Dispersed generation impact on distribution network losses

Electric Power Systems Research 97 (2013) 10–18

Contents lists available at SciVerse ScienceDirect

Electric Power Systems Research journal homepage: www.elsevier.com/locate/epsr

Dispersed generation impact on distribution network losses Maurizio Delfanti, Davide Falabretti ∗ , Marco Merlo Politecnico di Milano, Dipartimento di Energia, via La Masa, 34, Milano, Italy

a r t i c l e

i n f o

Article history: Received 30 October 2011 Received in revised form 14 November 2012 Accepted 28 November 2012 Available online 11 January 2013 Keywords: Dispersed generation Distribution network Energy losses Monte Carlo algorithm

a b s t r a c t Distribution systems are subject to increasing penetration of dispersed generation. The energy injections of the generators impact on many technical issues, including energy losses occurring in the grid. Given the technical and economic importance of losses, several statistical approaches are proposed in the literature to evaluate the effects of dispersed generation on the energy lost in distribution networks. In principle, these approaches require a great number of load flow calculations. This paper provides a novel index aiming to avoid complex and computationally expensive statistical analysis for loss assessment. Such an index does not require any detailed characterisation of the energy flows over the network (load flow calculations). The performance of the index is tested using a model of a real medium voltage network in a wide set of generation scenarios. Each scenario is defined by a Monte Carlo algorithm that was developed ad hoc for this study. Moreover, advanced convergence criteria are provided to optimise the number of scenarios to process. © 2012 Elsevier B.V. All rights reserved.

1. Introduction Climate issues, energy saving requirements and general energy needs are the impetus driving national governments to offer incentives to the large-scale deployment of Renewable Energy Sources (RES). This energy policy has significantly increased the amount of dispersed generation (DG) over the past decade and has resulted in several benefits, such as a reduction of the greenhouse gas emissions. However, DG diffusion has important consequences for the operation of electrical networks, which must be properly addressed in order to avoid deteriorating power quality, reliability and supply efficiency [1]. One of the previously mentioned consequences is the increase of network losses in the presence of a strong DG penetration. The relationship between DG power injections and network losses is difficult to assess because of the wide set of parameters that affect energy flows on the grid, such as the DG location, the generators injection profiles and load consumption profiles. Many statistical approaches are proposed in the literature to evaluate DG effects on network losses, and most of them aim to define the optimal DG location and sizing [2,3]. Because of the existing regulations on network connection (in Italy [4,5]), the Distribution System Operator (DSO) must accept all the requests for DG connections forwarded by the users. In this scenario, the DG siting, sizing and technology are not dependent

∗ Corresponding author. Tel.: +39 02 2399 4106; fax: +39 02 2399 8566. E-mail addresses: [email protected] (M. Delfanti), [email protected] (D. Falabretti), [email protected] (M. Merlo). 0378-7796/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.epsr.2012.11.018

on the DSO, and consequently, the approaches usually proposed in the literature (i.e., optimisations) [6,7] are not applicable. However, the knowledge of DG impact on losses is cardinal in order to select the structural improvements to implement in distribution grids (planning purposes). Additionally, the losses evaluation on wide networks databases (very critical because of the number of load flow procedures to perform) is essential to estimate effectively the average losses behaviour on distribution systems at a national level (standard coefficients for economical losses repayment; general regulatory purposes) [8]. In the literature, one of the critical issues is the simulation of a multi-generator scenario (i.e., the DG siting on the network); accordingly, simplified hypothesis are typically adopted (e.g. DG location and network structure not investigated [9], only one generation technology taken into account [10], only one feeder considered [11]). Because of the poor performance of the classical indices for the loss assessment (DG rated power, DG energy injections, reverse power flow time), in the present paper, a novel index is proposed that does not require a detailed characterisation of the energy flows over the network (i.e., it does not require the knowledge of the injection/load profile of each user for at least one year) or load flow calculations (i.e., the computational effort can be managed with a simple spread sheet). Consequently, this index is able to avoid complex and computationally expensive statistical approaches to DSOs when evaluating the impact of DG on network losses. The performances of this indicator are tested with respect to the actual losses occurring on a real distribution grid (load flow calculations). The analysis involved over 24,000 network scenarios generated by a Monte Carlo (MC) approach. Each scenario was

M. Delfanti et al. / Electric Power Systems Research 97 (2013) 10–18

characterised by its unique DG distribution where the losses are assessed by load flow calculations over a whole year. The MC algorithm is developed to represent properly the randomness of DG spreading in distribution systems. Each of the variables characterising the DG is defined by appropriate probabilistic distributions and assessed consistently with the Italian national framework. The generation scenarios defined through the MC algorithm are used to test the performance of some indices in the loss assessment. The paper is organised as follows: first, the traditional and novel indices for the loss estimation are presented; then, the MC approach utilised to evaluate the indices performances through comparison is depicted; it is used to build the database of operating conditions for the network under study; with this database, the impact of DG on network losses is evaluated and the performances of the losses estimation indices are investigated; finally conclusions are provided.

