J. Mod. Power Syst.

Clean Energy (2017) 5(1):105–116

DOI 10.1007/s40565-016-0253-0

Green neighbourhoods in low voltage networks: measuring impact

of electric vehicles and photovoltaics on load profiles
Laura HATTAM1, Danica Vukadinovic GREETHAM1

Abstract In the near future, various types of low-carbon 1 Introduction

technologies (LCTs) are expected to be widely employed
throughout the United Kingdom. However, the effect that From 2005 onwards, the national electricity demand in
these technologies will have at a household level on the the UK and other developed countries has stagnated or
existing low voltage (LV) network is still an area of even decreased, despite the population increase. The UK
extensive research. We propose an agent based model that energy statistics show that total electricity consumption
estimates the growth of LCTs within local neighbourhoods, year on year has reduced on a UK, GB and south-east level
where social influence is imposed. Real-life data from an [1]. The current predictions are that the UK domestic
LV network is used that comprises of many socially electricity demand will continue to decrease in the next ten
diverse neighbourhoods. Both electric vehicle uptake and to fifteen years [2] due to the better efficiency of electrical
the combined scenario of electric vehicle and photovoltaic appliances and lighting [3, 4]. The anticipated addition to
adoption are investigated with this data. A probabilistic the domestic demand will come mostly from the new
approach is outlined, which determines lower and upper builds and the low carbon technologies (LCTs) employed
bounds for the model response at every neighbourhood. in existing buildings.
This technique is used to assess the implications of modi- In particular, the electrification of transport and heating,
fying model assumptions and introducing new model fea- which is forecasted for the near future, are predicted to be
tures. Moreover, we discuss how the calculation of these the main contributors to the changes in the electricity
bounds can inform future network planning decisions. demand. The whole picture is made more complex by the
variability of renewable energy sources, which result in
Keywords Agent based modelling, Low voltage networks, new peaks and troughs forming in the aggregated con-
Electric vehicles, Photovoltaics sumption. However, not all the projected changes threaten
to worsen this situation. A big mitigating factor is energy
storage, which can help smoothen generation and demand,
as well as offer cost efficient local solutions. Our proposed
model can estimate future loads at the feeders, and as a
CrossCheck date: 18 November 2016
result, street level storage solutions can be appropriately
Received: 6 September 2016 / Accepted: 18 November 2016 / employed.
Published online: 30 December 2016 We are interested in measuring the combined impact of
Ó The Author(s) 2016. This article is published with open access at electric vehicles (EVs) and solar panels on low voltage
Springerlink.com
(LV) networks. Several possible issues that might arise
& Laura HATTAM
l.hattam@reading.ac.uk
from the described smart grid developments are already
recognised, these include frequent peak loads that reduce
Danica Vukadinovic GREETHAM
headway, voltage drops and phase unbalance. Due to the
d.v.greetham@reading.ac.uk
complexity of human societies, any predictions on the
1
Department of Mathematics and Statistics, University of uptake of EVs and photovoltaics (PVs) comes with large
Reading, Reading RG6 6AX, UK

123
106 Laura HATTAM, Danica Vukadinovic GREETHAM

uncertainties. Different models that determine uptake larger data sets based on trials become available. Focused
already exist and are used by network planners. However, mostly on LV networks impact, in [7] the authors created
by their nature it is quite difficult to validate these models, generic local networks to assess the neighbourhood impact
and to decouple the influence of different modelling of EV charging. Using a realistic distribution network
parameters. simulation, in [8] the authors evaluate a range of different
Our contribution is two-fold. Firstly, we present an residential EV charging strategies, highlighting their
agent based model of load profiles for the uptake of LCTs strengths and weaknesses.
in local neighbourhoods when social influence is present. In [9], the authors aimed to measure the impact of PVs
Our neighbourhoods are based on real-life LV networks on LV networks in New Zealand. They were looking in
containing multiple substations and feeders. This model particular at over-voltage and overload of conductors and
uses a sample of realistic EV and PV profiles to simulate transformers. They created a power-flow model of a LV
future uptake. Secondly, we demonstrate techniques that network and simulated varying percentages of PV uptake.
allow for a thorough mathematical analysis of results. PV was based on a specific installation with an output
Probabilistic methods based on multiple simulations enable power of 3.7 kW. These uniform PVs were then randomly
the calculation of upper and lower bounds for the model distributed through different parts of the LV network
response, which we refer to as confidence bounds. These classified as rural, urban, industrial and city. Their results
bounds are used to understand the inner-workings of the showed that only very high PV penetration (over 45%)
model and to measure the effects of introducing/changing caused an overload of conductors, and in most cases
the model’s parameters. In particular, these bounds are overvoltage was not much higher than the existing statu-
used here to quantify the difference between applying a tory limit. In [5], the major technical impacts of small PV
clustered or random initial LCT distribution amongst our installations were discussed. This included excessive
sample population. reverse power flow, overvoltages along distribution feed-
The outline of the paper is the following. In Section 2 ers, increased difficulty of voltage control, increased power
we give an overview of the recent relevant results. In losses (caused by reverse power flow) and severe phase
Section 3 the model is described in detail, as well as the unbalance.
data that is used for the initialisation and calibration of the A microgrid case-study from a neighbourhood in
model. The focus is initially EV adoption only. In Section 4 Utrecht in Netherlands, looking at the combination of PVs
the confidence bounds creation is explained with some and EVs throughout a year was described in [10]. Based on
simulation results shown. Then in Section 5 the adaptation simulations, and using February demand projected over the
of the model to include socio-demographic information is whole year, the authors compared several control algo-
discussed. As well, confidence bounds are used to assess rithms. Their results showed a potential for relative peak
the impact of this new model feature. Next in Section 6 the reduction and increased self-consumption when using
model is further modified to investigate the combined smart charging and vehicles to grid technology.
uptake of EVs and PVs. Again, confidence bounds are In [11], Monte-Carlo simulations were used to measure
computed to determine the effect of changing our model the impact of several low carbon technologies, including
assumptions. Finally in Section 7 we discuss the implica- EVs and PVs. Similar to our approach, the authors used a
tions of our results and their possible use in design, plan- realistic LV network with 7 feeders and sampled from
ning and policy. realistic profiles for load and for LCTs. Note that the net-
work area examined in our paper is significantly larger,
with 44 feeders considered. Their focus is on identifying
2 Previous work thermal and voltage problems in different feeders. While
they use a random allocation of LCTs, we compare a
There is a fast growing amount of literature [5, 6] random allocation with a clustered one using socio-demo-
concerning the different impacts EVs and PVs will have on graphic information, although here the focus is the load
the future power grids. More specifically, these studies profile impact.
concentrate on load profiles, system losses, voltage pro-
files, phase unbalance, harmonic and stability impacts. 2.2 Agent based modelling of PVs and EVs uptake
Here, we focus on load profiles within a LV network.
In [12], a simple agent-based model of EV uptake was
2.1 Impact of PVs and EVs to LV networks detailed and their impact on a local grid was examined.
Their analysis was based on governmental scenarios of
For EVs, most of the existing work is based on predic- future UK EV uptake and a small pilot project that gave
tions, simulations or small pilot trials. Only recently have incentives to participants to charge overnight. As expected,

