Accepted Manuscript

Evaluation of economic regulation in distribution systems with distributed generation

Yalin Huang, Lennart Söder

PII: S0360-5442(17)30377-8
DOI: 10.1016/j.energy.2017.03.019
Reference: EGY 10486

To appear in: Energy

Received Date: 13 October 2016

Revised Date: 28 February 2017
Accepted Date: 4 March 2017

Please cite this article as: Huang Y, Söder L, Evaluation of economic regulation in distribution systems
with distributed generation, Energy (2017), doi: 10.1016/j.energy.2017.03.019.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Evaluation of economic regulation in distribution systems with distributed

generation

Yalin Huanga,∗, Lennart Södera

a
Osquldas vägen 6, KTH Royal Institute of Technology, Stockholm, Sweden

PT
Abstract

RI
The economic regulation impact on distribution system investment is evaluated by a network expansion
planning model in this paper. Distributed generation (DG) integration has been taken into consideration

SC
in network investment worldwide. In most studies DG units are planned by distribution system operators
(DSOs). However, in some countries DSOs are not allowed to own generation due to unbundling regula-
tion. In the proposed model formulation, DG units are not owned by the DSOs. Moreover, fluctuation

U
from load and DG in the planning periods, DG curtailment possibility and regulation on losses and DG
AN
connection fees are altogether considered. Different regulation arrangements are studied in the same test-
ing network and the resulting network expansion costs are compared. The main value of this paper lies
in the application of network planning model to the economic regulation analysis, in the quantification
M
of the impact of different economic regulation frameworks, and in the implications of different regulation
choices concerning distributed generation integration.
D

Keywords: network planning, distributed generation, distributed generation integration, reinforcement,

TE

connection fee, economic regulation, curtailment, loss

1 1. Introduction
EP

2 Distribution system operators (DSOs) are natural monopolies; therefore, regulation is implemented to
3 direct their investments and operation decisions. This paper studies the impacts that different economic
C

4 regulation frameworks have on the DSOs’ cost for integrating distributed generation (DG).
AC

5 1.1. Motivation

6 More DG is expected to be connected to the distribution network. Studies have shown that DG
7 integration has the potential benefits of improving system reliability, efficiency [1], and deferring network
8 investment [2, 3]. These benefits have been shown to be significant when DG units are planned by
9 DSOs [4, 5, 6, 7]. However, the unbundling rules defined by European Directive [8] prohibit DSOs from

∗
Corresponding author
Email address: yalin.huang@ee.kth.se (Yalin Huang )

Preprint submitted to Journal of Energy March 6, 2017

 ACCEPTED MANUSCRIPT

10 owning generation plants. In order to ensure efficient and non-discriminatory network access, independent
11 management structures are put in place between the distribution system operators and the transmission
12 system operators, and any generation/supply companies have independent management structures. It
13 requires that the DSOs and TSOs be independent at least in terms of their legal forms, organizations
14 and decision making from other activities not relating to distribution and transmission respectively.

PT
15 Furthermore, the EU Electricity Directive Article 14/7 requires that DSOs should consider DG in-
16 tegration as an alternative to network expansion [8]. Much work has been found on network operation
17 considering DG integration [9, 10, 11]. However this paper focuses on the network expansion planning

RI
18 considering economic regulations that are related to DG integration. The main issue here is how the
19 expansion cost would change given the economic regulation. Thus, the evaluation is interesting for DG

SC
20 owners, DSOs and regulators.

21 1.2. Literature Review and Contributions

U
22 In the literature, three kinds of methods to study economic regulation in power system can be found.
23

24
AN
One is to use historical cases or existing practices. For example in Refs. [12, 13] innovative regulations
for DSOs considering DG integration are analysed using existing examples. A second one is to use
25 economic or statistical methods to compare different networks under the same or different regulations.
M
26 The need for new network regulation due to increasing DG is studied in Ref. [14] by DSO spreadsheet
27 business model. Network efficiencies are compared using economic measures in Ref. [15]. The third
D

28 kind of method is to use a reference model which represents the real network. The reference model
TE

29 provides a solution which fulfils all technical requirements and regulatory requirements at the minimum
30 cost, for example the reference network model developed in Ref. [16, 17, 18]. However, none of these
31 reference models considers uncertainties in load or DG and addresses DG ownership unbundling regulation
EP

32 or economic regulations. The impacts of DG curtailment and connection have been studied for active
33 network management schemes in Refs. [19, 20]. DG curtailment can reduce the reinforcement cost is
C

34 shown by a network planning model in Ref. [21]. The impact of regulatory support schemes on DG
expansion investment is studied in Ref. [22]. However, the regulation concerning losses, DG curtailment
AC

35

36 and connection fees have not been addressed all together in one network planning model nor uncertainties
37 from load and DG.
38 Literature on distribution network planning models is rich. Models developed before year 2000 were
39 reviewed in Ref. [23]. The shortcomings of these models are summarised in Ref. [23] as:(i) planning
40 problems not formulated from a practical point of view; (ii) ignored or inappropriately applied voltage
41 constraints; (iii) ignored or incorrectly quantified reliability; (iv) no budgetary constraints; (v) alternatives
42 on routes and (new and existing) conductor sizes not properly addressed; (vi) no true multi-stage approach

2
 ACCEPTED MANUSCRIPT

43 that guarantees global optimality.

44 With the improvement of optimisation solvers and the changes in distribution networks (e.g. more
45 DG), new formulations for distribution expansion planning have been developed, and most of the short-
46 comings have been overcome [24, 25]. Most models take voltage constraints and different connection
47 routes into account, for example models in Refs. [5, 6, 26, 27, 28, 29]. However, some of them only

PT
48 consider direct current (DC) power flow, and some of them only consider few scenarios of load and DG
49 or only the worse scenario. Reliability issues and budgetary constraints are also commonly considered in
50 the planning formulation, for example Ref. [6] and Ref. [26]. Some models consider DG as part of the en-

RI
51 ergy system design, therefore DG operation and installation are optimised together with energy storage,
52 boilers and heat production, for example Ref. [30, 31, 32]. Some models use a multi-stage approach, for

SC
53 example see Ref. [6] and Ref. [27]. The multi-stage approach has the advantage of modelling the decision
54 making under uncertainty compared to single stage. For example, with a two-stage model it is possible

U
55 to consider that the decisions of the investment are made at the first stage before the revealing of the
future generation and load. The uncertainties of load and DG then are represented by a large number
56

57
AN
of possible scenarios at the second stage. However, the DG integration and economic regulation are still
58 not properly addressed in the planning formulation. So far in network expansion planning literature, DG
M
59 is considered as an assisting tool which is controlled by the DSOs to meet the load growth. In reality
60 many DSOs cannot site DG units nor plan their capacity due to DG ownership unbundling regulation
D

61 [24, 33, 34].

62 In this paper a reformulation of the distribution network planning model in Ref. [27] is proposed
TE

63 to consider the uncertainties from load and DG instead of few scenarios of load and DG or only worse
64 scenario, and alternative current (AC) power flow. The network planning decisions are modelled as a
EP

65 two-stage stochastic programming problem with recourse. The model also encodes economic regulatory
66 issues, i.e. the regulation concerning losses, DG curtailment and connection fees. AC power flow is
highly non-linear. The non-linear problem with integers is difficult to solve due to its high computational
C

67

68 intensity [35]. Mixed integer linear programs (MIPs) are generally faster to solve. Therefore, the iterative
AC

69 linear current voltage alternative current optimal power flow (ILIV-ACOPF) algorithm, which linearises
70 the non-linear constraints in AC power flow, is adopted here to solve the problem [36, 37].
71 The contributions of this paper are:

72 1. The impact of some economic regulations is evaluated on a network planning model with DG
73 integration.

74 2. A new network planning model which considers load and DG uncertainties is formulated as a two-
75 stage stochastic programming with recourse.

3
 ACCEPTED MANUSCRIPT

76 3. DG ownership unbundling is properly addressed and economic regulatory on losses, DG curtailment

77 and DG connection fee are explicitly modelled.