2.1. The novel indicator proposed Although the indices reported in the literature and in the national resolutions typically provide important information about the trend of losses as a function of DG penetration, none of these indices is able to estimate effectively the losses occurring in a distribution system in the presence of DG (these indices present an excessive dispersion in cases with large DG amounts, see Section 6). This limitation is mainly attributable to the inability of the selected losses indicators to consider the DG localisation within the network, which is a parameter that affects the path followed by the energy produced by power plants and, consequently, the losses occurring in the various branches of the network. To obviate this lack of information and to provide an improved assessment of the energy flows in the grid (taking into account the compensation between load and DG), the following novel loss indicator is proposed: ◦

2. The losses indicators

El = E0tr +

n Br  i=1

The assessment of DG impact on network losses is a particularly critical issue. In fact, the accurate evaluation of such an impact is typically difficult, because the analysis usually requires to take into account the grid topology and the time dependency of energy flows. The complexity of the analysis limits its applicability to a small number of networks and DG configurations: this makes the losses evaluation for planning/regulatory purposes impossible (owing to the computational effort required for a generalised application). In this situation, because of the high number of scenarios to assess, the only way to estimate losses is to use simplified methods, based on the observation of losses indicators. The simplest losses indicator which can be used for this purpose is the total rated power of the DG plants connected to the distribution network (in absolute value and in percentage w.r.t. the peak value of the exchanged profiles at the HV/MV interface in the passive scenario). An alternative index is the amount of energy produced by DG units over a whole year, which is easy to evaluate almost as the DG overall rated power. On the one hand, such index gives more information about the energy losses; on the other hand, it needs a proper characterisation of the various generation technologies with the relevant annual production. In the past few years, the need for indices able to assess DG impact on the Quality of Service in distribution networks has risen. The goal is to identify the DG threshold beyond which the network control, protection and automation systems require updating (introducing the solutions proposed by the so-called smart grids paradigm [12]). Recently, to this purpose, some Italian regulations assumed the yearly hours of RPF (reverse power flow time, RPFT) to be a suitable indicator. The national standard CEI 0-16 [13] states that, when the RPFT at the HV/MV interface is greater than 5% over the year, the protection and control devices equipping the primary substation have to be revised. In fact, it is necessary to adopt protection and control systems that are able to operate with an “active” grid. Similarly, Resolution ARG/elt 39/10 [14], which provides incentives for the implementation of smart grid projects in the national distribution grid, establishes that smart grid developments are incentivised only if the network feeders present an RPFT at their point of connection to the primary substation, equal to, or greater than, 1%. The relevance of the RPFT index in the Italian regulation for DG impact assessment motivates an interest in its performance also in losses estimation.

11





◦

ri ·

n Ty 

2

(kc · Phours,k,i ) (hk − hk−1 )

(1)

k=1

In Eq. (1): • El represents the energy losses estimated by the proposed indicator [p . u .]; • E0tr is the total yearly no-load losses of the HV/MV transformer [p . u .]; • ri is the series resistance of each branch of the network (n◦ Br is the overall number of branches) [p . u .]; • kc is a simultaneity coefficient, between 0.8 and 1; • hk and hk−1 are the equivalent yearly hours of production at the maximum power, referring to the kth and k − 1th technology of generation or load profile (h0 = 0, technologies ordered according to increasing heq , n◦ Ty overall number of DG and load typologies). The element Phours,k is the difference between the power injections and withdrawals of the active and passive users downstream of the ith branch of the network, for the kth equivalent time interval [p . u .]. It is equal to: n◦ DG

Phours,k,i =

 n=1

◦

PDG,n −

n Lo 

Pload,n

(2)

n=1

where • PDG,n is the rated power of each DG plant downstream of the ith branch of the network with heq equal to, or greater than hk (n◦ DG is the total number of power plants downstream of the ith branch with a number of equivalent hours equal to, or greater than hk ) [p . u .]; • Pload,n is the conventional power of each supply point downstream of the ith branch of the network with heq equal to, or greater than hk (n◦ Lo is the total number of supply points downstream of the ith branch with heq equal to, or greater than hk ) [p . u .]. Eq. (1) is equivalent to applying a cumulative exchange profile to each power plant and load in the network. As additional simplifying assumption, the exchanges are supposed concentrated in the peak hours. These hypotheses are motivated by the following two aspects: • the lack of knowledge usually concerning the actual DG/load injection/withdrawal profiles during the year;

12

M. Delfanti et al. / Electric Power Systems Research 97 (2013) 10–18

Table 1 DG technologies probabilistic distribution in terms of number of power plants [%]. PV

