Energy & Buildings 213 (2020) 109794

Contents lists available at ScienceDirect

Energy & Buildings

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

Building as a virtual power plant, magnitude and persistence of

deferrable loads and human comfort implications
Mohammad Royapoor∗, Mehdi Pazhoohesh, Peter J. Davison, Charalampos Patsios,
Sara Walker
School of Engineering, Newcastle University, Newcastle upon Tyne, NE1 7RU

a r t i c l e i n f o a b s t r a c t

Article history: This work uses high resolution data from 130 electricity sub-meters to characterise a 12,500m2 com-
Received 23 September 2019 mercial building as a virtual power plant (VPP) by assessing magnitude and duration of electrical loads
Revised 10 January 2020
suitable for demand response (DR). In 2018, the building had a peak hourly demand of 48 W/m2 and
Accepted 16 January 2020
Available online 30 January 2020
its electricity consumption (183.2 kWh/m2 /yr.) was within low to medium range of air-conditioned UK
portfolio. Deferrable loads from heat pumps, air handling units, lifts, lighting, circulating pumps and dry
Keywords: air coolers were used to illustrate building’s DR capability over a maximum duration of 4 h per DR cycle.
Demand response On average, deferrable loads form 46.4% of total building electricity consumption and across a 4-hour DR
Deferrable electrical loads cycle can be characterised as having an initial power (and stored energies) of 28 kW (401 ± 117 kWh);
Human comfort 109 kW (571±82 kWh); and finally 138 kW (625±18 kWh) for 100%, 41.5% and 24.6% of time respec-
Virtual power plant
tively. Following a DR event, the HVAC ability to restore original indoor climate was found to be at least
Big data
twice as fast as climatic drift during the event. A linear regression model was found to be weak in using
Energy storage
external temperature to predict the magnitude of aggregated deferrable loads.
© 2020 Elsevier B.V. All rights reserved.

1. Introduction operations of electrical and natural gas networks rather than be-
ing operated as isolated energy entities [6]. While previously only
Buildings are complex thermodynamic objects that account for large industrial power consumers were configured for DR, substan-
a large proportion of energy consumption in developed and devel- tial capabilities can also be offered by clusters of buildings. The
oping countries. Building sector has been reported to be responsi- next 2 subsections outline recent progress in building-enabled DR,
ble for 20% of total delivered energy worldwide [1], 31% of global the acceptable boundaries of occupant comfort under a DR event
final energy demand [2] and finally 40% of primary energy con- and finally objectives of this work.
sumption in International Energy Agency (IEA) member countries
[3]. At the same time an ageing electricity infrastructure in the UK 1.1. Literature review
and much of the developed world are undergoing radical changes
which are driven by decarbonisation needs, renewable and storage Peak electrical demand occurs at different seasons across the
integration, operational reliability and support for electric heating world. In US - home to the largest electricity grid in the world
and electric vehicles (EVs) [4,5]. It would be impossible to achieve [7] - median peak electrical demand in all commercial buildings
these goals without a greater interoperability between buildings was reported as 58.1 W/m2 (with offices as 64.6 W/m2 ). Two third
and electrical grids, where buildings can modulate their consump- of US buildings’ peak demand occur at summertime and when
tion in aid of grid operators to guarantee network stability and combined, commercial buildings accounted for 45% of summer-
optimise operational cost and carbon footprint. Most commercial time peak demand [8]. However while in warmer climates the
buildings are characterised by high thermal inertia and host flex- refrigerant-driven cooling load mean that transmission and distri-
ible and non-critical loads, thermal storage, existing automation bution networks are strained most at summer-time, heating dom-
systems and in some cases on-site generation and electrical stor- inated regions such as those in Northern Europe experience peak
age. These characteristics enable buildings to actively participate in electrical demand during colder winter months [9,10]. In domestic
sector, EV penetration and electrification of heating have the po-
tential to nearly double electricity demand, which while imposing
∗
Corresponding author. challenges will also mean that each household will have the ability
E-mail address: [email protected] (M. Royapoor). to exercise much greater flexibility given adequate economic and

https://doi.org/10.1016/j.enbuild.2020.109794
0378-7788/© 2020 Elsevier B.V. All rights reserved.
2 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794

