Agent-based analysis of dynamic access ranges to

the distribution network

S. Mathieu, D. Ernst, B. Cornelusse
University of Li`ege, Belgium
AbstractThere is a need to clearly state an interaction model
that formalizes interactions between actors of the distribution
system exchanging flexibility. In previous works we quantitatively
evaluated the performance of five interaction models devised
with industrial partners using the agent-based testbed DSIMA.
Simulation results showed that these interaction models relying
on active network management suffer from a lack of coordination
between the distribution and the transmission system operator,
activating flexibility simultaneously in opposite directions. This
paper introduces a new interaction model fixing this issue based
on dynamic access bounds to the network changing throughout
the day and preventing the activation of flexibility leading
to congestions. This new interaction model is implemented in
DSIMA and compared to a model restricting the grid users to a
very restrictive but safe access range. Results show that this new
model allows to safely increase by 55% the amount of distributed
generation in the network.
Index Termsdistribution networks, market design, load flexibility, demand-side management, agent-based model, smart grid

I. I NTRODUCTION
A current trend in distribution systems is to use the flexibility of generation means (e.g. curtailment) and consumers (e.g.
load shifting), in addition to regular control assets such as tap
changers, for balancing or local congestion management. For
instance, distribution system operators (DSOs) rely on active
network management to anticipate and correct, using flexibility, operational limits violations when this is more economical
or easier to perform than straightforward grid reinforcement.
However, there are many possibilities to organize interactions
between the actors of the system when flexibility comes into
play. Thus, there is a need first to clearly state the interaction
models that formalize these interactions, and second to set up
a way for comparing their economical and technical merit.
In previous works we evaluated the performance of five
interaction models devised with industrial partners using the
open-source agent-based testbed DSIMA [1]. The scope of the
interaction models covers several stages from day-ahead exchanges of flexibility until settlement. In DSIMA, the actors of
the system are modeled as individual agents that solve a nested
sequence of optimization problems to take their decision in
order to maximize their individual objective. At each decision
stage the problem is constrained by the decisions taken at
the previous stages, and subsequent recourse possibilities are
taken into account. The agents considered are the DSO, the
transmission system operator (TSO), producers and retailers.
The impact of the agents decisions is evaluated through a
measure of the social welfare, the share of the welfare between
c 2016 IEEE
978-1-5090-3358-4/16/$31.00

the actors, and a measure of the service level.

Simulation results on an expected system in 2025 show
that the interaction models proposed in [1] relying on active
network management, in the way they have been stated, suffer
from a lack of coordination between the DSO and the TSO.
The current paper introduces a new interaction model fixing
this issue. The model is based on dynamic access bounds to the
network changing through the day and computed by the DSO.
These bounds are computed using baseline proposals from the
grid users at the medium voltage level. This new interaction
model is implemented in DSIMA [1] and compared to the
previous safest proposal, an interaction model restricting the
grid users to a very restrictive but safe access range.
The paper is organized as follows. Section II reviews the relevant literature. A summary of the agent-based testbed DSIMA
is given in Section III. The coordination issue arising in the
interaction models analyzed in [1] is explained in more details
in Section IV, which also describes a new interaction model
that solves this issue. Section V describes the implementation
of these interaction models in DSIMA. Test results are reported
in Section VI, and Section VII concludes.
II. L ITERATURE REVIEW
The common method to guarantee that distribution networks
can be safely operated is to restrict the access of distributed
generation to the distribution network. The power systems literature already contains many solutions to operate the network
considering the curtailment of distributed generation units
[2][5]. Regulated non-firm generation access contracts are
introduced in [3] where units are subject to curtailing and/or
shedding, depending on the requirements from the network
operation. The DSO may define the operating margins of the
units with the latter contract to reach a safe operating point.
A method to determine these margins is proposed and demonstrated in [3]. Article [2] reviews renewable energy curtailment
schemes and principles of access contracts. Three potential
strategies for congestion management based on the generic
interaction model of [4] are presented in [5]: distribution
grid capacity market, advance capacity allocation, dynamic
grid tariff. The authors provide a qualitative analysis of their
models. Their model advance capacity allocation is close to
the interaction model 6 proposed in this paper.
On the demand side, flexibility is mostly obtained through
dynamic pricing [6] or direct control of the loads [7], [8].
Both methods have drawbacks. Dynamic pricing exposes the
end-user of the load to volatile prices, which may not be