Run-of-the-river hydroelectric

CHP

96.48

2.42

1.09

• the need to avoid complex hourly analysis necessary to take into account the dependence on the time of the energy flows over the network. It is important to point out, however, that the proposed method can be applied, with only small changes, also to more detailed cumulative profiles: in this case the only difference is in the number of time intervals to be evaluated. The index proposed estimates the energy flows over the grid evaluating the energy injections/absorptions in the underlying buses, for each branch: the flows are determined as the algebraic difference of DG injections and load withdrawals. In order to apply the proposed method, a conventional value for load is needed (Pload in Eq. (2)). This value is determined as the number which, applying Eqs. (1) and (2) with no DG injection (PDG equal to zero), allows one to obtain the same energy losses actually occurring on the grid in the passive scenario. If the distribution system is equipped with advanced metering systems, as in the Italian framework [15], the DSO can evaluate the network losses without DG directly as the difference between the energy supplied to the users (or measured at the MV/LV substations) and the energy exchanged at the HV/MV interface. Additionally, network losses without DG can be assessed by using mathematical models of the network (since the analysis is circumscribed to only one scenario) or by the appropriate estimation coefficients (i.e., the conventional losses given by the Italian regulation [16]). In Table 2, Eqs. (1) and (2) are applied to the example network of Fig. 1. For each branch of the network (ri ) and for each time interval hk − hk−1 , the relevant DG power plants and loads are reported. 3. The probabilistic Monte Carlo approach In the following section, a method to assess the electrical losses affecting an MV distribution grid in the presence of DG is described. This method will be adopted as a reference metric in order to quantify the reliability of the index proposed. Using an MC approach [17], the algorithm defines a large number of DG scenarios (typically over 2000) where the losses are assessed. DG configurations, similar to real life scenarios, are defined on the basis of the actual DG spreading that can be observed in Italian distribution systems. The MC procedure assigns a size, technology and location on the grid to each power plant. Appropriate generation and load profiles are applied to the DG units and to final users. Finally, classical load flow procedures are implemented in order to calculate the electrical losses occurring in a year (the analyses are detailed as often as every hour).

Fig. 1. Example network.

Fig. 2. Flow chart of the proposed MC algorithm.

The flow chart of the adopted MC method is presented in Fig. 2. Through a stochastic roulette-wheel, one-by-one, each generator is initially defined (B) and characterised based on its technology (C). The technologies determined to be numerically relevant (according to their actual and prospective diffusion on Italian Medium Voltage networks) are the photovoltaic, the run-of-the-river hydroelectric and the CHP. By resorting to the Italian national generation shared among the various production technologies [18,19], and removing the technologies not suitable for the specific network under study (because of geographical location and population density of the electric network analysed, see Section 4), the percentages reported in Table 1 are achieved. According to the production technology selected by the first roulette-wheel, a second extraction process defines the DG power category of the plant (D). Each type of DG has a specific probability distribution; these are represented in the MC algorithm by a set of classes with different probabilities of being selected. The PV distribution is composed of 32 different categories whereas

M. Delfanti et al. / Electric Power Systems Research 97 (2013) 10–18

13

Table 2 DG power plants and loads taken into account in the evaluation of Pload in Eqs. (1) and (2). The results refer to the example network of Fig. 1. Branch (ri )

h1 = 1500

h2 = 3000

h3 = 5000

r1 r2 r3 r4

LA , GDA LB LA , GDA , LB , GDC LA , GDA , LB , GDC , LD

LA , GDA LB LA , GDA , LB LA , GDA , LB , LD

LA LB LA , LB LA , LB , LD

hydroelectric and CHP distributions are composed of 12 and 4 different categories, respectively. After the selection of the class in which the DG unit power falls, the effective size of the plant is randomly determined within the limits of the power category previously selected (E). The presence of DG with a size greater than 3 MW along MV feeders is not considered in the light of Italian regulations. Power plants above this size are directly connected to the primary substation MV busbars [13]. Consequently, their power injections would affect only the HV/MV transformer losses (the penetration of such units is quite low, about 0.5% of the total number of dispersed generators). Finally, a last MC roulette-wheel defines the DG connection point to the grid. According to the Italian standards [13], the power plants with a rated power equal to, or greater than, 100 kW can be connected only at the MV level (G). Plants with a power lower than 100 kW can be connected to both MV and LV buses (H). The DG extraction process within a given scenario continues until the sum of the generators power reaches the desired value (I). Consequently, all DG configurations share the same total DG rated power but not necessarily the same number of generators (the number of generators typically varies from a few units, up to 200–300 units). The scenario creation ends when a specific convergence criterion is satisfied (L). 3.1. The MC convergence criterion The convergence criterion used in the present study is determined by checking the value assumed by an index that represents the degree of instability of the probability distribution of the losses resulting from the MC analysis. If the index, in the last 100 DG configurations, exists and is lower than 1%, then the MC process is assumed to have reached convergence. At the nth MC scenario, the convergence index is determined as described below. • The losses evaluated in all the DG configurations, up to the nth (included), are divided into 20 losses categories. They are obtained by subdividing, into 20 equal intervals, the difference between the maximum and minimum losses observed in all of the MC scenarios. • The number of samples falling in the kth losses category (sk,n ), reported in percentage w.r.t. the total number of DG scenarios (n), is compared to the same quantity assessed at the scenarios from the n − 1th to the n − 500th. An index of the stability from the results of the kth losses category is obtained:

  sk,n ek,n = max  n

  sk,n−1   sk,n , − n−1  n

  sk,n−2   sk,n ,..., −  n n−2

 sk,n−500  − n − 500 

(3)