123
Green neighbourhoods in low voltage networks: measuring impact of electric vehicles and... 107

having a variety of EV charging patterns helped to reduce 120

the peaks, as opposed to when all the domestic charging 100

Feeder population
occurred after work and overnight. By then comparing
random and clustered uptake simulations, it was shown that 80

some local grids might see a substantial increase of peak 60

loads faster than expected. In [13], the impact of different
40
EV charging behaviours on the electricity grid was as well
considered, except their area of study was Western Aus- 20
tralia. They concluded that by promoting off-peak charg-
0 5 10 15 20 25 30 35 40 45
ing, the effect of EV uptake on the grid would be
Feeder number
minimised.
An agent-based model using San Francisco as a test city Fig. 1 Feeder populations
was presented in [14]. The paper considered how different
policies and battery technologies might affect the uptake We have three data sets that were created using metered
and usage of EVs. The model included a set of agents with data from this LV network. This information was collected
socio-demographic properties, attitudes and an EV on Thursday the 15th of January 2015 (winter), Thursday
ecosystem that considered the cost of gas/electricity, the 7th of May 2015 (spring) and Thursday the 9th of July
rebates and public charging stations. Each three months, 2015 (summer). The data sets consist of a combination of
agents assessed whether they needed a new vehicle. Based metered and predicted daily demand energy profiles (kWh)
on their properties, attitudes and state of their social net- for every household, where a genetic algorithm was used to
work, they acquired (or not) an EV to use for their daily allocate monitored endpoints to unmonitored customers
commute. The social network was created randomly based [15]. These profiles have readings every half hour and
on similarities in age, income and residential locations of therefore for each household we have a load profile as a
agents. This enabled the exploration of different scenarios vector of length 48. Throughout this paper, we refer to
(for example, increasing or decreasing rebates for EVs and these three data sets as ‘baseloads’.
increasing battery sizes) and then looking at the impacts on Thursdays are of particular interest here since presum-
the average daily load. ably most household members will be at home and there-
fore, the demand will be maximised. As well, if they
possess an EV, we assume Thursdays are a likely time to
3 Agent based model charge, just before the weekend.
Initially, the focus here is the clustered allocation of EVs,
The impact of future LCT adoption is predicted with our although later the combination of EV and PV uptake is
agent based model that applies a clustered distribution of investigated. The EV charging profiles used in our model
technologies to a sample UK population (Bracknell, UK). were generated during the 55 week trial conducted by My
The clustering follows the Joneses effect such that house- Electric Avenue [16], where the number of participants
holds are influenced by their neighbours’ choices through increased as the trial progressed. These profiles consist of the
observation, which is one of the causes for the development two values ‘0’ and ‘1.85’ kWh, which represent when the EV
of ‘green neighbourhoods’. This means a household is is not charging and charging respectively. They have read-
more likely to acquire a LCT if their neighbour already has ings also every half hour. Three days from this trial are
one. Here, we model EVs and PVs as they are visible from selected, which are Thursday May the 8th 2014 (week 16 of
the street, and can be observed by a neighbour. the trial), Thursday July the 10th 2014 (week 25 of the trial)
Our network is based on a realistic LV network situated and Thursday January the 15th 2015 (week 52 of the trial).
in Bracknell, UK, which comprises of 44 feeders. The These dates are chosen since they correspond seasonally to
household population at each feeder varies considerably. the baseload dates. There are 79 households that consistently
Figure 1 demonstrates the variation in feeder size, where participate during weeks 16–52 of the trial and therefore, we
the feeder number and feeder population is given. The have 79 EV daily profiles that are representative of winter,
feeders have been sorted according to feeder size. Note that spring and summer charging behaviour. As an example, the
each feeder corresponds to one neighbourhood and all EV profiles for one randomly selected household on the
households along a particular feeder are considered chosen days are shown in Fig. 2.
neighbours. Overall, there are 1841 properties, where 7 are Note that EV profiles will have variability within each
households with PVs installed and 71 are commercial season. However, here we assume that by taking a snapshot
properties. of charging behaviour on Thursdays occurring at the same

