed. The branch currents consist of the different currents drawn from the nodes and the magnetizing currents of the transformers (Section 3.3). These node currents originate from the linear and nonlinear loads in the nodes or from the capacitance in case of a cable segment (Section 3.4). Converter-connected DG units are an additional source of node currents (Section 3.5). 3.2. Distribution network model The above mentioned load ow algorithm will be applied on the network model presented in Fig. 2. This network is based on the IEEE 13 node test feeder [39] and has, just like the majority of the LV public distribution networks, a radial layout, with a number of LV feeders starting from the point of common coupling (pcc). Each feeder may include one or more subfeeders and the loads and DG units can be connected anywhere along the feeder or it subfeeders. On this network several single-phase consumers can be connected, so the simulation method has to allow unbalances. The LV network lines are either underground cable lines, encountered mainly in urban and suburban areas, or overhead lines. In the network used for the simulations both types are present. The cables are represented by grey lines and the overhead lines are represented by black lines (Fig. 2). The distances between the nodes are the same as in the original 13 node network topology but the properties of the line segments have been altered in order to

L. Degroote et al. / Electric Power Systems Research 80 (2010) 13321340

1335

Fig. 2. Topology of the considered radial distribution network.

Table 1 Feeder length and nominal power. Feeder S I II III IIIa IIIb IIIc Length (m) 620 155 248 806 310 155 248 Snom (kVA) 90 38 36 40 9 25 18

Fig. 3. Single node network without reactance pair (a) and with reactance pair (b).

3.4. Load and cable model meet the characteristics of a real-life LV network. Every line section consists of four wires (three phase and neutral) and the properties of the lines are given in Tables 1 and 2. The properties listed in Table 2 are valid for the fundamental component, for the higher order harmonics the resistance has been adjusted according to [40]. The network used is a three-phase four-wire network and consequently, the earthing scheme used is of the TN or the TT type, as mentioned in the classication of IEC 60364. 3.3. Transformer model In order to obtain an accurate network model, the transformers present in the network are not considered as passive elements. The magnetizing current of the transformers has been taken into account. This model was presented in [41] and is obtained by coupling the electrical equations with a magnetic equivalent circuit. This allows computing the currents in the primary and the secondary windings, and thus the magnetizing current, based on the applied voltages. The equations are developed in the frequency domain and symmetrical components are used. The saturation of both legs and yokes is represented by anhysteretic magnetization curves. In every backward current sweep the magnetizing current is injected between the primary and the secondary impedance of the transformers. The MV/LV transformer connection group is Dyn5 and has a nominal power of 400 kVA. The HV/MV transformer is connected in YNd11 and has a nominal power of 25 MVA. The linear loads are modelled as RL impedances. Based on the fundamental direct component of the voltage at the specic node, the impedance is calculated to represent a constant power load. For these linear loads the displacement power factor (dPF) is given, which is used to describe the power factor (PF) using the fundamental frequency components only [42]. When the Total Harmonic Voltage Distortion (THDV ) is negligible, the relation between the PF and the dPF is given by: PF = dPF 1 + (THDI /100)
2

(1)

where THDI is the Total Harmonic Current Distortion (%). Numerical problems can occur when incorporating these linear loads in a harmonic load ow program. For the higher order harmonics, the forward/backward method can become mathematically unstable. To avoid this problem the eigenvalues of the matrix describing the relationship between the currents and the node voltages in iteration k and k + 1 have to be located in the unity circle. In general, the magnitude of the load impedance and the impedance between the node and the sending bus of the feeder determines the magnitude of the eigenvalues. This can be shown by considering the simple case of a network with a single node (Fig. 3(a)). The relation between the current and the voltage in the k th and the (k + 1) th iteration can be expressed as:

ik+1

vk+1

0 z n,b

1 z load z load z n,b z load

ik

vk

(2)