The loss categories change during the MC process because of their dependence on the maximum and minimum loss values evaluated on all the previously generated scenarios. The comparison among the number of samples falling in a given class, reported in (3), is significant only if the categories remain unchanged for the last 500 scenarios (in the first 500 scenarios the convergence index is not evaluated). After the identification of a DG working condition with an energy losses value lower or greater than all

Fig. 3. Example of trend of the MC convergence index (the characteristic reported refers to the simulation, detailed in the following, with an overall DG rated power equal to 6 MW).

the previous DG configurations, the MC convergence index does not exist until 500 new scenarios are generated. • Therefore, for each category, the maximum variation assessed among the ek,n of the last 100 scenarios is selected:

vk,n = max{ek,n , ek,n−1 , . . . , ek,n−100 }

(4)

• Finally, the MC convergence index at the nth scenario is defined as the maximum variation shown by the vk,n on all the 20 losses categories: MCindexn = max{v1,n , v2,n , . . . , v20,n }

(5)

The use of a complex convergence criterion is justified by the performances required. This criterion makes it possible to stop the MC process when the results of the analysis reach the desired accuracy while not overestimating the number of scenarios to process.1 This ultimately results in an optimisation of the required computational time. An example of the MC convergence index of Eq. (5) is reported in Fig. 3: between the first scenario generated by the MC algorithm and the configuration no. 850, the index is not evaluated because of one or more changes of losses categories (i.e. an MC scenario with losses lower, or greater, than all the previous scenarios is created, so for the following 500 configurations the convergence indicator is not defined). Between scenarios nos. 851 and 2204, the index is assessed, but it is greater than the admitted tolerance (1%). At scenario no. 2204, there is again a change of losses classes, so, until the configuration no. 2704 the index is not evaluated. Between scenarios nos. 2705 and 2794, the MC convergence index is assessed, but it results greater than 1%. Finally, after scenario no. 2795 it shows a variability lower than the target value, so, after 100 more scenarios in which the index stability is verified, the MC convergence criterion is satisfied (scenario no. 2894). 4. Network model adopted To provide accurate evaluations, the losses analyses have been conducted on the model of a real distribution network (rated voltage 20 kV). This network is located in central Italy, is operated by A.S.SE.M. S.p.A. [20] and covers an urban area. The grid has a radial structure, starting from an HV/MV primary substation equipped with two 10 MVA transformers; the MV busbars underlying the transformers operate separately. The paper will refer only to the part of the network departing from one MV primary busbar, with 5

1 This possibility results from the use of the index to determine the error affecting the loss estimation.

14

M. Delfanti et al. / Electric Power Systems Research 97 (2013) 10–18

Fig. 5. Yearly profile of MV/LV substations (in p.u. w.r.t. the substation rated power).

yearly characteristic of the overall national load, suitably shared between the two user classes. Each MV user is assigned a profile with the same shape of the overall national MV characteristic (Fig. 6). Similarly, each secondary substation is attributed the same load trend of the LV system national profile (Fig. 5). The peak-values of the MV and LV energy supply characteristics of each user/substation are established by assuming that the energy absorbed yearly, by all the loads of the grid under study (evaluated by the conventional characteristics adopted), has to be equal to the energy measured at the network HV/MV interface: kp = Fig. 4. The MV network utilised for the analysis.

feeders underlying. The exchange profile with the HV network has a peak load value of 4.69 MW and 2.27 Mvar. The MV network under analysis is composed of 287 busbars: • • • • •

a bus representing the HV primary substation busbars; a bus representing the MV primary substation busbars; 10 buses to which MV users are connected; 109 buses underlying MV/LV substations; 166 transit buses (required to represent feeders discontinuities, e.g. transitions from cable to overhead lines).

The grid model is implemented in Matlab and utilises function libraries derived from the MATPOWER project [21]. Because of the complexity and spreading of LV networks, only the MV level is modelled in detail (Fig. 4 and Table 3). The LV loads are introduced in the model as equivalent power exchanges at the MV/LV interface. To estimate accurately the losses that occur in the distribution grid, a detailed energetic model is required. Thus, all the energy flows in the grid are represented on an hourly basis and over a whole year (8760 h). The loads are classified into two different categories: MV users and MV/LV substations. All users belonging to the same category are assigned the same load profile. These profiles are obtained from the

EMV ·

MVusers i

EPS Ai + ELV ·

Substations j

(6) Bj

where • kp is the load scale coefficient; • EPS is the energy measured yearly at the network HV/MV interface [MWh/year]; • EMV is the energy absorbed yearly by an MV user with a contractual power equal to 1 MW, according to the profile obtained from the national characteristic [MWh/(year MW)]; • ELV is the energy absorbed yearly by an MV/LV substation with a size of the MV/LV transformer equal to 1 MW, according to the profile obtained from the national characteristic [MWh/(year MW)]; • Ai is the contractual power of the ith MV user of the network [MW]; • Bj is the size of the MV/LV transformer of the jth secondary substation of the network [MW].