123
108 Laura HATTAM, Danica Vukadinovic GREETHAM

2.5 1

Energy unit (kWh)

2.0
1.5
5 2
1.0
0.5
0 4
5 10 15 20 25 30 35 40 45 50 3
Half hours
(a) Red circles signify EV allocation and the numbered cir-
Week 16 (Spring); Week 25 (Summer); Week 52 (Winter) cles correspond to eligible households (The connecting lines in-
dicate neighbours)
Fig. 2 EV profiles for one household
0.40
0.35
time of year as the baseloads, the model result will be more

Probability of selection
realistic. 0.30
All profiles are depicted in energy units (kWh). To 0.25
instead consider power units (kW), all profiles must be 0.20
multiplied by 2 (since the time step is 0.5 hours). This
0.15
gives the average load over 30 minutes.
The following outlines the clustering algorithm applied 0.10
to forecast EV uptake. 0.05
0
1) Firstly, we establish the percentage of households in 1 2 3 4 5
the sample population that will adopt EVs and the Household number
number of years it will take (This is set to be 8 years). (b) Probability of EV assignment by the random
number generator
2) Next, an initial random distribution of EV seeds is
performed to simulate the first year of EV uptake. Fig. 3 Illustration of selection process
3) Then, during the remaining years, EVs are assigned to
households according to the score s (refer to (2)).
selection process with a simplified network, comprising of
4) The number of EV households (households that
red and numbered circles that represent EV and eligible
adopted an EV) increases linearly every year until
households respectively, with their probability of EV
the specified amount is attained.
assignment by the random number generator also shown.
5) Lastly, EV profiles are assigned to the EV households,
This figure suggests that household 1 is the most probable
where
to acquire an EV due to observing 2 of its 3 neighbours
EV household profile ¼ base load þ EV profile ð1Þ
with EVs.
It is important to note that all 71 commercial properties in Once a household is selected, they become an EV
our data set never receive LCT load since our focus is LCT household for the remaining years of the simulation, with
household uptake. As well, there is one feeder comprised s updated every year. Using s to inform EV allocation leads
of only commercial properties, thus this site is always to clusters of EVs forming around the initial seeds.
given zero LCT load. Therefore, we are modelling the formation of green
In 2015, OFGEM (a UK regulatory authority for gas and neighbourhoods due to social influence. This method is an
electricity markets) increased the period covered by a adaptation of the algorithm proposed in [12], which was
single price control review to 8 years [17]. As detailed also applied to model EV uptake.
investment decisions must be outlined in this time horizon, There is an assumed link between increased neigh-
we also use 8 years in our simulations. bourhood diversity and a heavily populated feeder. As a
The score, s, assigned to eligible households is the result, when transforming these larger sites into greener
percentage of PVs and EVs in its neighbourhood presently, neighbourhoods, the impact from one EV household should
where be comparatively small. To account for this, s depends
  upon the feeder population and therefore, the influence of
Number of neighbours with an EV and=or PV
s ¼ 100 one household on its neighbours is relative to the neigh-
Number of neighbours þ 1
bourhood size.
ð2Þ Note that when a household is given an EV, the EV
This score is proportional to the probability of selection by profile assigned to them is randomly selected from 79
a random number generator. Figure 3 illustrates this possible profiles. If the baseload applied is representative