78 2. Network planning formulation

79 The overview of the method is depicted in Fig.A.1. The relevant equations are indicated as well. The

PT
80 inputs for the proposed model are stochastic load and DG profiles for the whole planning horizon and
81 the regulatory requirements. The regulation requirements determines the formulation of the objective
82 function (1a) and some of the regulatory constraints (21,22). The optimisation problem is non-linear

RI
83 and is solved by equivalent linear subproblems iteratively. The linearised constraints are updated in each
84 subproblem. When the solutions from the subproblems are converged, a solution pool is applied in order

SC
85 to obtain a number of best solutions. Reliability indices are then calculated for all selected solutions.

86 2.1. Assumptions

U
87 In this paper DSOs are assumed to be under DG ownership unbundling regulation, which means that
88

89
AN
DSOs cannot own DG units but are obliged to connect them. This regulation is applied in many developed
countries, for example Sweden and UK. Furthermore, DSOs are assumed to minimize their total cost,
90 which contains the net present value (NPV) of investing new connection lines, replacing existing lines,
M
91 and operational costs. These operational costs may include losses and DG curtailment depending on
92 economic regulation. Cost minimisation assumption is the first step for a more sophisticated decision
D

93 making criterion. In this model it is assumed that the DSO has a good knowledge about the location and
TE

94 connection time of DG units beforehand. Given the fact that the DG owners need to apply for network
95 connection in advance, possible connection routes for these units are predefined in this model.

2.2. Decision Variables

EP

96

97 In order to account for the impact of load and DG uncertainties on DSOs’ expansion decisions, a
98 two-stage recourse approach with multiple scenarios in multiple planning periods is used. In this network
C

99 expansion planning model, the first stage decision variables are the choice of new DG connection routes,
AC

100 conductors of the connection lines, substations updates and reinforcements in the existing network, as
101 well as the optimal timing of these decisions are represented by integer variables [5, 27]. These decisions
102 are made at the first stage under uncertainties of load and DG. When the uncertainty is disclosed, the
103 second stage decisions are taken. The second stage decisions are the operational resources of the DSOs
104 in the real time. Regulating the output of DG is one of the resources [6]. For the DG that cannot
105 increase the production on demand, curtailing the production is the focus. In this model the second stage
106 decisions are the timing and amount of DG curtailment. There decisions depend on current in each line

4
 ACCEPTED MANUSCRIPT

107 and voltage on each node, which are determined when the load consumption and DG production are
108 realized.

109 2.3. Uncertainties and Scenarios

110 Uncertainties of load and DG are assumed to be probabilistic in order to consider the possibility
111 and probability of DG curtailment. These uncertainties are represented by a large number of possible

PT
112 scenarios at the second stage. Different ways to generate load and wind scenarios can be implemented
113 in this model, for example as in Refs [6, 38]. Scenarios of load and DG generated by random variables

RI
114 [39] are taken as an example. In this model the planning horizon consists of T planning periods (e.g. 3
115 periods, 1-4 years in each period). In each period, different load and DG levels are defined according to

SC
116 the historical patterns. Each level is modelled by random variables. Therefore, the fluctuations of load
117 and DG production are modelled by scenarios generated from these random variables. With the increase
118 of number of scenarios considered, the computation burden increases rapidly. Therefore, a scenario

U
119 reduction process is applied. A scenario tree can be constructed and reduced by the method described
120

121 2.4. Objective Function

AN
in [40] and [41]. The toolbox SCENRED in GAMS [42] is used to perform it.
M
122 The objective function in the planning model is the present value of capital expenses and operational
123 expenses of the DSO as shown in (1a-1f). The expenses include the expenses for the network reinforcement
D

124 and the DG connection. The objective function can differ according to the regulatory framework as
125 presented in Section 3.
TE

126 The capital cost in (1b) is determined at the beginning of the planning horizon and is given by the
127 investment of replacement of or updating the existing lines or substations and installation of new lines.
EP

128 The cost of operation in (1c) is considered at the beginning of each planning period and corresponds to
129 the average cost of losses and DG curtailment during this planning period. However, whether the cost of
130 losses and curtailment are included in the objective function depends on the economic regulation. The
C

131 generation cost is not considered as part of the DSO’s cost since DG is considered unbundled from the
AC

132 DSOs’ business.

5
 ACCEPTED MANUSCRIPT

Minimize
T
X
Ctotal = [δtcap Ctcap + δtoper Ctoper ] (1a)
t=1

Ctcap = (C R,AL 0
) EtR,AL + (C K,AL )0 EtK,AL + (C NS ,AL )0 EtNS ,AL (1b)
Ctoper = Ctloss + Ctcur (1c)
ns c
X
Ctloss = loss
Pt,sc ∗ λloss ∗ P bt,sc (1d)

PT
sc=1
Xns c
Ctcur = cur
Pt,sc ∗ λcur ∗ P bt,sc (1e)
sc=1

RI
N
XDG
cur N,cur
Pt,sc = Pt,sc (1f)

SC
133

134 2.5. Constraints and optimisation

U
135 Power balance at each node, Kirchhoff’s current law (KCL), Kirchhoff’s voltage law (KVL), voltage
136 and thermal limits and logical limits are constraints considered in this model.
137
AN
The optimisation problem is solved by the iterative linear approximation of the current voltage alter-
138 native current optimal power flow (ILIV-ACOPF), which is developed and tested in a series of papers
M
139 [36, 37]. In this algorithm non-convex equations are approximated by their first-order Taylor series.
140 Convex constraints are approximated by a set of linear equations to create an outer bound. Later the
D

141 algorithm was extended to consider integers and was compared with nonlinear ACOPF with integers in
142 standard test systems [35]. It was shown that the iterative linear OPF usually can find the solution within
TE

143 1% of the benchmarking nonlinear ACOPF and decrease computation time substantially. However, the
144 algorithm has been only applied for static power flow studies. The algorithm is extended to stochastic
EP

145 power flow studies in current paper.

146 The formulation of constraints is adjusted for the solving algorithm. The algorithm is better explained
together with the formulation of constraints. Constraints (2-13) depend on both time period t and scenario
C

147

148 sc, but the subscripts t and sc are not shown in these equations for readability. In the iteration h, the
AC

149 constraints are as shown below.

150 1. power balance
N N N N N,d N,d N,d N,d
f (Vre , Vim , Ire , Iim , Vre , Vim , Ire , Iim ) = (P N − P N,cur ) + j ∗ (QN − QN,cur )for d = h − 1 (2)

2. KCL and KVL

L N
AIre − Ire =0 (3a)
L N
AIim − Iim =0 (3b)
151
L,AL
|A> ∗ Vre
N
− RL,AL Ire
L,AL
+ X L,AL Iim | ≤ M (1 − DL,AL ) (4)

6
 ACCEPTED MANUSCRIPT

152 3. losses
NS NS
|Vre Ire + P NDG − P NLD | = P loss (5)

153 4. physical limits

N N N
coscut ∗Vre + sincut ∗Vim ≤ Vmax (6)
154
L,AL L,AL L,AL L,AL
coscut ∗Ire + sincut ∗Iim ≤ Imax D (7)
155
NS ,AL NS ,AL NS ,AL NS ,AL
coscut ∗Ire + sincut ∗Iim ≤ Imax D (8)
156

PT
N,d N N,d N N
Vere ∗ Vre + Veim ∗ Vim ≤ (Vmax )2 if itdd = 1, where d=h−1 (9)
157
L,AL,d L,AL L,AL,d L,AL L,AL 2 L,AL
Iere ∗ Ire + Ieim ∗ Iim ≤ (Imax ) D or / and
NS ,AL,d NS ,AL
(10)
NS ,AL,d NS ,AL NS ,AL 2 NS ,AL
Iere ∗ Ire + Ieim ∗ Iim ≤ (Imax ) D if itcd = 1, where d=h−1

RI
158
N,d N N,d N N
Vere ∗ Vre + Veim ∗ Vim ≥ (Vmin )2 if itxd = 1, where d=h−1 (11)
159
N
−2 ∗ Vmax (a/hb ) ≤ N,d
Vre − N
Vre ≤2∗ N
Vmax (a/hb )for d=h−1 (12)