Table 2 The line properties. Feeder S I II III IIIa IIIb IIIc R ( /km) 0. 0439 0. 351 0. 226 0. 0585 0. 351 0. 226 0. 226 L (mH/km) 0. 0573 0. 344 0. 223 0. 086 0. 344 0. 223 0. 223 C ( F/km) 0. 29 0. 29 0. 29

This gives the following eigenvalues:

1 2

=1 z n,b = z load

(3)

A higher load impedance has a positive effect on the stability, where a higher impedance between the node and the sending bus of the feeder has a negative effect. Therefore, a reactance pair [43] is inserted Fig. 3(b). This pair consists of opposite reactances. The voltage used to calculate the current drawn by the linear load is not

1336

L. Degroote et al. / Electric Power Systems Research 80 (2010) 13321340 Table 3 The load distribution. Node 1 2 3 4 5 6 7 8 9 RL-load (kVA) 20 15 7 4 7 3 2 6 2 dPF 0.9 0.95 0.9 0.9 0.95 0.85 0.9 0.9 0.9 NL-load (kVA) 6 5 3 1 2 3 Phase ABC ABC A ABC ABC ABC

the node voltage, but the voltage of the ctitious node (Fig. 3). Physically, the network does not change, but from a mathematical point of view, the load impedance is higher and the line impedance has a lower absolute value. In this way, the forward/backward method can be used for the harmonic analysis of distribution networks. The nonlinear loads used in the simulation are considered to be xed harmonic current sources. To some nodes, single or three phase currents sources will be connected to represent a large number of computer loads. The THDI of these loads is 100.63%. Typical values of these currents are presented in [44]. The currents due to the capacity of the cable are calculated analogously to the linear loads. The main problem, from a mathematical point of view, for the capacitive current calculation is the decrease of the impedance of the capacitive element for higher frequencies while the line impedance increases for these higher order harmonics. For this reason the currents absorbed by these capacitors cannot be calculated [35]. In this paper, the problem is solved as follows. At rst, a reactance pair has been inserted between the capacitor and the line impedance, but this does not solve the problem completely. Therefore, the increase of the capacitive current between two iterations has been limited. The relaxation of the capacitive current results in an injected current that lies between the current of the previous iteration and the calculated current. Consequently, the overall iterative procedure does not end when the voltage has converged, also the currents drawn by the capacitors need to have converged. 3.5. Converter-connected DG model The converter is modelled as a voltage-dependent current source. The fundamental current depends on the voltage in the specic node, so a constant power is injected in the grid. The harmonic currents are dened as a percentage of the fundamental current (k) and have the same phase angle ( 1 ) as the fundamental component. 1 cos(2 hft + 1 ). So, the harmonic currents are given by Ih = kI The harmonic spectra of the converter-connected photovoltaic (PV) systems used for these simulations are based on [45]. These harmonic spectra depend on the ratio of the active power to the nominal power, the lower this percentage, the higher the share of the harmonic currents in relation to the fundamental component. For an active power equal or higher than 60% of the nominal power, the THDI amounts to 2.71%. The THDI rises for lower output powers, up to 11.98% for an output power between 5 and 10% of the nominal power. 3.6. Loss calculation The fundamental (PJ,1 ) and harmonic (PJ,h ) line losses in every feeder are calculated considering the currents in the branches. The losses are calculated separately for every phase and the neutral conductor. The phase currents are determined by using symmetrical components and the neutral current (IN ) equals three times the
Table 5 Test cases, connected DG units. Case Node 1 Node 4 Node 6 Node 8 P (kW) phase P (kW) phase P (kW) phase P (kW) phase 3 A 9 A 1 2 3 A 3 9 A 4 15 A 5 6 7

Table 4 The base case. Node 1 2 3 4 5 6

v1 (pu)
0.973 0.976 0.975 0.974 0.973 0.973

THDVA (%) 2.142 1.908 1.909 3.213 3.431 3.441

THDVB (%) 1.123 0.915 0.915 1.092 1.148 1.154

THDVC (%) 1.152 0.929 0.928 1.097 1.156 1.162

zero sequence component:

2 2 2 2 PJ,1 = Rph,1 (IA,1 + IB,1 + IC,1 ) + RN,1 IN,1 40

(4) (5)

PJ,h =
i=2

2 2 2 2 Rph,i (IA, + IB, + IC, ) + RN,i IN, i i i i

In the calculations of the harmonic line losses the skin effect has been taken into account according to [40]. 4. Simulation scenarios In this section, the presented fast harmonic simulation method will be applied to investigate the inuence of one or more converter-connected DG units on the fundamental and harmonic line losses. The load distribution for the base case is given in Table 3. Due to the nonlinear asymmetrical load in node 5, the distribution network is in an unbalanced state. At rst, a base case will be presented where no DG units are connected to the distribution network (Fig. 2). The absolute value of the direct voltage component and the THDV for the base case are given in Table 4. The inuence of the asymmetrical nonlinear load in node 5 can be seen in this table. Starting from this base case, the inuence of one or several converter-connected DG units on the network losses will be investigated. By considering several cases, some more general conclusions will be drawn about the consequences of the connection of DG units to the grid. This model can thus be used by the DNOs to study the inuence of the DG units. At rst, the inuence of increased output power of a singlephase DG unit will be investigated. Therefore a DG unit is connected

8 3 A

10

11

12

13

14

15 3 A

3 A

9 A 3 A 9 A 3 B 3 C

3 A 3 A 3 A

15 A

3 A

L. Degroote et al. / Electric Power Systems Research 80 (2010) 13321340

1337

to phase A in node 1. The output power of this converter is increased from 1 kW until 30 kW in steps of 1 kW. Furthermore, the inuence of the location on the feeder, the inuence of the unbalanced load distribution and the (in)validity of the superposition principle will be investigated. Therefore, fteen different cases will be considered. The cases all have the same load distribution as presented in Table 3 but in one or several nodes a converter-connected DG unit is connected as shown in Table 5. The DG can be connected in the following nodes: 1, 4, 6 and/or 8.

5. Simulation results In what follows, the losses of the different scenarios presented in Section 4 will be discussed. In general, the fundamental and harmonic losses in the three phases and neutral conductor will be discussed. The results of the losses in the different parts of the distribution network are given in Figs. 47. The losses in Feeder II are not shown, because the different scenarios have almost no inuence on the losses in this feeder. Per case, four bars are depicted. The rst three bars give the total losses for the phases A, B and C respectively, where the fourth bar represents the losses in the neutral conductor. The height of the bars represents the total losses, where the grey part represents the losses caused by the fundamental component of the current and the black part originates from the harmonic currents. First the losses in the base case (case 1) will be discussed and then the inuence of the converter-connected DG units on the losses will be examined and compared with the base case.
Fig. 6. The losses in the supply feeder (Feeder S). Black: harmonic losses, grey: fundamental losses.

5.1. The losses in the base case With a total power consumption of 86 kVA, the Joule losses in the entire network amount to 2.36 kW or 2.7% of the power demand in the network. The harmonic losses claim 10.3% of these total losses. From Figs. 46 can be concluded that for case 1 the main losses are situated in the supply feeder that transports all power from the central generation. The harmonic losses in the phases B or C of the supply feeder run up to 7.2% of the total losses in these phases, but to 15.16% for the total losses in phase A, due to the nonlinear asymmetrical load in node 5. The losses in the neutral conductor (due to unbalance) in this base case account for 0.3% of the total losses in the entire network. 5.2. Inuence of increased output power The inuence of increased output power will be examined by simulating a DG unit in node 1 connected to phase A. The output power of this converter-connected DG unit will be increased in steps of 1 kW. The losses in Feeder I can be found in Figs. 8 and 9. In Fig. 8, the fundamental (full line and left axis) and the harmonic (dashed line and the right axis) losses in phase A of Feeder I can be found. The fundamental losses decrease with increasing output power. From the point where the active power consumption in phase A equals the output of the DG unit, the losses start to increase again. For higher output powers, from approximately twice the active power consumption in phase A, the losses in phase A of Feeder I are higher than in the base case. However, the losses in entire network are still lower than in the base case, till the output power reaches the level of 26 kW. This is caused by the losses in the Supply feeder. The harmonic losses in phase A always grow with increasing output power, and in terms of percentage they can be

Fig. 4. The losses in Feeder I. Black: harmonic losses, grey: fundamental losses.