123
Green neighbourhoods in low voltage networks: measuring impact of electric vehicles and... 109

of spring, summer or winter then the EV profile chosen will comparable EV loads, where feeders 15, 17, 39 and 40
also correspond to spring, summer or winter respectively. have 25, 27, 82 and 86 households respectively. Further-
more, less populated feeders have greater EV peaks,
demonstrated by the blue curves. This can be attributed to
4 Confidence bounds increased neighbourhood diversity when the feeder popu-
lation is larger and therefore, it is more difficult to
For some fixed model parameters, there are many fea- influence your neighbours and form an EV majority. Also,
sible outcomes. This is due to the initial random distribu- it is evident that the spread between the 10% and 50%
tion of seeds highlighting different neighbourhoods every trends is far less notable for smaller feeders. Additionally,
model run. As well, one of 79 possible EV profiles are the red curve sits along the baseload in Fig. 4a. This
randomly assigned to households, causing further variation suggests that less populated feeders do often avoid EV
in the model result. Consequently, we aim to determine assignment.
upper and lower bounds of the model response for a fixed Hence, similar to [11], multiple simulations are used to
set of parameters, which we label confidence bounds. determine the impact of LCTs on a LV network. However,
These bounds will be calculated by undertaking 500 con- here the area examined is comparatively large, social
secutive model runs and will therefore relate to the EV load influence is imposed and our focus is the load profile.
variance, not the baseload. Since the clustering is based on Moreover, this methodology enables bounds for the
neighbourhoods, which are defined by feeders, the bounds expected load at each feeder/neighbourhood to be found,
will be computed at each feeder. which are extremely informative measures for network
The following details the method used to calculate planners.
confidence bounds. Only the results for four of the 44 feeders considered by
the model have been shown. Refer to the Appendix A for
1) Specify the model parameters, which are the uptake
the depiction of the results across the entire network. In this
percentage and the number of years i.e. 30% EV
paper, 30% EV uptake has been selected. However, this is a
uptake ensures d0:3  1841e properties receive an EV
model parameter that can be varied. Furthermore, the
each simulation.
network examined here with our approach covers a fixed
2) Complete 500 simulations.
area, although, the analysis can be easily adapted to study
3) After each simulation, record the aggregate result at
networks of a greater or smaller size.
the feeder. The 44 feeders are considered together so
The number of simulations conducted, n, to calculate the
that 0%–100% of households along a particular feeder
bounds shown was n ¼ 500. Choosing n ¼ 500 was
can receive an EV each simulation.
believed appropriate since the variance was captured but
4) The aggregate data is then used to calculate 10%, 50%
the computation time was minimised. This was deduced by
and 90% quantiles at the feeder. The feeder lower and
comparing the confidence bounds at various feeders when
upper bounds correspond to the 10% and 90%
n ¼ 200; n ¼ 500 and n ¼ 1000. These results revealed
quantiles.
only minor differences between the quantiles for n ¼ 500
5) The quantile with the baseload subtracted represents
and n ¼ 1000. Therefore, considering the additional time
the variation in EV load at the feeder. Then, dividing
required to undertake 1000 simulations, it was determined
the quantiles by the number of households along the
unnecessary to perform this many runs and setting n ¼ 500
feeder, we can compare the 44 feeders and their EV
was optimal for a network of this size.
loads.
These particular four feeders were chosen for closer
In Fig. 4a, b, the results for feeders 15, 17, 39 and 40 are analysis because feeders 15 and 17 have approximately the
depicted when 30% uptake overall is assumed amongst the same number of households, similarly for feeders 39 and
44 feeders. The aggregate result at the feeder is displayed 40. As well, household numbers at feeders 15 and 17 are
(including baseload), where the black dots represent the less than the average feeder population (which is 40
response from 500 model runs. The red, green and blue households), whereas they are greater at feeders 39 and 40,
curves are the 10%, 50% and 90% quantiles respectively and therefore, two extreme cases are studied. Furthermore,
calculated from the black dots. The far right plots show the the distribution of council tax bands (see Section 5)
quantiles with the baseload subtracted, divided by the amongst these feeders varies considerably, which becomes
number of households along the feeder. Only the results important in Section 5.
corresponding to winter are given here, where the winter The simulations and confidence bounds calculations
baseload and winter EV profiles discussed in Section 3 are were performed using MATLAB on a standard workstation
used. From the far right plots of Fig. 4a, b, it is apparent with Intel i5 processor on 2.50 GHz with 8 GB RAM. The
that feeders with similar household numbers receive computation time required to conduct 500 consecutive

123
110 Laura HATTAM, Danica Vukadinovic GREETHAM

35 35 0.8
30 30 0.7

Energy unit (kWh)

Energy unit (kWh)

Energy unit (kWh)

25 25 0.6
0.5
20 20
0.4
15 15
0.3
10 10 0.2
5 5 0.1
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Half hours Half hours Half hours
90% feeder 15; 50% feeder 15 90% feeder 17; 50% feeder 17 90% feeder 15; 50% feeder 15
10% feeder 15 10% feeder 17 10% feeder 15; 90% feeder 17
50% feeder 17; 10% feeder 17
(a) Result of 500 simulations with 30% EV winter uptake (random seed allocation)
80 80 0.8
70 70 0.7

Energy unit (kWh)

Energy unit (kWh)

Energy unit (kWh)

60 60 0.6
50 50 0.5
40 40 0.4
30 30 0.3
20 20 0.2
10 10 0.1
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Half hours Half hours Half hours
90% feeder 39; 50% feeder 39 90% feeder 40; 50% feeder 40 90% feeder 39; 50% feeder 39
10% feeder 39 10% feeder 40 10% feeder 39; 90% feeder 40
50% feeder 40; 10% feeder 40
(b) Result of 500 simulations with 30% EV winter uptake (random seed allocation)
35 35 0.8
30 30
Energy unit (kWh) 0.7
Energy unit (kWh)

Energy unit (kWh)

25 25 0.6
0.5
20 20
0.4
15 15
0.3
10 10 0.2
5 5 0.1
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Half hours Half hours Half hours
90% feeder 15; 50% feeder 15 90% feeder 17; 50% feeder 17 90% feeder 15; 50% feeder 15
10% feeder 15 10% feeder 17 10% feeder 15; 90% feeder 17
50% feeder 17; 10% feeder 17
(c) Result of 500 simulations with 30% EV winter uptake (seed allocation informed by CTB)
80 80 0.8
70 70 0.7
Energy unit (kWh)
Energy unit (kWh)

Energy unit (kWh)

60 60 0.6
50 50 0.5
40 40 0.4
30 30 0.3
20 20 0.2
10 10 0.1
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Half hours Half hours Half hours
90% feeder 39; 50% feeder 39 90% feeder 40; 50% feeder 40 90% feeder 39; 50% feeder 39
10% feeder 39 10% feeder 40 10% feeder 39; 90% feeder 40
50% feeder 40; 10% feeder 40
(d) Result of 500 simulations with 30% EV winter uptake (seed allocation informed by CTB)