SC
160
N N,d
−2 ∗ Vmax (a/hb ) ≤ Vim N
− Vim N
≤ 2 ∗ Vmax (a/hb )for d = h − 1 (13)

161 5. logical constraints X

EtL,AL ≤ 1, ∀L, NS (14)

U
AL,T
162
DtL,AL ≤ Dt−1
L,AL
+ EtL,AL , ∀L, NS (15)
163 AN
X R,AL
Dt
AL
=1 (16)
164 X dg
DtK ,AL
≤1 (17)
M
K dg ,AL
165 X dg
DtK ,AL
≥1 (18)
D

K dg ,AL,T
TE

P NDG ,cur ≤ P NDG (19a)

NDG ,cur NDG
Q ≤Q (19b)
EP

166 Constraint (2) is the first order approximation for the power balance at each node (20a) evaluated at
167 the previous results. Constraint (6) to (10) are linear approximation of (20b, 20c) for the upper limits of
168 voltage and current. (20b) is the upper and lower limits for voltage at each node. (20c) is the upper limit
C

169 for current. This constraint is only valid if the line exists. The calculation of corresponding parameters
AC

170 (coscut , sincut ) to formulate the outer bound in (6-8) can be found in [36]. The bound is tighten iteratively
171 by adding more tangent lines (also called Kelly’s cutting planes) based on results from previous iterations
172 as shown in constraint (9-10). In each iteration, the cutting planes are calculated for scenarios where
173 there are constraint violations.

N N N N
(Vre + j ∗ Vim )(Ire − j ∗ Iim ) = (P N − P N,cur ) + j ∗ (QN − QN,cur ) (20a)
N
(Vmin )2
≤ (VreN 2
) + N 2
(Vim ) ≤ N
(Vmax )2 (20b)
L,AL 2 L,AL
(Ire ) + (Iim )2 L,AL 2 L,AL
≤ (Imax ) D (20c)

7
 ACCEPTED MANUSCRIPT

174

175 The lower voltage limits are not convex sets, therefore accumulating cutting planes is not applicable.
176 However, in the case of optimal power flow, the voltage tends to be as high as possible to decrease losses.
177 Therefore, the lower voltage limits are only considered when a violation occurs, as shown in constraint
178 (11). The cutting plane is added only when there occurs a lower voltage violation in the previous iteration,

PT
179 as opposed to the other linearised constraints where cutting planes are accumulated from all previous
180 iterations.
181 Equations (3a-3b) which represent the Kirchhoffs current law (KCL) insure that the sum of currents

RI
182 flowing into a node is equal to the sum of currents flowing out of that node. Equations (3a) shows the
183 real part of current and (3b) shows the imaginary part of current. Constraint (4) which represents the

SC
184 Kirchhoffs voltage law (KVL) insures that the sum of the voltage around a closed network is zero. This
185 constraint is valid only if the line exists between two nodes. This is ensured by adding a big number M on

U
186 the right hand side of the inequality. Therefore, only if the line does exist, there is an equality constraint
that represents KVL. Constraint (5) calculates the thermal losses in the network, which is the difference
187

188
AN
between the injected energy, including energy from the substation and DG, and consumed energy.
189 Constraints (12) and (13) are used to control the voltage differences between iterations for each
M
190 scenario. These constraints exhibit periodic behaviour in iterations [37].
191 Constraints from (14) to (18) show the logical constraints for the investment and availability of each
D

192 branch and alternatives. Equation (14) shows that maximum one investment on each branch (and sub-
193 station) is permitted in the planning horizon. Equation (15) shows that the alternatives for replacement
TE

194 or connection branches (and substation) are available only after the corresponding investment has been
195 done. Equation (16) shows that only one alternative of replacement branches will be available in a period.
EP

196 Equation (17) shows that at most one branch among the connection branches is built to connect DG in
197 one period. Equation (18) shows that all DG should be connected in the end of the planning period.
Constraint (19a-19b) ensures that the curtailment is not more than the production of DG.
C

198

199 The iteration starts with defining the starting voltages for all nodes as shown in Fig.A.1. The starting
AC

200 value for current is calculated by (20a) assuming there is no curtailment. Solve the optimisation with
201 the objective function (1) and constraints (2-8, 14-19) in the first iteration. Results are checked with
202 the original nonlinear constraints (20a-20c). If the results are not feasible to the original constraints, the
203 bound is tighten and modified by (9-13) in the next iteration, moreover the results are used to update
204 the starting values. The iteration ends when the average performance measures, which are defined in the
205 ILIV-ACOPF algorithm [36, 37], over all scenarios are satisfied.

8
 ACCEPTED MANUSCRIPT

206 2.6. Reliability

207 Reliability indices, expected energy not served (EENS), system average interruption duration index
208 (SAIDI) and system average interruption frequency index (SAIFI) are calculated after the optimisation.
209 The optimal solution based on the cost minimisation may not fulfil given reliability requirement since
210 reliability is not explicitly considered in the optimisation problem. Therefore, a solution pool with several

PT
211 best solutions in the optimisation are saved to further study the system reliability.
212 The switching devices in the system which are used for preventing the failure expanding in the system
and maintenances, are not modelled and therefore the reliability is evaluated through a compromise

RI
213

214 solution between an upper and a lower bound on the reliability level. The upper bound is calculated
215 assuming that no switching devices are installed in the network, while the lower bound is calculated

SC
216 assuming that all branches are equipped with switching devices. The compromise is modelled by an
217 improvement coefficient [43]. It is a coefficient to simulate the effect of a compromise solution in switching

U
218 device location. The reliability level is estimated by an analytical method considering all line failures and
load levels [5, 6]. DG units are considered as alternative supplying sources when an outage occurs [5].
219

220 3. Modelling of regulatory issues

AN
M
221 3.1. Regulatory issues

222 DSOs are natural monopolies, therefore the decisions that they make are directed by regulation.
D

223 Different economic regulations are modelled in order to analyse their impacts on network investment. A
224 first aspect is the regulation on losses. With the increase of DG integration, the impact on losses will
TE

225 increase. More DG units which are close to the consumption points can decrease losses in the network.
226 However, losses in the network nowadays is usually treated by a pass through cost [44]; therefore the
EP

227 possible loss reduction benefit for DSOs from DG is not apparent. Economic regulation on losses are
228 justified by the fact that DSOs have the ability to control losses to some extent [44].
229 Another aspect is the regulation on DG curtailment. Generation curtailment, including conven-
C

230 tional generation and renewable generation, is a common practice in transmission levels [45]. Curtailment
AC

231 can also occur in the distribution system due to the increasing DG penetration level. If curtailment is
232 allowed by regulation, it would affect the planning decision of DSOs. On one hand, if the network was
233 dimensioned according to the extreme scenario, the investment cost could be high without curtailment
234 and the network would be redundant most of the time. On the other hand, high permissible curtailment
235 would lead to under-invest in the network, less renewable energy and maybe more losses.
236 The third aspect is the regulation on connection fees for DG. The approaches for connection
237 fees for DG are usually categorized as deep and shallow connection charges [24]. Deep connection charges
238 mean that DG owners are obliged to pay for the connection line and the grid upgrade and reinforcement.

9
 ACCEPTED MANUSCRIPT

239 Shallow connection charges mean that DG owners only pay for the connection line. How to link these
240 regulatory aspects to the DSO’s investment model is discussed below.