Fig. 5. The losses in Feeder III and subfeeders. Black: harmonic losses, grey: fundamental losses.

Fig. 7. The losses in the entire network. Black: harmonic losses, grey: fundamental losses.

1338

L. Degroote et al. / Electric Power Systems Research 80 (2010) 13321340

Fig. 8. The harmonic (dashed) and fundamental (full) losses in phase A of Feeder I.

important because at a certain point they equal the fundamental losses. In Fig. 9 the losses in phase B are shown. The fundamental losses always decrease with increasing output power. This is caused by the unbalance created by the DG unit, and the regulation principle of the constant power loads used in the simulations. These constant power loads are controlled in such way that the absorbed power over one period is xed and the current drawn by the load is proportional to the line voltage in the node. Thus, when the DG unit supports the voltage prole in phase A, the power consumption of the other phases decreases resulting in lower line losses. From cases 24 in Fig. 4 can be seen that the losses in the neutral conductor become very signicant. This is caused by the asymmetrical input of power. In case 4, the losses in the neutral conductor of Feeder I are even higher than the losses in phase A. Taking into account the entire network, the neutral conductor losses in case 4 reach the level of 4.7% of the total losses. From cases 57 in Fig. 5, the same conclusion can be drawn. The only difference can be found in the evolution of the neutral conductor losses. These losses are lower in case 5 than in the base case, caused by the compensation of the asymmetrical nonlinear load on the same feeder. 5.3. Inuence of location on the feeder When case 9 and 10 are compared with case 11 and 12, the inuence of the DG placement on a feeder can be investigated. In

Fig. 10. The total losses for the cases 912.

the cases 9 and 10, a 3 and a 9 kW DG unit respectively is placed in node 4. For the cases 11 and 12 the same has been done but more downstream on Feeder III, namely in node 6. The total losses in these cases, subdivided in the fundamental (grey) and the harmonic (black) losses, can be seen in Fig. 10. For a DG unit with an output power of 3 kW, the losses are higher when the DG unit is connected to node 4. The opposite can be noticed when the output power rises to 9 kW. There, the losses are higher when the DG unit is connected more downstream to node 6. The reason for this phenomenon can be found in Table 3. The load consumption in node 4 and 6 lies between 3 and 9 kW. So, in the cases where the output power of the DG unit is 3 kW, less transport losses will be observed when the DG unit is connected more downstream. So, the total losses in case 11 are lower than for case 9. When the output power of the converter equals 9 kW, the opposite can be discerned. The remainder of power produced by the DG unit has to be transported upstream, consequently, less transport losses will be noticed if the DG unit is connected more upstream (case 10). From Fig. 10, the same conclusions as in Section 5.2 can be drawn, namely, the total losses decrease with increasing output power and the harmonic losses become more important with increasing output power.

5.4. Inuence of an unbalanced load distribution The consequence of load unbalance combined with an unbalance in the injected power is revealed by cases 11, 13 and 14. In case 11, the DG unit is connected to the most loaded phase, where for the cases 13 and 14, the DG unit is connected to the least loaded phase. The total losses in case 13 and 14 are 2243 W, where the losses in case 11 are 2210 W. So, a higher reduction compared to the base case (2361 W) is discerned in case 11. This higher reduction has two reasons. Firstly, the higher losses in the neutral conductor for the cases 13 and 14 as can be seen in Fig. 5, and secondly, the connection of a DG unit on a more loaded phase has a larger inuence on the losses, which are quadratic with the current, as shown in the following derivation. To prove the higher loss reduction in case 11, the following condition has to be met (with IDG the current

Fig. 9. The harmonic (dashed) and fundamental (full) losses in phase B of Feeder I.