appropriate for retail consumers [9]. Dynamic pricing also

leaves uncertainty on the quantity of flexibility that is actually
deployed and therefore on the cost of this flexibility. This
uncertainty does not arise with the direct control of the
load but requires direct access to the controlled loads which
implies challenging control issues and intrusive equipment.
One alternative to these two methods is to rely on dynamic
fuses at the network connections and leave to the connected
equipment the optimization of its consumption in order to
satisfy the power limit of the fuse [9]. An aggregator may
control the fuse limit in case the total net load should be
limited to ensure the well functioning of the network [10].
This kind of load control perfectly fits with the interaction
model presented in Section IV.
III. D EFINITION OF INTERACTION MODELS AND THEIR
IMPLEMENTATION IN DSIMA
This section recalls the main elements of the system
modeled in DSIMA. More details can be found in the full
publication [1]. The analysis is centered on the short-term
operation of a medium-voltage network. In the scope of this
paper, we consider the short-term flexibility exchanges in an
operational planning phase. The actors simulated in DSIMA
are the DSO, the TSO, producers and retailers, and may
fulfill more than one role, depending on the interaction model.
The roles considered are balance responsible parties (BRP),
flexibility services providers (FSP) and flexibility services
users (FSU). To operate its network, the DSO may act as a
FSU. The TSO is, from the point of view of this simulation,
a FSU with given needs of flexibility services. Producers and
retailers both act simultaneously as BRPs, FSPs and FSUs.
The interaction models implemented in DSIMA follow the
procedure below.
1) DSO analysis and TSO imbalance settlement are based
on reference baselines for every bus. Right after the
clearing of day-ahead energy market, BRPs submit their
reference baselines to the DSO and the TSO.
2) The DSO assesses the state of the system and announces
its flexibility needs to the FSPs. Other FSUs also announce their needs to the FSPs.
3) FSPs provide flexibility offers sequentially to each potential FSU, the first FSU being the DSO.
4) FSUs contracts some flexibility offers for their needs.
5) Closer to real-time, FSUs request the activation of their
flexibility services.
6) Right before real-time, each FSP optimizes its realization taking into account the request of its FSUs, the
penalty incurred if a flexibility service is not provided
and its imbalance with respect to its BRPs.
7) The distribution network is operated using these realizations taking last resort actions if necessary, such as
shedding buses. If such actions are needed and FSPs
did not provide their service to the DSO, the FSPs are
penalized at a regulated tariff.
These interactions between the actors of the system are summarized in Fig. 1. The courses of the events is described from

left to right going from short-term interactions to settlement.

Each type of actor is represented by a horizontal line. Each
vertical arrow represents an interaction between two types
of actor. The actor on the head of the arrow receives the
information from the ones at the tails of the arrow indicated by
the circles. The flexibility platform is an intermediary between
FSUs and FSPs to exchange flexibility information.
A grid user makes first a request to the DSO to get access
to a range [ga,n , Ga,n ] : ga,n 0 Ga,n . Based on
these requested access bounds, the DSO computes safe access
bounds [ba,n , Ba,n ] : ba,n 0 Ba,n . The safe access
range ensures that no congestion occurs if every grid user
accesses to the network in the limits [ba,n , Ba,n ]. We assume
that there is an access contract between a BRP a and the DSO
for each bus n the BRP has access to. A contract specifies a
full access range [la,n , La,n ] and a wider flexible access range
[ka,n , Ka,n ], cf. Fig. 2. Note that these bounds are constant
with respect to time. The BRP can produce or consume
without any restriction within the full access range [la,n , La,n ],
but in the flexibility intervals [ka,n , la,n ] and [La,n , Ka,n ] the
DSO can order a restriction of the production or consumption
if necessary. Following the practice of electrical European
market design, it is the responsibility of the grid user to take
into account its technical constraints to be able to satisfy the
requests of the DSO. The link between [ba,n , Ba,n ], [la,n , La,n ]
and [ka,n , Ka,n ] is particular to each interaction model [1].
One simple method which is adopted for the simulation and
was proposed in [1] is described in Section V. A more detailed
definition of this procedure is outside the scope of this paper.
We strongly encourage the reader to get more information
about the agent-based model DSIMA, designed to study
quantitatively interaction models for the exchange of flexibility
within the distribution network, in paper [1].
IV. N EW INTERACTION MODEL WITH DYNAMIC RANGES
In the previously proposed interaction models, stakeholders
are free to exchange flexibility services. However, if the profile
of a BRP is within the flexibility intervals, it has to propose
flexibility offers to the DSO via FSPs. The DSO acts as a
standard FSU to contract and to activate the flexibility services.
Since every FSU activates flexibility simultaneously, some
actions may counterbalance the action of the DSO. Fig. 3
illustrates this issue on a small example where an action of
the TSO counterbalances and action of the DSO which was
supposed to solve a congestion. Assume that the flow exceeds
the capacity of line 3 by 1 MW. To solve this issue, the DSO
curtails a wind turbine by 1 MW. In the same time, assume that
the TSO asks a storage unit to increase its production by 0.4
MW. The simultaneous activation of these flexibility services
leads to a remaining congestion of 0.4 MW even though every
agent stays within its access bounds and satisfies its flexibility
request. We can distinguish two situations:

The storage unit is in its safe range: the DSO anticipates

that the profile of the storage unit could deviate from its
baseline, for instance to provide a service to the TSO;

Producers
& Retailers
Dynamic
Ranges

Flexibility
contracts

Global
baselines

Baseline
proposals

Local
baselines

TSO
Flexibility
offers

Settlement

DSO
Flexibility
activation
requests

Flexibility
needs

Flexibility
platform
Time

Fig. 1: Timeline of the interactions between the agents. The full lines correspond to interactions of the original DSIMA
interaction models. The dotted lines correspond to new interactions of the proposed interaction model.
ga,n

Ga,n

Requested range
ba,n

la,n
ka,n

Ba,n

Safe range
0
Full access range

Power
La,n

Flexibility intervals

Ka,n

Fig. 2: Definition of access bounds.

Fig. 3: Illustration of a flexibility activation coordination issue

between the DSO and the TSO.

The storage unit is in a flexibility interval: this flexibility

interval is limited so that no flexibility activation can
counterbalance the DSO actions.
The contribution of this paper lies in the proposal and the
analysis of an interaction model with dynamic access ranges
solving the previous issue. These ranges change each quarter
based on the distribution network limitations computed by the
DSO. These dynamic ranges are denoted [da,n,t , Da,n,t ], and
are constrained to be larger than the full access range. The
procedure to obtain the dynamic ranges is the following.
1) BRPs provide baseline proposals that may belong to the

full access range or the flexibility intervals.

2) Based on these proposals, the DSO computes the dynamic ranges so that its network is secure and communicates them to the BRPs.
3) BRPs submit new baselines, constrained to lie within
the dynamic ranges, to the DSO and the TSO.
4) These baselines are used as reference for the provision
of flexibility services. If the realization of a BRP violates
the dynamic ranges, the BRP is penalized at a regulated
tariff higher but of the same order of magnitude than the
imbalance price.
These additional steps are represented by the dotted lines in
Fig. 1. Coming back to our example, assume that the wind
turbine has the safe range [0, 5] MW and provide the baseline
proposal of 7 MW. The storage unit has a flexible range of
[1, 1] MW and a baseline proposal of 0 MW. To solve the
congestion problem, the DSO grants the dynamic range [0, 6]
MW to the wind turbine and [1, 0] MW to the storage unit.
As a result, the TSO cannot ask the storage unit to increase
its production and the DSO prevents the congestion.
Note that a baseline proposal is mandatory to discourage a
grid user to produce or consume outside of its dynamic range
established on a definitive baseline. For instance, assume that
a producer has a flexible range of [5, 8] MW in one bus and
decides to produce 8 MW, so that its baseline its 8 MW. After
its computations, the DSO grants the dynamic range [0, 6] MW.
The producer has to choose between paying an imbalance of 2
MW to the TSO or paying a penalty to the DSO. Now assume
that the producer may sell a flexibility service to another BRP
for a downward modulation of 1 MW, modulation for which
the producer is remunerated. The producer has incentive to
produce 7 MW, provide the flexibility service and pay only