241 3.2. Regulatory issues modelling

242 R-1: regulation on losses. Losses in the system are assumed to be either a) associated with a cost,
243 for example as in France, Germany and many other EU member states [44]; or b) constrained by a certain

PT
244 permissible level, for example as in Portugal [44]. The difference is that for a) the cost of losses (Ctloss ) is
245 in the objective function, while for b) the losses will be upper bounded in the constraints. In the model

RI
246 introduced in Section 2 the cost of losses is minimised. The cost of losses depends on electricity price or
247 a specific market for losses [44]. To model the price is beyond the scope of this paper. To apply upper
bounds on losses is modelled as following:

SC
248

ns c
X ns c
X
loss NL D
Pt,sc ∗ P bt,sc ≤ η Pt,sc ∗ P bt,sc ∀t ∈ T (21)
sc=1 sc=1

U
249 where a certain percentage (η) of the total consumption is set as the upper bound on losses. η can
250 also be dependent on the planning period considering that the regulation may change between planning
251 periods.
AN
252 R-2: regulation on DG curtailment. The curtailed energy in the system is assumed to be either a)
M
253 associated with a cost, for example a certain amount of curtailed energy will be compensated in Portugal
254 [45]; or b) discouraged by a certain quota, which is equivalent to encourage a certain penetration level of
D

255 DG. In the model introduced in Section 2 cost of curtailment (Ctcur ) is minimised. The cost of curtailment
256 can be the value of the curtailed power, the price that DSOs pay for it or a regulatory penalty. To apply
TE

257 upper bounds on curtailment is modelled as following:

ns c
X ns c
X
EP

cur NDG
Pt,sc ∗ P bsc ≤ γ1 Pt,sc ∗ P bsc ∀t ∈ T (22a)
sc=1 sc=1
Xns c Xns c
Qcur
t,sc ∗ P bsc ≤ γ2 QN
t,sc ∗ P bsc
DG
∀t ∈ T (22b)
C

sc=1 sc=1

258 where γ is a parameter for the highest permissible curtailment level as a share of available DG.
AC

259 R-3: regulation on connection fees for DG. The connection fee is based on either a) deep fees
260 or b) shallow fees. In the deep fee regime, it is assumed that DG owners choose a connection plan which
261 has the least price offered by DSOs. The price is assumed to be related with the total cost for DSOs to
262 integrate DG. Therefore, in this case the lines to connect DG are determined by the least cost for DSOs,
263 which is as in the model introduced in Section 2. The model can however be readily adapted to consider
264 shallow fees by defining decision variable which represents the connection lines (EtK,AL ) as parameters.
265 This enables the model to minimize reinforcement cost on the network given the connection lines that
266 DG owners propose.

10
 ACCEPTED MANUSCRIPT

267 In addition to all the above aspects, the unbundling of the DSOs should be valid in any case. Due
268 to the unbundling regulation the decision of DG connection is separated from the decision of network
269 investment. Therefore the DG connection is the input of the network investment optimisation instead of
270 the output. However, in the case of the deep connection charges, despite the DG owners make decisions
271 independently from the DSOs, their costs are closely linked. In this paper, the DG owners are modelled

PT
272 as cost minimizer, and it is assumed that they will choose the solution that minimizes the DSOs cost
273 due to the DG connection. Therefore, the DG connection is modelled as the output of the network
274 investment optimisation. A summary of modelling these three kinds of regulation is presented in Table

RI
275 A.1. In reality each power system depends on all three kinds of regulation. Therefore eight regulatory
276 cases (all combinations of the discussed regulation) are carried out in Section 4.

SC
277 4. Case study

U
278 To apply the proposed model considering different regulatory cases, a radial network where DG plants
have asked for connection is created. The system details (node data and line data) can be found in [46].
279

280
AN
This network was created because data for alternatives for upgrading standard networks are not available.
281 The computation is performed in a PC Intel Core i7 2.70 GHz, 8 GB RAM. The iterative procedure is
M
282 performed in MATLAB 7.11. The mixed integer optimisation programming is performed in GAMS 23.9
283 using CPLEX solver with relative optimality criterion 0.001. The solution pool is obtained during the
D

284 optimisation by CPLEX solver. The CPLEX option “solnpool” triggers the collection of solutions [47].
285 Cplex also offers different ways to filter the solutions. In this paper, the collected solutions in the solution
TE

286 pool are the best 20. Total execution time for each case is around 350 seconds and it obtains the optimal
287 solution after around six major iterations (index h in Section 2.5).
EP

288 4.1. System characteristics

289 The network is a distribution system with 21 load nodes displayed in Fig. A.3. The network operates
C

290 at 24.9 kV. Two wind farms are in the pipeline to be connected. For each wind farm, three possible
connection routes are predefined. In the figure, the square node represents the feeding substation, the
AC

291

292 wind turbines represent wind farms, and the circles are the load points and wind farm locations (N22
293 and N23). The branches between nodes represent the electrical connections between nodes. Continuous
294 lines denote the existing network and dashed lines are candidate routes for new connections. The base
295 values for the whole network are 2.5 MVA and 24.9 kV.
296 Investment for three planning periods (two years, two years and four years) is considered in the case
297 study. The time line is demonstrated in Fig. A.2. A DG owner applies for a connection to N22 in the
298 beginning of the period 2 and another DG owner applies for a connection to N23 in the beginning of

11
 ACCEPTED MANUSCRIPT

299 period 3. The two DG units are assumed to maintain a power factor 1 and 0.95 respectively, and they
300 act as negative load. The costs of lines are discounted to the beginning of the planning horizon.
301 In this grid, all lines have two alternatives for upgrade, AL1 is the initial line and AL2 AL3 are the
302 alternatives (thicker conductors and higher costs), and there are three paths to connect each new wind
303 farm into the network. L8, L13, and L21 are for N22. L9, L15, and L24 are for N23. Each new path

PT
304 has three alternatives. The decisions of alternatives (total three) of each line (total 26) are made for each
305 time period (total three periods), therefore, the optimisation problem has 26 ∗ 3 ∗ 3 = 234 binary variables
306 for investment and another 234 binaries to represent the availability of each line in each period.

RI
307 4.2. System uncertainties

SC
308 In each period eight groups of random variables are generated (seasons and weekday/weekend). In
309 the group there are one random variable that represents the total load and one random variable that
310 represents the total wind power generation. The load and wind power are then spread on each node in

U
311 the network according to the installed capacity. This corresponds to assuming that load consumptions
312

313
AN
are fully correlated and wind power distributions at all sites are fully correlated.
The load and wind power random variables are assumed to follow Beta distributions. Beta distribution
314 can be applied to model the behaviour of random variables limited to intervals of finite length. In network
M
315 planning model, it is realistic to model load and wind power limited to a finite interval, which means that
316 the minimal and maximum values are assumed to be well-defined. The maximum load for each period
D

317 depends on the growth in that period, the type of the day and the type of the load mix(residential,
TE

318 commercial or industrial). Other distribution functions can be easily applied as well. For example wind
319 speed distributions with a standard power curve could be used to model wind power DG.
320 In each group, twenty equally probable realizations are generated from both the load and the wind
EP

321 power probability density function. Therefore, the probability for each realization is 0.05. According to
322 the group the density functions belong to, a scenario is generated by matching the realizations of the load
C

323 with that of the wind power. Therefore, in total 481 scenarios (160 scenarios for each planning period
and 1 scenario represents now) have been considered. The probability of a scenario is the product of the
AC

324

325 probability of the realization in the group and the probability of this group in a year. After applying
326 the scenario reduction tool, the relative distances between the original scenario tree and reduced scenario
327 trees are calculated. The expected costs of the reduced trees are also calculated to show the impact of
328 scenario reduction on the objective function. With the number of scenarios in the reduced scenarios trees
329 increases, the relative distance decreases first then stay stable while the difference between expected costs
330 increases first then decrease and finally stay stable. By comparing these two parameters, a 40-scenario
331 tree is considered as an appropriate choice. Using a 40-scenario tree instead of a 100-scenario tree incurs

12
 ACCEPTED MANUSCRIPT

332 a 1.2% deterioration in the relative distance, while the expected cost difference is just 0.22%. The time
333 of computation reduces from 3377 seconds to 350 seconds.

334 4.3. Value of Predefined Parameters

335 The cost of losses is set as 0.23 e per kWh. This can be possible if losses are valued by electricity
336 market price [48]. In cases which the regulation sets an upper bound on losses, η in (21) is set to 5%.