Fig. 4 Simulation results

123
Green neighbourhoods in low voltage networks: measuring impact of electric vehicles and... 111

model runs with this network (44 feeders consisting of The last column of Table 1 details the spread of CTBs
1841 properties) was approximately 15 minutes. within our sample population of 1841 properties. As well,
in Table 1 a comparison is given of 100 seeds that are
selected using CTB information, with j ¼ 1; 2; 3; 4, where
5 Adding socio-demographic information the distribution of CTBs for the 100 nominated households
is displayed. We propose that by setting j ¼ 4, the subse-
The simulations performed in Section 4 randomly allo- quent initial EV population reflects the survey findings
cated EV seeds. Next, we aim to improve our model by [20], since approximately 70% of seeds now have a CTB
introducing council tax band (CTB) information to instead greater than C (A  C typically represents small dwell-
inform seed distribution. Council tax is a local tax on ings). Note that j is a model parameter that can be
domestic properties in England, Scotland and Wales, varied.
introduced in 1993. Each property is assigned one of eight Confidence bounds can be used to measure the effect of
bands (A to H) based on the property’s capital value [18], changing our model assumptions. Here, we analyse the
where A corresponds to the smallest value and H to the influence of using CTB to inform the initial seed distri-
highest. Here, it is assumed that larger homes have higher bution, instead of random initial distribution. The winter
CTBs. We note here that other socio-demographic infor- results for feeders 15, 17, 39 and 40 are shown in Fig. 4c, d.
mation can be used if available. We use CTB as it is It is evident that feeders with about the same sized popu-
publicly available [19] and it allows us to identify neigh- lations are no longer given a similar EV load. The upper
bourhoods that have a higher percentage of larger bounds depicted along Fig. 4c reveal that feeder 15
properties. receives a significantly larger load than feeder 17. This is
A survey of Californian EV owners [20] revealed that due to 60% of properties along feeder 15 having a CTB
generally they owned and lived in single family dwellings greater than D, whereas for feeder 17 it is 0%. Similarly,
that had parking and space to install a charging point. They Fig. 4d suggests that feeder 39 has been assigned a greater
also had higher incomes, which typically relates to living in EV load compared to feeder 40, which is due to 54% of
larger homes. Furthermore, present EV owners commonly households along feeder 39 having a CTB of more than D,
had a PV installed at their property. Also acknowledged when feeder 40 has 0%. Furthermore, of these four feeders,
was that neighbour influence was an important factor in EV feeder 15 overall has the largest EV peak, which is a result
adoption since clusters of EV households had formed in of both feeder size and its households’ CTBs. In Fig. 4a, c:
California. This study therefore supports initialising the feeder 15 has 25 households and the number of households
seeds guided by CTB information and then imposing with CTB [ D ¼ 60%; panels a and c: feeder 17 has 27
neighbour influence to determine the growth of EV households and the number of households with
ownership. CTB [ D ¼ 0%; panels b and d: feeder 39 has 82 house-
The model is now adapted to firstly favour PV properties holds and the number of households with
and larger households, signified by higher CTBs. To CTB [ D ¼ 54%; panels b and d: feeder 40 has 86
implement this, we assign to every household an initial households and the number of households with
score, si , such that: CTB [ D ¼ 0%.
j Now that socio-demographic data has been incorpo-
100CTBhh
si ¼ ð3Þ rated, higher potential peaks are exhibited at certain feeders
8j
than previously predicted. This modelling suggests that
where CTBhh ¼ 1; 2; . . .; 8 when the household’s CTB is
A; B; . . .; H respectively and j is some positive integer. As
well, PV (resp. commercial) properties are given the score Table 1 CTB Distributions
si ¼ 100 (resp. si ¼ 0). The score is proportional to the CTB j¼1 j¼2 j¼3 j¼4 All
likelihood of selection by a random number generator. This
0 0 0 0 0 71
relationship is consistent with that portrayed in Fig. 3. It
A (1) 0 0 0 0 1
should be noted that si is only used during the first year
B (2) 2 2 2 0 103
when seeds are allocated, then s as given by (2), applies for
C (3) 59 47 46 31 1135
the remaining years.
D (4) 25 32 24 29 373
Choosing j determines how dependent seed assignment
E (5) 5 9 14 23 107
is on the CTB information, where CTB influence increases
F (6) 8 5 8 8 37
with j. Moreover, higher values of j will result in prominent
clusters developing in neighbourhoods with large house- G (7) 1 5 4 7 12
holds. Here we set j ¼ 4. H (8) 0 0 2 2 2