L. Degroote et al. / Electric Power Systems Research 80 (2010) 13321340

1339

injected by the DG unit).

2 2 2 2 R(I I RIA DG ) >RIB R(IB IDG ) A ?

IA >IB This last statement is true because in the base case the load is unbalanced with a higher consumption in phase A, so the current in that phase IA is larger than the current in phase B (IB ). 5.5. Invalidity of superposition principle In case 8, the inuence of superposition has been tested. This case is a combination of case 2, where the reduction of the total loss is 138 W, and case 5, where the DG unit results in a reduction of 146 W. The reduction in case 8, however, is not 284 W, but 270 W. This can be seen in Fig. 6, where can be seen that superposition of the reduction of losses in phase A for the cases 2 and 5 is larger than the reduction in case 8. From case 15 the same conclusion can be drawn, namely, that the superposition principle is not valid. The reduction of the losses in this case run up to 449 W and not to 582 W. 6. Conclusions This paper presented a fast harmonic calculation model based on the forward/backward sweep based technique. Problems concerning the divergence of the forward/backward technique have been solved by adding relaxation (for capacitive elements) and reactance pairs where needed. The distribution model presented in the paper has the characteristics of a traditional LV network and the magnetizing current of the transformers has been taken into account. The model has been used in order to simulate the inuence of converter-connected DG units on the fundamental and the harmonic losses. The converter-connected DG units give an increase in the harmonic losses, while, until a specic level, the fundamental losses decrease. Therefore, harmonic losses can become signicant when converter-connected DG units are used. It can also occur that the losses in a feeder increase with a high amount of injected power but that the losses in the entire network decrease. When small single-phase DG units are used, the losses in the neutral conductor can become higher. The losses per phase decrease most when the DG unit is connected to the most loaded phase. As additional conclusion one can state that the loss reduction of different cases cannot be summed to nd the loss reduction for the combined case. Acknowledgements The research of L. Degroote is funded by the Special Research Fund (BOF) of Ghent University. The research was carried out in the frame of the Interuniversity Attraction Poles, IAP-VI-021, funded by the Belgian Government. References
[1] T. Thiringer, Power quality measurements performed on a low-voltage grid equipped with two wind turbines, IEEE Trans. Energy Convers. 11 (September (3)) (1996) 601606. [2] P. Barker, R. de Mello, Determining the impact of distributed generation on power systems. i. Radial distribution systems, in: IEEE PES Summer Meeting, 3, Seattle, Washington USA, July 1620, 2000, pp. 16451656. [3] R. Brown, J. Pan, X. Feng, K. Koutlev, Siting distributed generation to defer T&D expansion, in: IEEE/PES Trans. and Distr. Conf. and Expos., vol. 2, Atlanta, Georgia, USA, October 28November 2, 2001, pp. 622627. [4] Z. Chen, E. Spooner, Grid power quality with variable speed wind turbines, IEEE Trans. Energy Convers. 16 (June (2)) (2001) 148154. [5] P. Dondi, D. Bayoumi, C. Haederli, D. Julian, M. Suter, Network integration of distributed power generation, J. Power Sources 106 (2002) 19.