Table 3 Type and length of the lines of the network under analysis (Fig. 4; lines numbered from right to left). Feeder number

Cable (km)

Overhead line (km)

1 2 3 4 5

3.73 13.03 3.91 10.51 0.13

19.35 28.60 15.12 25.67 0

Fig. 6. Yearly profile of MV users (in p.u. w.r.t. the user contractual power).

M. Delfanti et al. / Electric Power Systems Research 97 (2013) 10–18

15

Consequently, the coefficient (kp ) is the multiplying factor that, when applied both to the HV and LV profiles, ensures the energetic equivalence between the actual energy flows and their estimation: ai = kp · Ai bj = kp · Bj

(7)

where • kp is the load scale coefficient; • ai is the power of the ith MV user used in the network model [MW]; • bj is the power of the jth secondary substation used in the network model [MW]; • Ai is the contractual power of the ith MV user of the network [MW]; • Bj is the size of the MV/LV transformer of the jth secondary substation of the network [MW]. All loads are modeled as P,Q buses; according to the Italian regulation [16], the power factor is equal to 0.9 lagging. Similarly to the load, each technology of DG is characterised by a specific injection profile. Even if recent Grid codes, in Italy and in a few other EU systems, require DG power plants to manage reactive power in the case of overvoltage on the distribution grid [22], in our simulations all power plants are assumed to inject energy with a unitary power factor. This choice is consistent with operational practises: in fact, reactive power flows increase the losses level significantly; consequently, during operation, such reactive injections will be limited to those situations with severe overvoltage problems and to a small number of hours over a year. The exchange profile of the PV generators is obtained from data of solar irradiance registered by sensing satellites in that specific location [23], on an hourly basis for five years (1995–1999) [24]. The power injections of run-of-the-river hydroelectric power plants have been defined as the monthly average production of the hydroelectric units already connected to the network under analysis. Finally, since CHP injections vary greatly as a function of the considered power plant, a conventional characteristic is used for this DG technology. The CHP electrical injection is usually motivated by the heating demand. Thus, assuming for simplicity only industrial CHP generators, the power injections of this type of DG are strictly related to the trend of the industrial activities, concentrated in peak hours (CHP production at the rated power) and reduced in middle/off peak bands (80% of the rated power in the middle peak band and null in the other hours). All these assumptions are consistent with data publicly available for the Italian electricity framework.

Fig. 7. Loss distribution for the scenarios with an overall DG power equal to, or lower than, 4 MW (loss variation reported in percentage w.r.t. the network losses without DG).

number of load flow calculations required for the analysis is about 270 million. In the literature [25], the relationship between the losses and DG is commonly cited as being a U-shape, because a limited amount of DG usually causes a reduction in the network losses, but increasing DG causes reverse power flows (RPFs) on the feeder, from the final users up to the primary substation, and network losses progressively increase as a consequence. The MC approach allows a correct evaluation on the impact of different DG penetration levels on network losses. In particular, in Figs. 7 and 8, the variation of the yearly losses of the network w.r.t. the losses occurring without DG (passive scenario) is reported. The results concerning the scenarios with an overall DG power equal to, or lower than 4 MW are depicted in Fig. 7, whereas the results of the scenarios with greater DG amounts are presented in Fig. 8. For low DG penetration, a loss reduction is nearly always achieved. When DG is equal to 0.5 MW, there is a loss reduction in 100% of the cases. With 4 MW (i.e., approximately 85% of the peak value of the HV/MV exchange profile), the percentage drops to 85%. The percentage remains quite high even in the presence of a strong DG penetration (Fig. 8). With an overall DG amount equal to 10 MW, there is a loss reduction in 57.5% of the cases. Moreover, the graph of Fig. 7 highlights that increasing DG causes a greater variability in the resulting annual losses. With a power equal to 0.5 MW, the loss variation is always limited between −15% and 0%, but with a power of 4 MW, the losses change between −50% and +20% (with a mean value still negative).