PT
337 The curtailment cost is set as the same cost as for losses. In cases in which the regulation sets an upper
338 bound on curtailment, γ in (22) is set to 0.2; therefore, the total curtailed energy is less than 20% of

RI
339 generation. Wind power curtailment was 10.7% in Italy in 2009 and it decreased to 1.24% in 2012 [45].
340 This γ can be set lower to encourage higher energy penetration.

SC
341 In the optimisation constraints, M in (4) is set as 10000, and predefined parameters for step size in
342 the optimisation are a = 0.5, b = 2 in (12) and (13). The voltages are limited between [23.66 26.15] kV
343 (±5 %). The stopping criteria for iteration process are set as following. The maximum number of the

U
344 major iteration (index h in Section 2.5) is set to 20. The maximum percentage violations of all nonlinear
345

346
AN
constraints in all scenarios is set to less than 0.005 and the summed violations of all constraints in all
scenarios is set to less than 0.005. When any of these three criteria is satisfied the iteration stops.
347 Eight cases considering all combination of regulatory cases presented in Section 3 are studied here,
M
348 as shown in Table A.2. From Case 5 to Case 8 shallow connection fee regulation is applied. The DG
349 connection lines paid by the DG owners are L21.AL2.T2 and L9.AL2.T3. The cost is 2.5 Me and 2.5
D

350 Me respectively.
TE

351 4.4. Results and discussion

352 Table A.2 summarizes the minimum cost results for eight cases considering all combinations of reg-
EP

353 ulatory cases presented in Section 3. The investment plan shows the optimal decisions that are not in
354 common among in all cases. L1.AL2.T1 here means that Line 1 should be upgraded to alternative 2
355 at the beginning of planning period 1. The cost is in unit Me . The investment cost in Case 1-4 is the
C

356 DSOs’ capital cost and while in Case 5-8, where the connection cost is paid by DG owners, is the DSOs’
AC

357 capital cost and DG connection cost. The NPV of the connection cost is 4.46 Me . The cost of losses or
358 curtailment is zero in some cases because there is no relevant monetary regulation.
359 More integrated DG (less curtailment) can reduce losses in the system due to the network upgrading;
360 however, the reduction can lead to an increase on the investment. In Table A.2, comparing Case 1 and
361 Case 3, it can be observed that the constraint on losses does not help decrease losses but just increase the
362 DSO’s cost. This is expected since in Case 3 the constraint on losses is not binding (the upper bound for
363 losses is set as 5%). Comparing Case 2 and Case 4, the losses are lower in Case 2 than in Case 4. This
364 is because losses are considered as costs in Case 2 while the losses can be high as long as they are within

13
 ACCEPTED MANUSCRIPT

365 the constraint in Case 4. Moreover, more curtailment leads to higher losses in the system. However,
366 comparing Case 3 and Case 4, the more curtailment, the lower losses in the system. It also shows that
367 allowing more curtailment can defer the reinforcement in the existing network.
368 Comparing Case 1 and Case 2, the investment difference lies in the upgrading choices of L1, L14
369 and L21. In Case 2, the DG curtailment is higher, therefore, the capacity of the connection line L21 is

PT
370 smaller. However, the capacity of L1 is larger. This is because more electricity is transferred from the
371 substation to the load. The capacity of L14 is larger in Case 1 in order to maintain the voltage given a
372 larger voltage drop on L1. This analysis applies in comparing Case 3 and Case 4 as well.

RI
373 Comparing Case 5 and Case 6, the investment difference lies in the upgrading choices of L2, L4, L14
374 and L23. In these cases, the connection lines for DG are L21 and L9. In Case 5, the DG curtailment

SC
375 is higher, therefore, more electricity is transferred from the substation. This requires a larger capacity
376 of L2. However, the capacity of L4 is smaller because one DG is connected to Node 4 and the flow on

U
377 L4 is towards the substation in some scenarios. The higher capacity of L14 and L24 in Case 6 is for
maintaining the voltage given a larger voltage drop on L2. This analysis applies in comparing Case 7
378

379 and Case 8 as well.

AN
380 Comparing Case 1 and Case 5 shows that the total investment cost is higher in the cases with shallow
M
381 fees where the connection lines are given. It is still true if the cost of connection line (in total 4.46
382 Me ) is deducted from the total investment cost in Case 5. This applies in all cases. This is expected
D

383 since the DG owners chose the connection line from their point of view. In addition it can be seen that
384 curtailment is higher under shallow fees regulation. This is because DG owners have limited information
TE

385 on the overall impact of in a connection point in the distribution network.

386 Sensitivity on prices for curtailment is analysed for Case 1. In this case the curtailment is very low;
EP

387 therefore, the price for curtailment is set lower. The results are shown in Fig. A.4a and Fig. A.4b. It
388 shows that the curtailment is not sensitive to the price. The curtailment does not increase until when
the price is 10 times lower.
C

389

390 Reliability indices in each period are calculated by setting the improvement coefficient to 0.3. The
AC

391 reliability indices in Case 1 for all three periods for the optimal solution are shown in Table A.3. SAIDI
392 and SAIFI are increasing due to the network expansion and load growth. However, EENS decreases in
393 period 2. This is because the DG 1 is connected to the grid. It increases in period 3. One reason can be
394 that the load increases more than DG. EENS is calculated for the best 20 solutions in Case 1 Period 1.
395 Two non-dominated solutions are found as shown in Fig. A.5. Analysing reliability for the solution pool
396 provides DSOs a tool to achieve a trade-off between cost and reliability.

14
 ACCEPTED MANUSCRIPT

397 5. Strengths and Limitations

398 The main strengths of the developed model lie in adding regulation impact in formulation of network
399 planning, considering the uncertainties of load and DG in multi planning periods, the tractability of
400 the computation due to the linear formulation, and the flexibility in applying other networks which can
401 be done by changing the line to node matrix. However, many aspects can be improved in the future.

PT
402 One of them is the network planning decision making criteria. The planning decision is assumed to be
403 based on the NPV of the total cost, which is very ideal. One of them is in the calculation of reliability.

RI
404 The reliability is evaluated through a compromise solution between an upper and a lower bound on
405 the reliability level. Last but not least, the model needs to be applied to a larger network. The linear

SC
406 formulation has been applied in a larger network with 300 nodes in [36]. However, the application in [36]
407 only considers power flow in one snap shot. To apply the model which considers multi-scenario power
408 flow to a larger network needs to be further investigated. To apply the model to a larger network also

U
409 requires better load and DG models. This is because in a larger network, the correlation between load
410 and DG becomes more complicated. AN
411 6. Conclusion
M
412 This paper has shown that the interplay between different economic regulations for distribution sys-
413 tems. The impacts are evaluated by a network planning model. A new model formulation for distribution
D

414 network planning considering DG units which are not owned by the DSO is proposed. The formulation
TE

415 considers uncertainties from load and DG in the planning periods. It also encodes regulation on losses,
416 DG curtailment and DG connection fee. This model provides a tool to analyse how the regulations on
417 DG and losses are affecting each other, how the DSO’s expansion plan changes according to economic
EP

418 regulations and how the reliability can be affected.

419 The case study has shown that the impact of regulations on network expansion planning is intricate.
C

420 In particular, it was shown that the effectiveness of regulatory approaches on DG curtailment and losses
AC

421 is not straightforward. Therefore, in order to adequately analyse the impact or to design future regulation
422 schemes, advanced models such as the one developed here are needed. In this case study, where it is a
423 weak network with high DG penetration level studied, the total investment for the network is between
424 20-25% lower if the deep fee for DG connection is implemented. Moreover, regulation on curtailment
425 is linked with the regulation on connection fee. When it is deep connection fee applied, to charge the
426 DSO for curtailing energy is more effective to reduce the curtailment than to put an upper bound but
427 the total cost is higher. The curtailment is about 10% lower and the investment cost increases by 1.7%
428 when all the curtailed energy is charged. When it is shallow connection fee applied, to charge the DSO

15
 ACCEPTED MANUSCRIPT

429 for curtailing energy is less effective to reduce the curtailment than to put an upper bound and the total
430 cost is also higher. The curtailment is about 2.8% higher and the investment cost increases by 4% when
431 the curtailment is charged. The loss percentage does not vary much with different regulatory schemes.
432 The variation is around 0.2%. To charge losses is slightly more effective to reduce losses than to put an
433 upper bound for losses.