[6] J.G. Slootweg, W.L. Kling, Impacts of distributed generation on power system transient stability, in: IEEE PES Summer Meeting, vol. 2, Chicago, Illinois, USA, July 25, 2002, pp. 862867. [7] D.S. Forrest, A.R. Wallace, Accommodating high levels of domestic generation in the distribution network, in: 17th Int. Conf. on Electricity Distribution (CIRED 2003), Barcelona, Spain, May 1215, 2003. [8] G. Celli, E. Ghiani, S. Mocci, F. Polli, A multi-objective formulation for the optimal sizing and siting of embedded generation in distribution networks, in: IEEE Power Tech Conference, Bologna, Italy, June 2326, 2003. [9] A. Bhowmik, A. Maitra, S. Halpin, J. Schatz, Determination of allowable penetration levels of distributed generation resources based on harmonic limit considerations, IEEE Trans. Power Deliv. 18 (April (2)) (2003) 619624. [10] T. Niknam, A. Ranjbar, A. Shirani, Impact of distributed generation on Volt/VAr control in distributed networks, in: IEEE Power Tech Conference, Bologna, Italy, June 2326, 2003. [11] T. Tran-Quoc, C. Andrieu, N. Hadjsaid, Technical impacts of small distributed generation units on LV networks, in: IEEE PES General Meeting, vol. 4, Toronto, Canada, July 1317, 2003, pp. 24592464. [12] J.H.R. Enslin, P.J.M. Heskes, Harmonic interaction between a large number of distributed power inverters and the distribution network, IEEE Trans. Power Electron. 19 (November (6)) (2004) 15861593. [13] W. El-Khattam, M. Salama, Distributed generation technologies, denitions and benets, Electric Power Syst. Res. 71 (2004) 119128. [14] V. Van Thong, J. Driesen, R. Belmans, Dispersed generation interconnection and its impact on power loss and protection, in: IEEE Young Researchers Symposium in Electrical Power Engineering, Delft, The Netherlands, March 1819, 2004. [15] P. Chiradeja, R. Ramakumar, An approach to quantify the technical benets of distributed generation, IEEE Trans. Energy Convers. 19 (December (4)) (2004) 764773. [16] C. Wang, M.H. Nehrir, Analytical approaches for optimal placement of distributed generation sources in power systems, IEEE Trans. Power Syst. 19 (November (4)) (2004) 20682076. [17] M.E.H. Golshan, S.A. Arefar, Distributed generation, reactive sources and network-conguration planning for power and energy-loss reduction, IEE Proc. Gener. Transm. Distrib. 153 (March (2)) (2006) 127136. [18] V.H. Mndez Quezada, J. Abbad, T.S. Romn, Assessment of energy distribution losses for increasing penetration of distributed generation, IEEE Trans. Power Syst. 21 (May (2)) (2006) 533540. [19] A. Beddoes, C. Lynch, M. Attree, M. Marshall, The impact of distributed generation upon network losses, in: 19th Int. Conf. on Electricity Distribution (CIRED 2007), Vienna, Austria, May 2124, 2007. [20] A.D.T. Le, M. Kashem, M. Negnevitsky, G. Ledwich, Optimal distributed generation parameters for reducing losses with economic consideration, in: IEEE PES General Meeting, 2007 June, pp. 18. [21] S. Papathanassiou, A technical evaluation framework for the connection of DG to the distribution network, Electric Power Syst. Res. 77 (2007) 2434. [22] M. Cirrincioni, M. Pucci, G. Vitale, A single-phase DG generation unit with shunt active power lter capability by adaptive neural ltering, IEEE Trans. Ind. Electron. 55 (May (5)) (2008) 20932110. [23] B. Renders, K. De Gussem, W.R. Ryckaert, L. Vandevelde, Converter-connected distributed generation units with integrated harmonic voltage damping and harmonic current compensation function, Electric Power Syst. Res. 79 (2009) 6570. [24] S. Polley, Technical losses, the role of the regulator, in: IET Power Convention 2007, London, England, September 1011, 2007. [25] B. Renders, L. Degroote, J. Driesen, L. Vandevelde, Prots of power-quality improvement by residential distributed generation, in: Proceedings of the 42th Universities Power Engineering Conference (UPEC07), Brighton, U.K., September 46, 2007. [26] B. Renders, K. De Gussem, W.R. Ryckaert, K. Stockman, L. Vandevelde, M.H.J. Bollen, Distributed generation for mitigating voltage dips in low-voltage distribution grids, IEEE Trans. Power Deliv. 23 (July (3)) (2008) 15811588. [27] G. Singh, Power system harmonics research: a survey, Eur. Trans. Electr. Power 19 (2009) 151172. [28] D. Shirmohammadi, H. Hong, A. Semlyen, G. Luo, A compensation-based power ow method for weakly meshed distribution and transmission networks, IEEE Trans. Power Syst. 3 (May (2)) (1988) 753762. [29] T.-H. Chen, M.-S. Cheng, T. Inoue, P. Kotas, E. Chebli, Three-phase cogenerator and transformer models for distribution system analysis, IEEE Trans. Power Deliv. 6 (October (4)) (1991) 16711681. [30] C. Cheng, D. Shirmohammadi, A three-phase power ow method for realtime distribution system analysis, IEEE Trans. Power Syst. 10 (May (2)) (1995) 671679. [31] W.-M. Lin, Y.-S. Su, H.-C. Chin, J.-H. Teng, Three-phase unbalanced distribution power ow solutions with minimum data preparation, IEEE Trans. Power Syst. 14 (August (3)) (1999) 11781183. [32] Y. Zhu, K. Tomsovic, Adaptive power ow method for distribution systems with dispersed generation, IEEE Trans. Power Deliv. 17 (July (3)) (2002) 822827. [33] A. Augugliaro, L. Dusonchet, M.G. Ippolito, S. Mangione, E.R. Sanseverino, A modied backward/forward method for fast solving radial distribution networks, in: IEEE Power Tech Conference, Bologna, Italy, June 2326, 2003. [34] R. Ciric, A. Padilha-Feltrin, L. Ochoa, Power ow in four-wire distribution networks general approach, IEEE Trans. Power Syst. 18 (November (4)) (2003) 12831290.