5. Preliminary analysis: assessment of the network losses The electrical losses occurring on the MV network (including HV/MV transformers) in all the DG configurations generated by the MC algorithm are assessed. Because of the hourly basis of the energy profiles, the evaluation for each scenario requires 8760 load flow calculations (one for each hour of the year). The analysis involves generation conditions with different DG amounts. In particular, sets of scenarios with an overall DG rated power equal to 0.5, 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10 MW are studied (the exploration is limited to 10 MW, consistently with the rated power of the HV/MV transformer). Each set includes a different number of cases (approximately between 1900 and 3700) according to the aforementioned convergence criterion applied to the MC process. Considering the number of scenarios for each set (1900–3700), the number of sets investigated (11) and the time resolution of the exchange profiles (one hour, i.e., 8760 values in a year), the

Fig. 8. Loss distribution for the scenarios with an overall DG power equal to, or greater than, 5 MW (loss variation reported in percentage w.r.t. the network losses without DG).

16

M. Delfanti et al. / Electric Power Systems Research 97 (2013) 10–18

Fig. 9. Variation of the network losses w.r.t. the network losses without DG, according to the overall DG power connected to the grid (DG power in absolute value and in percentage w.r.t. the peak value of the exchanged profiles at the HV/MV interface in the passive scenario).

The variability of the results is more evident in the presence of high values of DG overall power (Fig. 8). In particular, in a limited number of cases (less than 1%), the increase in losses due to the DG is considerable (greater than 200%). The results confirm (and quantify) that the connection to the grid of small amounts of DG typically causes a reduction in losses. If DG penetration increases, losses also increase. If it increases significantly, losses can be even higher than without DG. However, for this general behaviour many exceptions could be identified due to DG size, technology, connection point, and so forth. Moreover, to evaluate the correlation between DG penetration and losses correctly, a chronological analysis (based on the actual hourly load and generation profiles) must be performed, which would result in a computational effort that is not compatible with DSOs needs, and with regulatory requirements. Thus, a more viable approach has to be identified.

Fig. 10. Loss variation w.r.t. the scenario without DG, according to the DG energy produced yearly (DG energy in absolute value and in percentage w.r.t. the energy yearly exchanged at the HV/MV interface in the passive scenario).

connected to the network increases (with a DG overall rated power equal to 10 MW, on average, the loss reduction is about −8.5%). Moreover, the spread of the samples for each DG penetration level is greater for the situations with a large generation. This fact is imputable to the presence, in these scenarios, of larger power plants. Considering the losses strictly determined by the DG connection point to the grid, the cases where the generation is connected far from the primary substation have greater losses. Additionally, the variation in losses, in the best scenarios (i.e., the DG configurations which cause the maximum losses reduction), does not improve as DG power increases. In fact, there are no DG configurations that are markedly better in reducing the amount of energy lost (the losses reduction is concentrated at about −50%). On the other hand, for higher DG penetration levels, there are few scenarios that present a substantial deterioration of losses (with values greater than +250% of the initial one). 6.2. Network losses & DG annual production

6. Indices for the loss estimation: numerical analysis In the following section, the performance of losses indicators in the assessment of energy lost in the presence of different DG penetration levels is evaluated. As already mentioned, the indicators taken into account include the DG overall rated power, the DG annual production, the RPFT measured at the HV/MV interface and, finally, the novel index proposed in this paper. All the indices have been evaluated in all scenarios obtained thanks to the MC procedure previously introduced. 6.1. Network losses & DG rated power In Fig. 9, the variation of network losses (w.r.t. the passive scenario) according to the overall DG power connected to the grid is reported. The red bars indicate the maximum and minimum values assessed over the database of MC scenarios. The black bars show the 5th and 95th percentiles, as well as the losses mean values. Despite the simplicity of the overall DG power as an indicator, it proves to be effective in the loss assessment. In particular, it yields important indications about their average value. For all the DG scenarios, the mean value of the losses (the larger tick on black bars) is lower than the energy lost in the case without DG (in the best case, the scenarios with 6 MW, i.e. 128% of the exchange profiles peak value, the losses reduction is equal to −23.5%); however, for high values of DG penetration, the benefits decrease as the generation

In Fig. 10, the loss variations w.r.t. the scenario without generation are reported according to the overall yearly DG energy production (DG production reported also in percentage w.r.t. the overall energy exchanged at the HV/MV interface in the passive scenario). Each colour indicates a different amount of DG connected to the grid (for example, the red colour refers to the set of scenarios with a generation equal to 8 MW). The graph of Fig. 10 highlights the correlation of this index with the losses. The results in this graph are better than the results obtained by the DG rated power. This fact was anticipated: in fact, DG rated power is unable to take into account the impact of the various DG technologies on system efficiency. For example, a 1 MW run-of-the-river hydroelectric power plant (5350 equivalent yearly production hours at the maximum power, heq ) has a larger effect on the energy flows over the grid when compared with a PV or CHP generator with the same rated power (with, respectively, 1650 heq and 3450 heq ). The use of the overall DG energy production as a loss indicator allows the results previously observed to be confirmed; that is, when the DG penetration increases, the energy lost in the grid usually reduces. Contrary to DG rated power, the DG annual production provides a better characterisation of each scenario in terms of impact on energy losses. In particular, it highlights that for production values greater than about 33 GWh/year (i.e., approximately 160% of the energy yearly exchanged at the HV/MV interface in the

M. Delfanti et al. / Electric Power Systems Research 97 (2013) 10–18

Fig. 11. Loss variation w.r.t. the network losses without DG according to the RPFT at the HV/MV interface.