PT
434 The model can further be developed to analyse expansion decisions considering other economic regu-
435 lation, for example incentive regulations for renewable energy and impact from other forms of regulation
436 which allow the DSOs to own DG. Furthermore, other scenario generation tool considering load and wind

RI
437 power correlation can be integrated with this model to more accurately represent the uncertainty.

SC
438 Appendix A. Nomenclature
439 1. Sets
L All fixed and possible branches, e.g. L1,L2,L3...;
N All nodes in the network, e.g. N1,N2,N3...;

U
NS The substation node, subset of N;
F Set of fixed branches, subset of L;
440
R
K
AL
NLD
AN
Set of reinforcement branches, subset of L;
Set of all connection branches for all DG units, subset of L;
Set of alternatives;
Set of load nodes, subset of N;
NDG Set of DG nodes, subset of N;
M
K dg Set of connection branches for connection of dg in NDG , subset of K
441 2. Parameters
442
D

T Number of periods in the planning horizon;

δ The annual discount rate;
δtcap , δtoper Present value factors for the investment and operation costs in planning period
TE

t;
nsc Number of scenarios;
N N
Vmin , Vmax Minimum and maximum node voltage limits;
EP

L,AL
Imax Maximum current limits on each branch;
RL,AL , X L,AL Resistance and reactance of the alternatives of each branch;
A node-branch incidence matrix;
N
Pt,sc , QN
t,sc Active and reactive power of demand or supply on each node in planning period
C

t scenario sc;
C R,AL C K,AL C NS ,AL Vector of investment costs of the alternatives of the replacement branches in R;
connection branches in K; substations in Ns respectively;
AC

λloss Price of power losses;

λcur Price of DG curtailment;
ncut Number of pre-processed cut constraints;
P bt,sc Probability of scenario sc in in planning period t;
coscut , sincut Parameters for pre-processed cut constraints;
N,d N,d
Vre,t,sc Vim,t,sc Real part and imaginary part of voltage in planning period t scenario sc in
iteration d;
N,d N,d
Vere,t,sc Veim,t,sc Rescaled real part and imaginary part of voltage in planning period t scenario sc
if the voltage limit in iteration d is violated;
L,AL,d eL,AL,d
Iere,t,sc Iim,t,sc Rescaled real part and imaginary part of current in all alternatives for each
branch in planning period t scenariosc if the current in iteration d is larger than
the thermal limit;
d, h Indices for previous and current iterations;
itcd Parameter is 1 if any real current constraint is violated in iteration d;

16
 ACCEPTED MANUSCRIPT

itdd Parameter is 1 if any real maximum voltage constraint is violated in iteration d;

itxd Parameter is 1 if any real minimum voltage constraint is violated in iteration d;
a, b Parameters for step-size;
η Maximum percentage on losses;
γ Maximum curtailment percentage;

443 3. Binary variables

dg
DtR,AL DtK ,AL
Vector of availability of alternatives of the replacement branches and connection branches

PT
444 in planning period t;
EtL,AL Vector of binary decision on investment of alternatives of the upgrading or installation in
planning period t;
4. Variables

RI
445

cur
Pt,sc Total curtailed power in planning period t scenario sc;
loss
Pt,sc Loss in the system in planning period t scenario sc;
L,AL L,AL
Ire,t,sc , Iim,t,sc Real part and imaginary part of current flows in all alternatives for each

SC
branch in planning period t scenario sc;
NDG ,cur
Pt,sc QN DG ,cur
t,sc Real part and imaginary part of curtailed power from DG in planning period
t scenario sc;
N N
446 Ire,t,sc , Iim,t,sc Real part and imaginary part of current injection from the load point or DG

U
connection point in planning period t scenario sc;
N N
Vre,t,sc , Vim,t,sc Real part and imaginary part of all nodal voltages in planning period t sce-
nario sc;
Ctcap , Ctoper

Ctloss , Ctcur
ning period t;
AN
Capital expenditure (CAPEX) and operational expenditure (OPEX) in plan-

Cost of losses and curtailment in planning period t;

Ctotal Total cost.
M
447 References
D

448 [1] M. Nehrir, C. Wang, V. Gerez, Impact of wind power distributed generation on distribution systems,
in: Proc. IEEE 17th Int. Conf. Electricity Distribution (CIRED), Barcelona, Spain, 2003.
TE

449

450 [2] A. Piccolo, P. Siano, Evaluating the impact of network investment deferral on distributed
generation expansion, Power Systems, IEEE Transactions on 24 (3) (2009) 1559–1567.
EP

451

452 doi:10.1109/TPWRS.2009.2022973.

[3] L. F. Ochoa, A. Padilha-Feltrin, G. P. Harrison, Evaluating distributed generation impacts with a

C

453

454 multiobjective index, Power Delivery, IEEE Transactions on 21 (3) (2006) 1452–1458.
AC

455 [4] M. Shaaban, Y. Atwa, E. El-Saadany, Dg allocation for benefit maximization in

456 distribution networks, Power Systems, IEEE Transactions on 28 (2) (2013) 639–649.
457 doi:10.1109/TPWRS.2012.2213309.

458 [5] H. Falaghi, C. Singh, M. R. Haghifam, M. Ramezani, DG integrated multistage distribution system
459 expansion planning, International Journal of Electrical Power & Energy Systems 33 (8) (2011) 1489–
460 1497. doi:10.1016/j.ijepes.2011.06.031.

17
 ACCEPTED MANUSCRIPT

461 [6] C. L. Borges, V. F. Martins, Multistage expansion planning for active distribution networks un-
462 der demand and Distributed Generation uncertainties, International Journal of Electrical Power &
463 Energy Systems 36 (1) (2012) 107–116. doi:10.1016/j.ijepes.2011.10.031.

464 [7] N. Khalesi, N. Rezaei, M. R. Haghifam, DG allocation with application of dynamic programming
465 for loss reduction and reliability improvement, International Journal of Electrical Power & Energy

PT
466 Systems 33 (2) (2011) 288–295. doi:10.1016/j.ijepes.2010.08.024.

467 [8] D. 2003/54/EC, of the european parliment and of the council of 26 june 2003 concerning common

RI
468 rules for the internal market in electricity and repealing directive 96/92/ec (2003).

SC
469 [9] C. Ma, P. Kaufmann, J. C. Tbermann, M. Braun, Optimal generation dispatch of distributed gen-
470 erators considering fair contribution to grid voltage control, Renewable Energy 87, Part 2 (2016)
471 946–953. doi:10.1016/j.renene.2015.07.083.

U
472 [10] A. V. Pombo, J. Murta-Pina, V. F. Pires, A multiobjective placement of switching devices in dis-
473
AN
tribution networks incorporating distributed energy resources, Electric Power Systems Research 130
474 (2016) 34 – 45. doi:http://dx.doi.org/10.1016/j.epsr.2015.08.012.
M
475 [11] J. Kennedy, P. Ciufo, A. Agalgaonkar, A review of protection systems for distribution networks
476 embedded with renewable generation, Renewable and Sustainable Energy Reviews 58 (2016) 1308 –
D

477 1317. doi:http://dx.doi.org/10.1016/j.rser.2015.12.258.

TE

478 [12] M. Scheepers, D. Bauknecht, J. Jansen, J. D. Joode, T. Gómez, D. Pudjianto, S. Ropenus, G. Strbac,
479 Regulatory improvements for effective integration of dg into electricity distribution networks, Tech.
rep. (2007).
EP

480

481 URL http://www.ecn.nl/docs/library/report/2007/e07038.pdf

C

482 [13] P. Frı́as, T. Gómez, R. Cossent, J. Rivier, Improvements in current European network regulation
483 to facilitate the integration of distributed generation, International Journal of Electrical Power and
AC

484 Energy Systems 31 (9) (2009) 445–451. doi:10.1016/j.ijepes.2009.03.001.

485 [14] J. de Joode, J. C. Jansen, A. J. van der Welle, M. J. J. Scheepers, Increasing penetration of renewable
486 and distributed electricity generation and the need for different network regulation, Energy Policy
487 37 (8) (2009) 2907–2915. doi:10.1016/j.enpol.2009.03.014.