123
112 Laura HATTAM, Danica Vukadinovic GREETHAM

2.0 the initial seed is randomly distributed. Then, the seed

Generation (kWh) 1.8 allocation is guided by CTB. There are now two scores
1.6
1.4 assigned to eligible households. These are sEV and sPV ,
1.2 where both are defined using (2) and updated every year. A
1.0 household’s likelihood for EV (resp. PV) selection by a
0.8
0.6 random number generator is proportional to sEV (resp. sPV ).
0.4 The dependence of selection on these scores is
0.2 demonstrated by Fig. 3.
0 5 10 15 20 25 30 35 40 45 50 Again the 71 commercial properties within our data set
Half hours do not receive a LCT. Also, we ensure that the 7 house-
holds with PVs installed already are not allocated an
Fig. 5 Summer PV generation profiles
additional PV.
In Fig. 6a, c: feeder 15 has 25 households and the
these feeders are likely sites for future network issues. number of households with CTB [ D ¼ 60%; panels a and
Refer to the Appendix A for an overview of the results at c: feeder 17 has 27 households and the number of house-
all 44 feeders. holds with CTB [ D ¼ 0%; panels b and d: feeder 39 has
82 households and the number of households with
CTB [ D ¼ 54%; panels b and d: feeder 40 has 86
6 Combination of electric vehicles households and the number of households with
and photovoltaics CTB [ D ¼ 0%.
In Fig. 6a, b, the results for feeders 15, 17, 39 and 40 are
The additional impact of PV adoption on the electricity displayed. Here, the initial seeds have been randomly
network is now examined by adapting our model to also allocated, where EV and PV seeds are distributed sepa-
consider PV uptake. rately. The quantiles depicted reveal significant troughs
As part of the New Thames Valley Vision project, (red curve) develop during the day and large peaks (blue
surplus generation and solar radiation data was recorded at curve) form at night. Furthermore, it is evident that feeders
12 households that had PVs installed. PV daily generation of a similar size again receive comparable EV/PV loads,
profiles were then created, defined every half hour, by where the red trough and blue peak are more prominent for
assuming that PV generation is proportional to solar radi- smaller feeders, suggesting increased variability at these
ation. It should be noted that other influences, such as feeders.
ambient and surface temperatures, would also contribute to Next, in Fig. 6c, d, the results for feeders 15, 17, 39 and
PV generation, but we ignore these to simplify our model. 40 are given, where now the seed distribution is informed
Thus, for our analysis only solar radiation is used. As a by CTB information. The allocation of EV and PV seeds
result, we obtained three sets of 12 PV daily generation are again separate. Consistent with previous findings, due
profiles that were representative of spring, summer and to introducing CTB, feeders 15 and 39 have greater
winter generation. For this investigation, only the summer extremes. Interestingly, these values are roughly the same
baseload and summer PV profiles are applied to simulate for feeders 15 and 39, and hence, independent of feeder
uptake. In Fig. 5, the 12 summer PV daily generation size. This is a result of now modelling two technologies,
profiles are shown. These profiles are scaled so that the which amplifies the clustering effect. Although, the result
maximum generation is 1.9 kWh to comply with UK variability is more pronounced for the smaller feeders,
standards. When a household is given a PV, one of the 12 indicated by the spread of the quantiles. As a result of using
possible profiles are randomly selected and then subtracted CTBs, the minimum and maximum loads obtained at
from their baseload i.e. a household with an EV and a PV is feeders 15 and 39 are now larger than initially estimated
assigned the profile: (see Fig. 6a, b).
Lastly, simulations of 30% EV and 30% PV uptake with
EV þ PV household profile ¼ baseload þ EV profile
 PV profile ð4Þ CTB information are again performed, except now we
ensure that all households which receive an EV with our
The confidence bounds discussed in Section 4 are now clustering algorithm are also given a PV. The results are
used to quantify the effect of both EV and PV adoption by actually extremely similar to those depicted in Fig. 6c, d.
our sample population. Simulations for 30% EV and 30% The most significant difference observed is at feeder 17 and
PV uptake are conducted (the uptake percentage chosen is is shown in Fig. 7. This is expected as our clustering
a model parameter, which can be modified). Firstly, the method already promotes the growth of EV ? PV
clustering algorithm outlined in Section 3 is applied, where groupings.

123
Green neighbourhoods in low voltage networks: measuring impact of electric vehicles and... 113

40 40
30 30 1.0
Energy unit (kWh)

Energy unit (kWh)

0.8

Energy unit (kWh)

20 20
0.6
10 10 0.4
0 0 0.2
-10 -10 0
-0.2
-20 -20
-0.4
-30 -30 -0.6
-40 -40 -0.8
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Half hours Half hours Half hours
90% feeder 15; 50% feeder 15 90% feeder 17; 50% feeder 17 90% feeder 15; 50% feeder 15
10% feeder 15 10% feeder 17 10% feeder 15; 90% feeder 17
50% feeder 17; 10% feeder 17

(a) Result of 500 simulations with 30% EV uptake and 30% PV uptake summer (random seed allocation)
80 80
60 60 1.0
Energy unit (kWh)

Energy unit (kWh) 0.8

Energy unit (kWh)

40 40
0.6
20 20 0.4
0 0 0.2
-20 -20 0
-40 -40 -0.2
-0.4
-60 -60 -0.6
-80 -80 -0.8
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Half hours Half hours Half hours
90% feeder 39; 50% feeder 39 90% feeder 40; 50% feeder 40 90% feeder 39; 50% feeder 39
10% feeder 39 10% feeder 40 10% feeder 39; 90% feeder 40
50% feeder 40; 10% feeder 40

(b) Result of 500 simulations with 30% EV uptake and 30% PV uptake summer (random seed allocation)
40 40
30 30 0.8
Energy unit (kWh)

Energy unit (kWh)

0.6
Energy unit (kWh)
20 20
0.4
10 10 0.2
0 0 0
-10 -10 -0.2
-0.4
-20 -20
-0.6
-30 -30 -0.8
-40 -40 -1.0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Half hours Half hours Half hours
90% feeder 15; 50% feeder 15 90% feeder 17; 50% feeder 17 90% feeder 15; 50% feeder 15
10% feeder 15 10% feeder 17 10% feeder 15; 90% feeder 17
50% feeder 17; 10% feeder 17

(c) Result of 500 simulations with 30% EV uptake and 30% PV uptake summer (seed allocation informed by CTB)
80 80
60 60 1.0
Energy unit (kWh)

Energy unit (kWh)