17

Fig. 12. Correlation between the actual losses on the network (w.r.t. the energy lost in the scenario without DG) and the losses estimated by the proposed indicator.

passive network scenario), there is usually a loss deterioration (for the previous indicator, all the DG penetration levels gave on average a loss reduction). Two additional aspects can be observed in Fig. 10. First, for high DG production levels (more than 40 GWh/year), the number of samples is small, so the mean value is not reliable (this is the cause of its apparent variability). Second, for large amounts of DG, the dispersion of the results is considerable. This finding proves, in many cases, that the correlation between DG annual production and the losses occurring in the grid is poor. 6.3. Network losses & RPFT The correlation between the indicator adopted by the standard CEI 0-16 (RPFT at the HV/MV interface) and the variation of electrical losses (w.r.t. the case without DG) is investigated.2 In Fig. 11, the correlation between the two quantities is presented. Despite the index adopted by the Italian regulation provides an effective estimation of the DG consequences on the network control, protection and automation systems, it is not able to evaluate the DG impact on the losses occurring in the grid. The correlation of the RPFT with the energy lost is weaker than in the previous cases. In fact, the samples show a considerable degree of dispersion, also for low DG penetration.

Fig. 13. PDF of the error introduced in the loss estimation by the proposed indicator.

The performances of the proposed indicator are highlighted in Fig. 13, where the loss estimation error is represented. For each scenario, the error (errn ), reported in percentage w.r.t. the losses without DG, is calculated by the following formula: err n =

6.4. Network losses & the novel indicator proposed

El,n − Ea,n E0

(8)

where In Fig. 12, the proposed index is used to estimate the energy losses occurring in the DG scenarios defined by the MC technique on the network model adopted (Fig. 4). The DSO is assumed to be capable of assessing the losses in the passive scenario by direct measurement. The graph in Fig. 12 shows a suitable correlation between the two quantities (actual losses and estimated ones). An effective loss assessment is performed in all scenarios, with both a high or a low DG penetration (there are no distortions or polarisations of the characteristic). For configurations with a low DG penetration (up to approximately 2 MW, i.e., about 43% of the peak load), the results follow the ideal characteristic (the black line in Fig. 12), which is subjected to a very limited dispersion. Therefore, the error in the loss assessment for these scenarios is almost negligible. At increasing levels of the DG power connected to the grid, the error slightly rises but remains at acceptable levels.

2 The indicator proposed by Resolution ARG/elt 39/10 is not taken into account, as it refers to single feeders of the network.

• El,n represents the energy losses in the nth scenario, as determined by the proposed indicator; • Ea,n is the actual value of energy losses in the nth scenario (evaluated by load flow calculations); • E0 is the energy lost in the passive scenario. Approximately 65% of situations depict a gap between the effective energy lost in the grid and the estimated losses lower than 5% (w.r.t. the losses in the passive scenario). The percentage of situations rises significantly, reaching 86%, if a tolerance equal to 10% is accepted between the actual losses in the grid and the estimated losses. Finally, in almost all the scenarios (95%), the error is lower than 16.9%. 7. Conclusion In this paper, a new index has been proposed that does not require complex and computationally expensive approaches in order to assess properly the energy losses occurring in distribution networks with DG. Using the proposed method, the DSO could

18

M. Delfanti et al. / Electric Power Systems Research 97 (2013) 10–18

detect the DG configurations that are harmful for system efficiency and plan the network expansion accordingly. Moreover, the evaluation of the DG impact on losses on wide sets of networks (with acceptable computational efforts) could be effectively used to define the losses regulatory principles: through this algorithm it is possible, for example, to identify the losses behaviour on distribution system at a national level with DG and, consequently, to calculate the coefficients for the economical losses repayment. Despite its relative simplicity w.r.t. more complex mathematical models including detailed characterisations of the power flows over the network (on which load flow calculations have to be performed), such an index has demonstrated good accuracy in the losses estimation. Considering the poor performances of the other indicators available for losses assessment (the DG rated power, the DG annual production and the RPFT at the HV/MV interface), the proposed index could be the preferred choice. The MC algorithm was exploited to generate a statistically relevant population of DG scenarios (over 24,000) that were used to compare the performances of the various losses indicators. The simple MC algorithm proposed allows losses to be evaluated without resorting to deterministic approaches. Despite their frequent use in the literature, these approaches are hardly applicable to such real life systems as the distribution grid used in this analysis. References [1] J. Pecas Lopes, N. Hatziargyriou, J. Mutale, P. Djapic, N. Jenkis, Integrating distributed generation into electric power systems: a review of drivers, challenges and opportunities, Electr. Power Syst. Res. 77 (2007) 1189–1203. [2] H. Iyer, S. Ray, R. Ramakumar, Assessment of distributed generation based on voltage profile improvement and line loss reduction, in: Transmission and Distribution Conference and Exhibition, 2005/2006 IEEE PES, Dallas, TX, pp. 1171–1176. [3] M. Kashem, A. Le, M. Negnevitsky, G. Ledwich, Distributed generation for minimization of power losses in distribution systems, in: Power Engineering Society General Meeting, 2006 IEEE, Montreal, Que. [4] Decree 79/99, Application of Directive 96/92/CE, containing common regulations for the internal market of the electricity, 1999. [5] Code for the terms and conditions for electricity grid connection with the obligation to connect third party electricity generation plants (Code for active connections – TICA), 2008. Available from: http://www.autorita.energia.it [6] T. Gözel, M. Hocaoglu, An analytical method for the sizing and siting of distributed generators in radial systems, Electric Power Systems Research 79 (2009) 912–918. [7] M. Moradi, M. Abedini, A combination of genetic algorithm and particle swarm optimization for optimal DG location and sizing in distribution systems, Electric Power Systems Research 34 (2012) 66–74.