488 [15] R. Turvey, On network efficiency comparisons: Electricity distribution, Utilities Policy 14 (2) (2006)
489 103–113.

18
 ACCEPTED MANUSCRIPT

490 [16] V. Levi, G. Strbac, R. Allan, Assessment of performance-driven investment strategies of distribu-
491 tion systems using reference networks, IEE Proceedings-Generation, Transmission and Distribution
492 152 (1) (2005) 1–10.

493 [17] M. B. Larsson, The network performance assessment model: a new framework of regulating the
494 electricity network companies, Licentiate thesis, KTH (2005).

PT
495 [18] C. Domingo, T. Romn, A. Snchez-Miralles, J. Gonzlez, A. Martnez, A Reference Network Model for
496 Large-Scale Distribution Planning With Automatic Street Map Generation, Power Systems, IEEE

RI
497 Transactions on 26 (1) (2011) 190–197. doi:10.1109/TPWRS.2010.2052077.

498 [19] S. Gill, I. Kockar, G. W. Ault, Dynamic Optimal Power Flow for Active Distribution Networks,

SC
499 Power Systems, IEEE Transactions on 29 (1) (2014) 121–131. doi:10.1109/TPWRS.2013.2279263.

500 [20] L. Kane, G. W. Ault, Evaluation of Wind Power Curtailment in Active Network

U
501 Management Schemes, Power Systems, IEEE Transactions on 30 (2) (2015) 672–679.
502 doi:10.1109/TPWRS.2014.2336862. AN
503 [21] Z. Hu, F. Li, Cost-Benefit Analyses of Active Distribution Network Management, Part II, IEEE
M
504 Transactions on Smart Grid 3 (3) (2012) 1067–1074. doi:10.1109/TSG.2012.2205412.

505 [22] A. Sheikhi Fini, M. Parsa Moghaddam, M. K. Sheikh-El-Eslami, An investigation on the impacts of
D

506 regulatory support schemes on distributed energy resource expansion planning, Renewable Energy
TE

507 53 (2013) 339–349. doi:10.1016/j.renene.2012.11.028.

508 [23] M. Vaziri, K. Tomsovic, T. Gönen, Distribution expansion problem revisited. part 1 categorical
analysis and future directions, Mathematical Programming 13 (2000) 26–34.
EP

509

510 [24] Y. Huang, K. Alvehag, L. Söder, Regulation impact on distribution systems with distributed gen-
eration, in: 9th International Conference on the European Energy Market (EEM), Florence, Italy,
C

511

512 2012, pp. 1–8. doi:10.1109/EEM.2012.6254721.

AC

513 [25] A. Keane, L. F. Ochoa, C. L. T. Borges, G. W. Ault, A. D. Alarcon-Rodriguez, R. A. F. Currie,

514 F. Pilo, C. Dent, G. P. Harrison, State-of-the-Art Techniques and Challenges Ahead for Distributed
515 Generation Planning and Optimization, Power Systems, IEEE Transactions on 28 (2) (2013) 1493–
516 1502. doi:10.1109/TPWRS.2012.2214406.

517 [26] M. Ahmadigorji, N. Amjady, A multiyear dg-incorporated framework for expansion planning of
518 distribution networks using binary chaotic shark smell optimization algorithm, Energy 102 (2016)
519 199 – 215. doi:http://dx.doi.org/10.1016/j.energy.2016.02.088.

19
 ACCEPTED MANUSCRIPT

520 [27] S. Haffner, L. F. A. Pereira, L. A. Pereira, L. S. Barreto, Multistage Model for Distribution Expan-
521 sion Planning With Distributed Generation ;Part I: Problem Formulation, Power Delivery, IEEE
522 Transactions on 23 (2) (2008) 915–923. doi:10.1109/tpwrd.2008.917916.

523 [28] A. Soroudi, M. Ehsan, A distribution network expansion planning model considering dis-
524 tributed generation options and techo-economical issues, Energy 35 (8) (2010) 3364 – 3374.

PT
525 doi:http://dx.doi.org/10.1016/j.energy.2010.04.022.

526 [29] X. Tao, H.-J. Haubrich, A Two-Stage Heuristic Method for the Planning of Medium Voltage

RI
527 Distribution Networks with Large-Scale Distributed Generation, in: International Conference on
528 Probabilistic Methods Applied to Power Systems, IEEE, Stockholm, Sweden, 2006, pp. 1–6.

SC
529 doi:10.1109/pmaps.2006.360314.

530 [30] E. D. Mehleri, H. Sarimveis, N. C. Markatos, L. G. Papageorgiou, A mathematical programming

U
531 approach for optimal design of distributed energy systems at the neighbourhood level, Energy 44 (1)
532

533
AN
(2012) 96 – 104, integration and Energy System Engineering, European Symposium on Computer-
Aided Process Engineering 2011. doi:http://dx.doi.org/10.1016/j.energy.2012.02.009.
M
534 [31] G. M. Kopanos, M. C. Georgiadis, E. N. Pistikopoulos, Energy production planning of a network of
535 micro combined heat and power generators, Applied Energy 102 (2013) 1522 – 1534, special Issue on
Advances in sustainable biofuel production and use - {XIX} International Symposium on Alcohol
D

536

537 Fuels - {ISAF}. doi:http://dx.doi.org/10.1016/j.apenergy.2012.09.015.

TE

538 [32] S. Samsatli, N. J. Samsatli, A general spatio-temporal model of energy systems with a detailed
539 account of transport and storage, Computers & Chemical Engineering 80 (2015) 155 – 176.
EP

540 doi:http://dx.doi.org/10.1016/j.compchemeng.2015.05.019.

541 [33] S. Repo, H. Laaksonen, P. Jarventausta, Statistical models of distributed generation for distribution
C

542 network planning, in: 18th Int. Conf. Electricity Distribution (CIRED), Turin, Italy, IET, 2005, pp.
AC

543 1–5.

544 [34] A. Al-Sunaidy, R. Green, Electricity deregulation in {OECD} (organization for economic coopera-
545 tion and development) countries, Energy 31 (67) (2006) 769 – 787, electricity Market Reform and
546 Deregulation. doi:http://dx.doi.org/10.1016/j.energy.2005.02.017.

547 [35] A. P. Lipka, P. R. O’Neill, S. S. Oren, A. Castillo, M. Pirnia, C. Campaigne, Optimal transmission
548 switching using the iv-acopf linearization, Tech. rep., Technical report, US FERC (2012).

20
 ACCEPTED MANUSCRIPT

549 [36] P. R. O’Neill, A. Castillo, M. Cain, The iv formulation and linear approximations of the ac optimal
550 power flow problem, Tech. rep., Technical report, US FERC (2012).

551 [37] C. Campaigne, A. P. Lipka, M. Pirnia, P. R. O’Neill, S. Oren, Testing step-size limits for solving the
552 linearized current voltage ac optimal power flow, Tech. rep., Technical report, US FERC (2012).

[38] M. Carrion, A. B. Philpott, A. J. Conejo, J. M. Arroyo, A Stochastic Programming Approach to

PT
553

554 Electric Energy Procurement for Large Consumers, IEEE Transactions on Power Systems 22 (2)
555 (2007) 744–754. doi:10.1109/TPWRS.2007.895164.

RI
556 [39] E. B. Cedeño, S. Arora, Performance comparison of Transmission Network Expansion Planning

SC
557 under deterministic and uncertain conditions, International Journal of Electrical Power & Energy
558 Systems 33 (7) (2011) 1288–1295. doi:10.1016/j.ijepes.2011.05.005.