0.8
Energy unit (kWh)

40 40
0.6
20 20 0.4
0 0 0.2
-20 -20 0
-40 -40 -0.2
-0.4
-60 -60 -0.6
-80 -80 -0.8
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Half hours Half hours Half hours
90% feeder 39; 50% feeder 39 90% feeder 40; 50% feeder 40 90% feeder 39; 50% feeder 39
10% feeder 39 10% feeder 40 10% feeder 39; 90% feeder 40
50% feeder 40; 10% feeder 40

(d) Result of 500 simulations with 30% EV uptake and 30% PV uptake summer (seed allocation informed by CTB)

Fig. 6 Simulation results

123
114 Laura HATTAM, Danica Vukadinovic GREETHAM

40 substation and feeder assignment. This allowed us to sort

30
the 1841 properties into 44 realistic neighbourhoods. Then
neighbour influence was imposed to determine uptake. The
Energy unit (kWh)

20 model also applied sets of EV and PV profiles that were

10 representative of spring, summer and winter usage. To
assess the model response, a probabilistic approach was
0
proposed that provided feeder confidence bounds, which
-10 were an upper and lower limit for the expected load at
-20 every neighbourhood. These were a result of 500 consec-
utive simulations and therefore, the bounds measured the
-30
0 5 10 15 20 25 30 35 40 45 50 variation in LCT load. Next, another aspect of social
Half hours influence was introduced with socio-demographic infor-
0.6 mation also guiding LCT selection. More specifically, we
0.4 ensured that bigger households were more likely to acquire
Energy unit (kWh)

0.2 a LCT. Confidence bounds were then utilised to quantify

0 the effect of implementing this change. In particular, the
-0.2 potential peaks/troughs at select feeders were amplified as
-0.4 these neighbourhoods comprised of clusters of larger
-0.6 homes. The modelling undertaken focussed on EV adop-
-0.8
tion and then the combination of EV and PV uptake. To
0 5 10 15 20 25 30 35 40 45 50 investigate these different scenarios and their possible
Half hours model outcomes, computing confidence bounds was
90% feeder 17; 50% feeder 17; 10% feeder 17 extremely effective. Moreover, the upper bound can also be
used to determine the available headroom at each feeder
Fig. 7 Result of 500 simulations with 30% EV uptake and 30% PV
uptake summer (seed allocation informed by CTB, every household for some specified uptake percentage. Identifying head-
that obtained an EV were also given a PV) room is essential for network planning since negative
headroom indicates transmission is greater than the maxi-
Modelling EV and PV uptake reveals significant night- mum available power, causing issues for the electricity
time peaks and daytime troughs, which enlarge at some provider. Hence, subsequent to the upper bound calcula-
feeders when the CTB information is applied. These results tion, certain feeders can be highlighted as likely sites for
can inform network operators about the need for reinforc- network malfunction when subjected to LCT demand.
ing some of the feeders so to cater for the load increases at Furthermore, when analysing PV uptake as well, the lower
night and voltage rises during the day, which is due to EV bound becomes an equally important measure since nega-
charging and PV generation respectively. Refer to the tive power at the feeder level is also problematic for the
Appendix A where an overview of the results at the 44 electricity provider. As well, future voltage and thermal
feeders is given. issues can be examined with these bounds if used as input
Analysing load profiles is a straightforward method to to a voltage and thermal constraints simulator. Confidence
assess the impact of LCTs and to highlight potential sites bounds therefore will be an important tool to inform new
that require further attention. However, other factors such policies and planning so that the future impact of LCTs on
as thermal and voltage issues must also be considered to the LV network can be mitigated.
fully understand the LCT impact. To explore these issues,
the feeder upper and lower bounds obtained through our Acknowledgements This work was supported by Scottish and
Southern Electricity Networks through the New Thames Valley
modelling are used by a network operator as input to a Vision Project (SSET203 New Thames Valley Vision), and funded by
voltage and thermal constraints simulator. the Low Carbon Network Fund established by Ofgem.

Open Access This article is distributed under the terms of the

Creative Commons Attribution 4.0 International License (http://
7 Conclusion creativecommons.org/licenses/by/4.0/), which permits unrestricted
use, distribution, and reproduction in any medium, provided you give
An agent based model was outlined that considered appropriate credit to the original author(s) and the source, provide a
social factors to predict the uptake of low-carbon tech- link to the Creative Commons license, and indicate if changes were
made.
nologies. The data used was taken from real-life, with real

123
Green neighbourhoods in low voltage networks: measuring impact of electric vehicles and... 115

Appendix A feeder does not receive any LCTs. Panel a depicts each
feeder’s average CTBhh (blue-dashed) and household
The confidence bounds results for the studies: population (red). Panel b relates to study a, where the
maximum value of the 90% EV feeder quantile (the
1) 30% EV uptake winter with
baseload subtracted), divided by the number of households
a) seeds randomly distributed; along the feeder, is shown. The red and blue curve
b) seed distribution guided by CTB information. correspond to a(i) and a(ii) respectively. Panel c displays
the results for study b, with the minimum value of the 10%
2) 30% EV and 30% PV uptake summer with
EV/PV feeder quantile (the baseload subtracted), divided
a) seeds randomly distributed; by the number of households along the feeder, given. The
b) seed distribution guided by CTB information. red and blue trend are linked with b(i) and b(ii) respec-
tively. There is an evident correlation between the feeder
are summarised in Fig. 8 for all 44 feeders. Note that feeder population and the red curves associated with a(i) and b(i).
1 is comprised of only commercial properties, so this This behaviour was discussed in Section 4, where less
populated feeders received larger LCT loads. When the
6 150 CTB data is introduced, certain feeders attain greater
Average feeder CTB hh