1340

L. Degroote et al. / Electric Power Systems Research 80 (2010) 13321340 [41] L. Degroote, L. Vandevelde, B. Renders, J. Gyselinck, Nonlinear transformer model in the frequency domain and with symmetrical components, Int. J. Comput. Math. Electr. Electron. Eng. (Compel) 27 (6) (2008) 14181437. [42] A. Zobaa, The optimal passive lters to minimize voltage harmonic distortion at a load bus, IEEE Trans. Power Deliv. 20 (April (2)) (2005) 15921597. [43] R. Carbone, M. Fantauzzi, F. Gagliardi, A. Testa, Some considerations on the iterative harmonic analysis convergence, IEEE Trans. Power Deliv. 8 (April (2)) (1993) 487495. [44] A. Mansoor, W. Grady, P. Staats, R. Thallam, M. Doyle, M. Samotyj, Predicting the net harmonic current produced by large numbers of distributed single-phase computer loads, IEEE Trans. Power Deliv. 10 (October (4)) (1995) 20012006. [45] J. Schlabbach, A. Gross, Harmonic current emission of photovoltaic inverters, in: 19th Int. Conf. on Electricity Distribution (CIRED 2007), Vienna, Austria, May 2124, 2007.

[35] J.-H. Teng, C.-Y. Chang, Backward/forward sweep-based harmonic analysis method for distribution systems, IEEE Trans. Power Deliv. 22 (July (3)) (2007) 16651672. [36] J. Peralta, F. de Len, J. Mahseredjian, Unbalanced multiphase load-ow using a positive-sequence load-ow program, IEEE Trans. Power Syst. 23 (May (2)) (2008) 469476. [37] J. Nanda, M. Kothari, M.S. Srinivas, On some aspects of distribution load ow, in: IEEE Region 10 International Conference on Global Connectivity in Energy, Computer, Communication and Control, vol. 2, December 1719, 1998, pp. 510513. [38] J. Nanda, M.S. Srinivas, M. Sharma, S.S. Dey, L.L. Lai, New ndings on radial distribution system load ow algorithms, in: IEEE PES Winter Meeting, vol. 2, 2000, pp. 11571161. [39] Radial test feeders. IEEE Distribution System Analysis Subcommittee. Available online: http://ewh.ieee.org/soc/pes/dsacom/testfeeders.html. [40] S.-J. Jeon, Non-sinusoidal power theory in a power system having transmission lines with frequency-dependent resistances, IET Gener. Transm. Distrib. 1 (March (2)) (2007) 331339.