[40] H. Heitsch, W. Romisch, Scenario tree modeling for multistage stochastic programs, Mathematical

U
559

560 Programming 118 (2) (2009) 371–406.

561
AN
[41] H. Heitsch, W. Romisch, Scenario tree reduction for multistage stochastic programs, Computational
562 Management Science 6 (2) (2009) 117–133. doi:10.1007/s10287-008-0087-y.
M
563 [42] B. A. McCarl, A. Meeraus, P. van der Eijk, M. Bussieck, S. Dirkse, P. Steacy, F. Nelissen, Mccarl
gams user guide, Version (2014).
D

564

[43] V. Miranda, J. V. Ranito, L. M. Proenca, Genetic algorithms in optimal multistage distribution net-
TE

565

566 work planning, Power Systems, IEEE Transactions on 9 (4) (1994) 1927–1933. doi:10.1109/59.331452.

[44] European Regulators Group for Electricity and Gas (ERGEG), Treatment of losses by network
EP

567

568 operators, Position Paper (2009).

[45] D. Lew, L. Bird, M. Milligan, B. Speer, X. Wang, E. M. Carlini, A. Estanqueiro, D. Flynn, E. Gomez-
C

569

570 Lazaro, N. Menemenlis, et al., Wind and solar curtailment, in: International Workshop on Large-
AC

571 Scale Integration of Wind Power Into Power Systems, Citeseer, 2013.

572 [46] Y. Huang, 23-node testing system, https://kth.box.com/s/6gsesmg05nlmb8lwquuh (2014).

573 [47] GAMS Model library, Cplex solution pool for a simple facility location problem (2017).

574 [48] W. El-Khattam, Y. Hegazy, M. Salama, An integrated distributed generation optimization model
575 for distribution system planning, Power Systems, IEEE Transactions on 20 (2) (2005) 1158–1165.
576 doi:10.1109/TPWRS.2005.846114.

21
 ACCEPTED MANUSCRIPT

Load data
DG data The network
Regulation planning
data(1)(21)(22) model (1-22)

h=0 Solution pool

Initialization
(20a)
Yes

Reliability indices
calculation for selected

PT
Solve MIP subproblem Stopping solutions
(2-19), (1)(21)(22) criteria

h=h+1 No
Update the linearized End

RI
constraints (9-13, 20a)

Figure Figure
A.1: 1 Outline
Outline of the method
of the method

SC
Table A.1: Summary of the modelling of regulatory issues

Regulatory issues Modelling

Associated with a cost (1d)

U
Losses
Constrainted by an
(21)
upper limit
Curtailment
Associated with a cost
Constrainted by an
upper limit
AN (1e)
(22)
Connection fees Shallow The connection lines (EtK,AL )are given
M
for DG Deep The connection lines (EtK,AL )are outputs
D

years
TE

0 1 2 3 4 5 6 7 8 9

E1 E2 E3
EP

first stage decision second stage decision

n,cur
(Et ) (Pt,sc )

Figure A.2: Planning periods.

C

L14 L16 L17

8 9 10 11
AC

L15
L5 16 17 18
L24
23 L23 L25

L9 L10
L1 L2 L4
1 2 3 4 5 21
L6 L11

L3 L8 L7

L18 L19 L22

6 12 13 14 15

L12 7 L20
L21 19 20
22
L26
L13

Figure A.3: Diagram of the 23-node network

22
 ACCEPTED MANUSCRIPT

Table A.2: Optimal solutions for eight different cases

Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7 Case 8

λ ∗ P loss λ ∗ P loss P loss ≤ η P loss ≤ η λ ∗ P loss λ ∗ P loss P loss ≤ η P loss ≤ η
λ ∗ P cur P cur ≤ γ λ ∗ P cur P cur ≤ γ λ ∗ P cur P cur ≤ γ λ ∗ P cur P cur ≤ γ
deep deep deep deep shallow shallow shallow shallow
Invest. plan L1.AL2.T1 L1.AL3.T1 L1.AL2.T1 L1.AL3.T1 L1.AL3.T1 L1.AL3.T1 L1.AL3.T1 L1.AL3.T1

PT
L2.AL1.T1 L2.AL1.T1 L2.AL1.T1 L2.AL1.T1 L2.AL3.T1 L2.AL2.T1 L2.AL3.T1 L2.AL2.T1
L4.AL1.T1 L4.AL1.T1 L4.AL1.T1 L4.AL1.T1 L4.AL1.T1 L4.AL3.T1 L4.AL1.T1 L4.AL3.T1
L5.AL3.T1 L5.AL3.T1 L5.AL3.T1 L5.AL3.T1 L5.AL2.T1 L5.AL2.T1 L5.AL2.T1 L5.AL2.T1
L6.AL1.T1 L6.AL1.T1 L6.AL1.T1 L6.AL1.T1 L6.AL3.T1 L6.AL3.T1 L6.AL3.T1 L6.AL3.T1

RI
L14.AL3.T1 L14.AL2.T1 L14.AL3.T1 L14.AL2.T1 L14.AL2.T1 L14.AL3.T1 L14.AL2.T1 L14.AL3.T1
L23.AL2.T1 L23.AL2.T1 L23.AL2.T1 L23.AL2.T1 L23.AL2.T1 L23.AL3.T1 L23.AL2.T1 L23.AL2.T1
L21.AL3.T2 L21.AL2.T2 L21.AL3.T2 L21.AL2.T2 L21.AL2.T2 L21.AL2.T2 L21.AL2.T2 L21.AL2.T2
L24.AL2.T3 L24.AL2.T3 L24.AL2.T3 L24.AL2.T3 L9.AL2.T3 L9.AL2.T3 L9.AL2.T3 L9.AL2.T3

SC
Invest. cost 25.19 24.77 25.19 24.77 31.43 32.73 31.43 32.73
Curt. 0.33% 10.78% 0.33% 10.40% 18.67% 15.78% 18.67% 15.78%
Curt. cost 0.13 0 0.13 0 6.76 0 6.77 0
Losses 3.47% 3.29% 3.47% 3.44% 2.21% 2.46% 2.21% 2.46%

U
Loss cost 1.90 1.86 0 0 1.31 1.45 0 0
Total cost 27.22 26.63 25.32 24.77 39.50 34.18 38.20 32.73

AN
M
Table A.3: Reliability indices in Case 1

Indices Period 1 Period 2 Period 3

D

SAIDI (hr/cust,yr) 7.07 7.76 7.83

SAIFI (int/cust,yr) 1.31 1.36 1.37
0,0575 0,046 0,0345 0,023 0,0115
EENS (MWh/yr) 66.76 49.58 74.83
TE

0,23
price for curtailment 10 5 4 3 2 1 20
per curtailment,percentag 0,351 0,347 0,347 0,347 10,38 10,38
per loss 3,471719 3,467158 3,468386 3,467 3,261 3,261817 10
opex 1,8574 1,822 1,815 1,809 2,075 1,911
capex 251,926 25,193 25,193 25,193 24,767 24,767
EP

total cost 270,5 27,014 27,008 27,001 26,841 26,678 0,189535

total_load 16,29
total_DG 8,426 12
curtailment percentage

opex capex total cost 10

C

8
Percentage (%)
AC

6
27,014 27,008 27,001 26,841 26,678

25,193 25,193 25,193 24,767 24,767 2

1,822 1,815 1,809 2,075 1,911 0

0,0575 0,046 0,0345 0,023 0,0115
0,0575 0,046 0,0345 0,023 0,0115 Prices for curtailment (€/kWh)
Prices for curtailment (€/kWh)
(b) Curtailment levels Vs. different prices for cur-
(a) Costs Vs. different prices for curtailment tailment

Figure A.4: Different prices for curtailment

Prices for curtailment

23
 ACCEPTED MANUSCRIPT

PT
RI
SC
78

U
77

76

75 AN
EENS(MWh/yr)

74

73
M
72

71
D

70

69
TE

68
26 28 30 32 34 36 38
Total cost (106 Euro)

Figure A.5: Expected energy not served for selected solutions

EP
C
AC

24
 ACCEPTED MANUSCRIPT
Highlights
1. The impact of economic regulations is evaluated

2. A new network planning model is formulated

3. Distributed generation ownership unbundling is properly addressed

PT
RI
U SC
AN
M
D
TE
C EP
AC