Feeder population extreme values, whilst at other feeders the load magnitude
4 100 is reduced. This is demonstrated by the blue curve along
the bottom two panels, which overall follows the top panel
blue trend. This is expected since when CTBs are applied,
2 50
the clustering algorithm favours feeders that have a higher
proportion of larger properties. The feeders that receive
0 0 amplified minimum and maximum values are especially
1 10 20 30 40 44
Feeder number vulnerable, therefore further analysis and possibly rein-
Average feeder; Feeder population forcement at these sites are needed.
(a) The average CT B hh at each feeder (blue-dashed) and the
variation in the feeder household population (red)
0.9
0.8 References
Maximum (kWh)

0.7
0.6 [1] Department of Energy & Climate Change (2014) UK energy
0.5 statistics: statistical press release. https://www.gov.uk/government/
0.4 news/uk-energy-statistics-statistical-press-release-september-2014
0.3 [2] National Grid (2015) Electricity ten year statement. National
0.2 Grid, London. http://www2.nationalgrid.com/UK/Industry-
0.1 information/Future-of-Energy/Electricity-ten-year-statement/
0 [3] Bilton M, Chike NE, Woolf M et al (2014) Impact of low
1 10 20 30 40 44
voltage-connected low carbon technologies on network utilisa-
Feeder number
tion. Report B4. Imperial College London, London, UK
Random seeds; CTB seeds
[4] Bossmann T, Staffell I (2015) The shape of future electricity
(b) The peak value of the feeder 90% quantile, divided by the demand: exploring load curves in 2050s Germany and Britain.
number of households along this feeder (study a) Energy 90(2):1317–1333
0 [5] ElNozahy MS, Salama MMA (2013) Technical impacts of grid-
-0.1 connected photovoltaic systems on electrical networks: a review.
J Renew Sustain Energy 5(3):116–120
Minimum (kWh)

-0.2
-0.3 [6] Yong JY, Ramachandaramurthy VK, Tan KM et al (2015) A
-0.4 review on the state-of-the-art technologies of electric vehicle, its
-0.5 impacts and prospects. Renew Sustain Energy Rev 49:365–385
-0.6 [7] Neaimeh M, Wardle R, Jenkins AM et al (2015) A probabilistic
-0.7 approach to combining smart meter and electric vehicle charging
-0.8 data to investigate distribution network impacts. Appl Energy
-0.9 157:688–698
1 10 20 30 40 44
[8] Liu MM, McNamara P, Shorten R et al (2015) Residential
Feeder number
electrical vehicle charging strategies: the good, the bad and the
Random seeds; CTB seeds
ugly. J Mod Power Syst Clean Energy 3(2):190–202. doi:10.
(c) The minimum value of the feeder 10% quantile, divided by 1007/s40565-015-0122-2
the number of households along this feeder (study b) [9] Watson JD, Watson NR, Santos-Martin D et al (2016) Impact of
solar photovoltaics on the low-voltage distribution network in
Fig. 8 Overview of results for all 44 feeders New Zealand. IET Gener Transm Distrib 10(1):1–9

123
116 Laura HATTAM, Danica Vukadinovic GREETHAM

[10] van der Kam M, van Sark W (2015) Smart charging of electric [18] Understand how council tax bands are assessed. http://www.
vehicles with photovoltaic power and vehicle-to-grid technology lancaster.gov.uk/council-tax/council-tax-bands/
in a microgrid; a case study. Appl Energy 152:20–30 [19] Council Tax. https://www.gov.uk/council-tax/working-out-your-
[11] Navarro-Espinosa A, Ochoa LF (2016) Probabilistic impact council-tax
assessment of low carbon technologies in LV distribution sys- [20] California Center for Sustainable Energy (2013) California plug-
tems. IEEE Trans Power Syst 31(3):2192–2203 in electric vehicle driver survey results
[12] Poghosyan A, Greetham DV, Haben S et al (2015) Long term
individual load forecast under different electrical vehicles
uptake scenarios. Appl Energy 157:699–709 Laura HATTAM is a Postdoctoral Research Assistant within the
[13] Mullan J, Harries D, Braunl T et al (2011) Modelling the Centre for the Mathematics of Human Behaviour at the University of
impacts of electric vehicle recharging on the Western Australian Reading. Her research interests include dynamical systems theory and
electricity supply system. Energy Policy 39(7):4349–4359 its applications, and mathematical modelling.
[14] Adepetu A, Keshav S, Arya V (2016) An agent-based electric
vehicle ecosystem model: San Francisco case study. Transp Danica Vukadinovic GREETHAM is Lecturer in Mathematics and
Policy 46:109–122 Director of the Centre for the Mathematics of Human Behaviour at
[15] Giasemidis G, Haben S, Lee T et al (2016) A genetic algorithm the University of Reading. Her main expertise is in algorithmic graph
approach for modelling low voltage network demands. https:// theory and network modelling and analysis. Her research interests
arxiv.org/abs/1612.06833 include mathematical modeling of low voltage networks.
[16] EA Technology Limited (2016) My electric avenue
[17] RIIO-ED1 price control. https://www.ofgem.gov.uk/electricity/
distribution-networks/network-price-controls/riio-ed1-price-control

123