for 1 MW of penalty to the DSO. Note that the producer is

not responsible for any imbalance with this decision.
V. DYNAMIC ACCESS RANGES COMPUTATION
Prior to the day by day simulation of the system, the DSO
computes the access bounds [ba,n , Ban ] to its system for each
agent a and each bus n. We opt for the following procedure,
which is performed only once for the whole simulation range,
e.g. one year. The bounds for each bus and each agent
[ba,n , Ba,n ] are computed by solving the problem (1), which
provides bounds that are as close a possible from the bounds
that were requested by each agent, [ga,n , Ga,n ]. These bound
are obtained by minimizing the sum of the auxiliary variables
g and G , representing the deviation from the requests, taking
into account the operational limits of the system. Here, we
consider the line capacities only in a network flow problem.
The extension of the procedure to the classic optimal power
flow problem considering reactive flow and voltage constraint
would follow the same principle but would lead to non-linear
optimization problems. Worst case conditions are considered
through the auxiliary variables f n,m and f n,m , but actual
power flows are not. Note that in a real system, this procedure

where a,n,t is a parameter quantifying the credible relative

deviation. If ppa,n,t is in the flexible access range, we set
pa,n,t = pa,n,t = ppa,n,t . Two additional parameters, dmax
a,n,t
max
and Da,n,t
give amount of curtailable power by the DSO
and are computed by
p
dmax
a,n,t = max{0, la,n pa,n,t }

(3a)

= max{0, ppa,n,t

(3b)

max
Da,n,t

La,n }

Together with the dynamic ranges, the DSO computes its

+

upward and downward flexibility needs as a FSU, rn,t

and rn,t
by solving optimization problem (4). These needs are computed as the maximum needs in each scenario. The positive
parameters n+ and n characterize the flexibility potential at
each bus and are obtained from the DSO knowledge of the
system (this may for instance be part of the access contract
procedure). The dynamic ranges are built using the solution
min

X

+ n rDSO,n,t
n+ rDSO,n,t

(4a)

nN

subject to, n, n0 N 2 , t T ,
Cn,n0 f n,n0 ,t , f n,n0 ,t Cn,n0

min g + G

(1a)

Cn,n0 f n,n0 , f n,n0 Cn,n0

(1b)

subject to, n, n N ,

n N

n N , t T ,
X
X
+
f n,n0 ,t
(pa,n,t Da,n,t + da,n,t ) r
n,t + r n,t =

ba,n =

aA(n)

X
aA(n)

Ba,n =

f n,n0

(1c)

n0 N (n)

aA

(4c)
X

(4b)

(pa,n,t Da,n,t + da,n,t )

r
n,t

r+
n,t

(4d)

n0 N (n)

f n,n0

f n,n0 ,t

n0 N (n)

aA

(1d)

n0 N (n)

n N , a A(n)

rn,t
+
rn,t

r
n,t , r n,t
+
rn,t , r+
n,t

(4e)
(4f)

a A, n N , t T ,

g (ga,n ba,n )/ga,n

(1e)

G (Ga,n Ba,n )/Ga,n

(1f)

ga,n ba,n 0 Ba,n Ga,n

da,n,t dmax
a,n,t
Da,n,t

max
Da,n,t

(4g)
(4h)

(1g)
of optimization problem (4) as

would not be performed in one round, since access contracts

are delivered within a certain delay after they were requested.
The aggregation of contracts of the low voltage contributors
should also be considered. A more realistic simulation would
thus consider the legacy of access contracts.
In the day by day simulation of the system, the DSO uses
the local baseline proposals to compute the dynamic access
ranges [da,n,t , Da,n,t ] of each agent a for each period t and bus
n. To compute the dynamic ranges, the DSO first determines
credible scenarios of deviation from the baseline proposal
ppa,n,t , denoted pa,n,t and pa,n,t . If ppa,n,t is in the safe access
range,
pa,n,t = max{la,n,t , ppa,n,t a,n,t |pba,n,t |}

(2a)

pa,n,t = min{La,n,t , ppa,n,t

(2b)

a,n,t |pba,n,t |}

da,n,t = min{la,n , ppa,n,t + da,n,t }

Da,n,t = max{La,n , ppa,n,t Da,n,t }.

(5a)
(5b)

To obtain the final baselines communicated to the DSO and

to the TSO, BRPs solve the same optimization problem they
used to get their baseline proposal taking into account these
dynamic ranges.
VI. R ESULTS
The interaction models are tested on a generic 11 kV
distribution network composed of 75 buses and hosting 22
distributed generation units [11]. Two interaction models are
simulated under the same conditions for 365 days individually,
which amounts to 32 hours of simulation by model. Each
day is divided in 24 periods and simulated independently.