[8] Approval of the Authority for Electricity and Gas code for the settlement of physical and monetary amounts of the dispatching services (settlement code) inclusive of the procedures for determining monetary amounts arising from adjustments to metering data (with amendments to Resolution No. 111/06), Resolution ARG/elt 107/09, 2009. Available from: http://www.autorita.energia.it [9] Y. Hegazy, G. Hashem, Probabilistic power loss analysis of electrical distributed generation systems, in: Conference on Electrical and Computer Engineering (IEEE CCECE’03), vol. 1, Montreal, Que., pp. 623–626. ˙ [10] J. Widén, E. Wäckelgard, J. Paatero, P. Lund, Impacts of distributed photovoltaics on network voltages: Stochastic simulations of three Swedish low-voltage distribution grids, Electric Power Systems Research 80 (2010) 1562–1571. [11] A. Marinopoulos, M. Alexiadis, P. Dokopoulos, Energy losses in a distribution line with distributed generation based on stochastic power flow, Electric Power Systems Research 81 (2011) 1986–1994. [12] ERGEG public consultation paper, Position Paper on Smart Grids, European Regulators’ Group for Electricity and Gas, ERGEG, 2009. [13] Reference technical rules for the connection of active and passive consumers to the HV and MV electrical networks of distribution Company, Standard CEI 0-16, 2008. Available from: http://www.autorita.energia.it [14] Procedures and criteria for the selection of investments admissible for incentives pursuant to paragraph 11.4, letter d) of Annex A to Authority for Electricity and Gas Resolution No. 348/07 of 29th December 2007, Resolution ARG/elt 39/10, 2010. Available from: http://www.autorita.energia.it [15] O. Bono, M. Cotti, P. Petroni, The new edge for the Enel Telegestore: an integrated solution for the remote management of electricity and gas distribution allowing a total management of the energy consumptions, in: 20th International Conference and Exhibition on Electricity Distribution (CIRED’09), Prague, Czech Republic, pp. 1–5. [16] Code of the Authority for Electricity and Gas provisions for the delivery of the electricity transmission, distribution and metering services for regulatory period 2008–2011 and provisions concerning the economic conditions for the delivery of the connection service, Resolution ARG/elt 348/07, 2007. Available from: http://www.autorita.energia.it [17] M. Marseguerra, E. Zio, Basics of the Monte Carlo Method with Application to System Reliability, LiLoLe-Verlag GmbH, Hagen, Germany, 2002. [18] Monitoring of the development of distributed generation plants in Italy for the year 2009 and analysis of the possible effects of distributed generation on the national electricity system, Resolution ARG/elt 223/10, 2010. Available from: http://www.autorita.energia.it [19] Atlasole, 2011. http://atlasole.gse.it/atlasole/ [20] A.S.SE.M. S.p.A., 2011. http://www.assemspa.it/ [21] R. Zimmerman, C. Murillo-Sánchez, R. Thomas, Matpower steady-state operations, planning and analysis tools for power systems research and education, IEEE Transactions on Power Systems 26 (2011) 12–19. [22] Reference technical rules for the connection of active and passive users to the LV electrical Utilities, Standard CEI 0-21, 2012. Available from: http://www.autorita.energia.it [23] A. Capone, M. Delfanti, D. Falabretti, L. Megalini, M. Merlo, PV production forecast for an effective VPP exploitation, in: Cigré 2011 Symposium Bologna, Bologna, Italy, pp. 1–7. [24] S@tel-Light, 2011. http://www.satel-light.com/ [25] V. Méndez Quezada, J. Rivier Abbad, T. Gómez San Román, Assessment of energy distribution losses for increasing penetration of distributed generation, IEEE Transactions on Power Systems 21 (2006) 533–540.