environmental incentive and information [11]. This emerging flex- [17] rather than in pursuit of full realisation of building potentials
ibility is now being investigated broadly in scientific communities as highly responsive energy entities. Incorporation of DR in build-
primarily via simulation-based techniques to develop dynamic and ing codes however is likely to accelerate the presence of buildings
real time control and optimisation algorithms. in the DR market as for instance the estate of California became
In addition to relieving network congestions, DR is reported to the first to require new buildings to incorporate automated utility-
provide competitive pressure to reduce wholesale electricity prices to-customer energy management in its 2019 Building Energy Stan-
and mitigate market dominance by limited players, increase energy dard Title 24.
use awareness and drive operational efficiency in energy markets.
It can also be a mechanism for integration and optimisation of re- 1.1.1. DR using hvac equipment
newables and distributed storage, assist EV take-up, facilitate bidi- Despite variations in complexity, heating, ventilation and air-
rectional electricity flow and ultimately ensure greater grid stabil- conditioning (HVAC) systems are present in most building types
ity. From the early 1980s, on-site thermal storage offering HVAC regardless of climatic condition and building function. Using a 12-
peak shaving/shifting and multi-tiered energy tariffs have been storey case study building, the ventilation system (that are broadly
used by building and utility operators as tools for load manage- implemented in colder Nordic countries) was examined as an asset
ment. However information infrastructure (i.e. pervasive sensing capable of offering DR. The ventilation system was found to be ca-
and advanced metering infrastructure), building automation and pable of ramping up or down which at aggregate level (796 similar
market tariffs have not so far been in place to enable DR to reach properties) could offer a peak power reduction of 1.57 MW [18].
its full potential. UK National Grid has a licence obligation to con- A similar study used HVAC modelling with a particular focus on
trol system frequency at 50 Hz ±1% and provide a supply voltage fast responding chiller systems to conclude that 32% to 66% of the
at 230 V (with a range of −6% to +10%) and a ramp rate that is electrical consumption of the building is available for DR while ac-
limited to around 0.2% of capacity per second for frequency rate knowledging that the ‘fixed’ reduction preferred by the grid might
of change of 0.1 Hz/s [12]. Unbalanced conditions (e.g. from si- not be achievable in buildings unless active thermal storage are de-
multaneous EV charging) can degrade power quality, increase har- ployed [19]. A simulation based study emphasises building-specific
monics and voltage issues and potentially lead to line losses and DR design to reflect the ability of different zones or loads to offer
equipment damage. These issues arise not only at instances of high DR over different durations without comfort penalty and concludes
demand, but also steep reduction in generations (i.e. wind tur- by identifying a maximum power demand reduction of 14.7% in a
bines hitting cut-off points). At such instances buildings, in par- sports and leisure facility if centralised chillers were targeted [20].
ticular those with larger electrical demand can switch-off or turn 5 large commercial and academic buildings in the pacific North
down interruptible electrical loads and in doing so perform a vir- West of US with demand ranges of 16–78 W/m2 were found to
tual power plant service. A number of mechanisms can oversee be able to shed 16% of peak summer and 14% of peak winter loads
the ‘building as a virtual power plant’ behaviour with electricity with overall aggregate winter DR capability (767 kW) being twice
price being one specific example. For instance in a real-time mar- that of summer (338 kW) [21]. The load shedding ability of 29
ket, a dynamic demand response controller (DDRC) was proposed buildings in California were found to reduce slightly with rising
to aid the regulation of a congestion-prone electrical network [13]. outdoor temperatures and this overall trend showed little varia-
The controller allows the consumer to set an energy price thresh- tion across 4 years of analysis [22]. EnergyPlus simulation of 11
old which in turn informs the DDRC to trigger set-point changes buildings was found to be able to predict the magnitude of build-
in a single dwelling to avoid periods when threshold is exceeded. ing DR with ±10% accuracy against a benchmark of actual trials