Production data of the 22 generation units are taken from

a production curve of 2013 scaled such that the maximum
production reaches 64.2 MW and the average is 16 MW.
Distributed generation units are clustered into three portfolios,
each managed by a different producer. The consumption of
the 53 connected loads is built from the Belgian total load
consumption of 2013 scaled to a mean of 10.8 MW and a
maximum of 21 MW. This consumption is divided in three
parts, each belonging to one retailer. The consumption of a
retailer is divided in a static part and a flexible part respectively
accounting for 80% and 20% of the total consumption of the
load. We assume that the TSO aims to contract a volume of
flexibility equal to 2% of the total installed production capacity
of the system in each period. The total activation request of
the TSO is drawn using a zero mean Gaussian distribution
with a variance equal to the target flexibility volume of the
TSO. Energy prices are taken from the clearing of 2013 of the
Belpex day-ahead energy market scaled to a mean of 53.64
e/MWh and a maximum of 93.03 e/MWh, excluding the
extreme 2.5% of the original data. The value of lost load taken
into account in the case of shedding due to the tripping of a
protection is set to 500 e/MWh for the production and 1000
e/MWh for the consumption. The imbalance prices come from
the Belgian TSO. A FSP not providing a contracted flexibility
service or violating its dynamic range is penalized at 150%
of the maximum imbalance price. The experiments are carried
out on a computer equipped with an Intel Core i7-3770 CPU
at 3.40 GHz with 32 GB of RAM. The optimization problems
are solved with SCIP.
First of all, we compare the proposed interaction model
referred to as dynamic baseline with respect to the safest
model of article [1], the Model 2, referred to as restrictive
in this article. An example of results is given in Table I
where the parameters a,n,t , defining the relative deviation
from the baseline proposals, are arbitrarily fixed to 20%. Table
I provides daily results obtained from the simulation of the
whole year. In this table, the welfare is defined as the sum of
the surpluses and the costs of all actors minus a shedding cost
proportional to the value of lost load. These results show a
significant growth of the welfare of 47%. This growth comes
from an increase of the total production by 56% which directly
impacts the revenues of the producers. The dynamic baselines
interaction model requires shedding about 122 MWh with
a maximum of 11.48 MWh on one day with the parameter
a,n,t = 20%. This amount of shedding is the result of actors
deviating from their proposal in the safe access range. Note
that they are free to deviate as they wish within this range, in
this case even more than the 20% security margin of the DSO.
To take into account these additional deviations, the DSO can
easily use some margins and increase the relative deviation it
considers. Fig. 4 shows that taking values greater than 30%
avoids the shedding issues. Fig. 5 depicts the impact of the
relative maximal deviation parameter on the welfare and the
total production with respect to the restricted model. Even
though the total production increases as a,n,t decreases, the
welfare reaches its maximum around a,n,t = 25% with a

welfare increased by 47.57%. Fig. 5 shows that the welfare is

not as sensitive with respect to a,n,t as we would expect.
Therefore, a practical choice for this application would be
a,n,t = 40% to make sure to avoid the shedding issues
without notably impacting the welfare. With this value of the
relative deviation, the welfare would be increased with respect
to the restricted model by 47% and the total production by
55%.

Fig. 4: Evolution of the yearly shed production with the

relative maximal deviation parameter considered by the DSO.

Fig. 5: Welfare and production increase as a function of the

relative maximal deviation parameter considered by the DSO.

VII. C ONCLUSION
This article proposes an interaction model organizing the
interactions between the actors of the distribution system
exchanging flexibility. This new interaction model is based
on dynamic access bounds to the network changing throughout the day and computed by the DSO. These bounds are
computed using baseline proposals from the grid users at the
medium voltage level and depends on an estimation of the
DSO of the maximal deviation of the realization of the actor
with respect to its proposal. This new interaction model is
implemented in DSIMA and compared to the previous safest
proposal, an interaction model restricting the grid users to a

TABLE I: Comparison of the interaction models on one year.

Welfare

Model restricted
min
mean
max
-7727.27 27410.94 68434.42

Units
e
e
e
e
e
e

Shedding costs
DSOs costs
TSOs surplus
Producers surplus
Retailers surplus

0
0
831.52
0
-14750.43

0
0
2878.34
24005.32
527.28

0
0
3849.78
68265.06
9852.33

0
0
831.52
0
-14750.43

167.31
0
2877.74
37188.73
527.22

5740.86
0
3849.78
151426.68
9852.33

Total production
Total consumption

0
-489.93

247.33
-257.07

622.26
-39.95

0
-489.93

385.42
-257.07

1333.51
-39.95

MWh
MWh

Total imbalance
Max. imbalance
Total usage of flex.
Total energy shed

0.15
-0.6
0
0

1.4
-0.13
14.94
0

5.11
0.3
65.1
0

0.15
-7.09
0
0

1.73
-0.36
14.92
0.33

15.16
0.3
65.1
11.48

MWh
MW
MWh
MWh

very restrictive but safe access range. Results show that this
new model achieves better performances than the restrictive
model, allowing to safely increase by 55% the amount of
distributed generation in the network and the welfare by
42.5%.
The work presented in this paper could be extended along
several lines, for instance by computing fair access ranges
to the network taking into account the current system situation and future evolution; refining the modeling level by
considering more detailed agents, considering the analysis
in a stochastic environment and alternating current power
flow equations; studying dynamics of the system, such as
the entry or exit of new players or production units; and
finally comparing the proposed interaction models to network
reinforcement decisions without change of interaction model.

[4]

[5]

[6]

[7]

ACKNOWLEDGMENTS
This research is supported by the public service of Wallonia
Department of Energy and Sustainable Building within the
framework of the GREDOR project.

[8]

R EFERENCES
[1]

Model dynamic baselines

min
mean
max
-7726.54 40426.38 151596.04

S. Mathieu, Q. Louveaux, D. Ernst, and B. Cornelusse,

Dsima: A testbed for the quantitative analysis of interaction models within distribution networks, Sustainable Energy, Grids and Networks, vol. 5, pp. 78 93,
2016, ISSN: 2352-4677.
[2] L. Kane and G. Ault, A review and analysis of renewable energy curtailment schemes and principles of
access: Transitioning towards business as usual, Energy
Policy, vol. 72, pp. 6777, 2014.
[3] R. Currie, G. W. Ault, C. Foote, and J. R. McDonald, Active power-flow management utilising operating
margins for the increased connection of distributed generation, IET generation, transmission & distribution,
vol. 1, no. 1, pp. 197202, 2007.

[9]

[10]

[11]

K. Heussen, D. E. M. Bondy, J. Hu, O. Gehrke, and L.

H. Hansen, A clearinghouse concept for distributionlevel flexibility services, in 4th Innovative Smart Grid
Technologies Europe, IEEE/PES, 2013.
P. Bach Andersen, J. Hu, and K. Heussen, Coordination strategies for distribution grid congestion management in a multi-actor, multi-objective setting, in 3rd
Innovative Smart Grid Technologies Europe, IEEE/PES,
2012.
L. Merciadri, S. Mathieu, D. Ernst, and Q. Louveaux,
Optimal assignment of off-peak hours to lower curtailments in the distribution network, in 5th Innovative
Smart Grid Technologies Europe, IEEE/PES, 2014.
S. Mathieu, D. Ernst, and Q. Louveaux, An efficient
algorithm for the provision of a day-ahead modulation
service by a load aggregator, in 4th Innovative Smart
Grid Technologies Europe, IEEE/PES, 2013.
E. Georges, B. Cornelusse, D. Ernst, Q. Louveaux,
V. Lemort, and S. Mathieu, Direct control service
from residential heat pumps aggregation with specified
payback, in Proceedings of the 19th Power Systems
Computation Conference (PSCC), IEEE, 2016.
S. S. Oren, A historical perspective and business model
for load response aggregation based on priority service,
in System Sciences (HICSS), 2013 46th Hawaii International Conference on, IEEE, 2013, pp. 22062214.
K. Margellos and S. Oren, A fuse control paradigm
for demand side management: Formulation and stochastic pricing analysis, in American Control Conference
(ACC), 2015, IEEE, 2015, pp. 56725677.
SEDG, Uk generic distribution system (ukgds) project,
2010. [Online]. Available: www.sedg.ac.uk.