The study reports that in a cooling dominated environment such that demonstrated 15–30% DR capability [23]. Reviewing a total of
as Texas which has rapidly rising demands and limited reserve ca- 22 field and climate chamber based thermal comfort studies, Agh-
pacities, houses have the capacity to offer significant peak electric- niaey, S. et al. conclude that using more flexible upper and lower
ity demand reductions and simultaneously achieve reduced energy boundaries of adaptive thermal comfort to control building climate
bills with negligible variation in occupant comfort. offers greater opportunities for successful DR implementations in
Regulatory and incentive based mechanisms in North America, buildings [24], where indoor summertime temperatures as high as
Europe and parts of Asia (China, Hong Kong, India, Japan and Ko- 28 °C were found to be tolerated with a relaxed uniform policy
rea) are increasingly deployed to realise aggregated DR potentials and occupant engagement. The aforementioned authors consult a
in buildings [14]. In USA buildings are a major part of DR resources number of previous studies to report that a temperature ramp of
that provide economic and reliability-based DR equivalent to 3% to between 0.5 °C to 0.6 °C an hour is hardly perceptible by the oc-
9% of peak summer demand across different states. This was put cupants with adverse effects on productivity reported only in one
to test in 2018 (due to 4th hottest summer on record in US) where climate chamber-based study [25] when the length of exposure to
200 buildings responded to real-time grid operator stress signals temperature ramps exceeds 4 h. Another reference source of ther-
to collectively reduce 4.5MW of demand across 4 large grid oper- mal drifts in buildings can be found in Table 5.2.5.2 of ASHRAE 55
ators. However better programme and aggregation designs, com- which defines maximum acceptable temperature ramps to maxi-
pensation rules and penalties for non-compliance are needed to mums of 2.2 °C, 2.8 °C and 3.3 °C respectively in the first, second
improve predicted DR capabilities that in their current form over- and forth hour of thermal drifts [26]. Strong empirical evidence ex-
estimate participation and performance [15]. In UK the participa- ists that ranks the perception of thermal environment as the most
tion of smaller energy users in DR markets and creation of asso- important determinant of occupant comfort [27], and given that
ciated technical solutions and business models have been pursued the only other component of human comfort to be impacted by a
through the roll out of 53 million smart gas and electricity me- DR event will be the visual element for which no time-limit is de-
ters. EU directive 2012/27/EU encouraged member states to facil- fined (see 1.1.2), this work borrows from the aforementioned bod-
itate DR and consumer participation in energy markets [16] and ies of work to limit the DR duration to 4 h and also consider 0.5 °C
digitise EU’s power sector (Sweden, Finland, Estonia and Spain all of temperature drift per hour as imperceptible, and 2.2 °C, 2.8 °C
have smart metre adaption exceeding 90%) . However notably the and 3.3 °C of successive hourly thermal drifts as the maximum per-
majority of DR products, services and business models are devel- missible limits. A 4 hour maximum DR period also reflects the typ-
oped in response to market and power system needs. Regulatory ical congestion periods on electrical grid. While demand peaks are
interventions have so far been infrastructure-focused and in re- likely to last for 15 to 30 min, periods of congestion on electri-
sponse to events that expose the vulnerability of power systems cal networks can last up to 4 h, in particular during peak cooling
 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794 3

Table 1
Sub-metered building assets categorised as interruptible loads in a DR event and the comfort thresholds that trigger DR termination.

Maximum Advance
turndown time Maximum
Abbr. Description No. ratio a (ω) needed DR duration Occupant comfort threshold

HP Heat pump 25 100% 0 4h [a] Tzone : Min: 19 °C Max: 28 °C

[b] Thermal drifts: 0.5 °C per hour of thermal drift per hour found to
be imperceptible, with maximum allowable drifts in the 1st, 2nd and
3rd hour as 2.2 °C, 2.8 °C and 3.3 °C.
C.P. circulating pump 21 HTG 100% 0 4h None
(Secondary circuits) 20 CHW
C.P. circulating pump 6 HTG 100% 5 minb 4h None
(Primary circuits) 6 CHW
AHU Air Handling Unit 22 100% 0 4h 1000 ppm CO2c
Lt. Lighting circuit 22 circuits 24%d 0 Not limited Illuminance level (Lux)
serving 421 Office: 400(N) 300 (DR) 100 (HSE)e
zones Labs: 400(N) 300 (DR) 200 (HSE)
Kitchen 400(N) 300 (DR) 50 (HSE)
Circulation 150(N) 100 (DR) 5 (HSE)
Lf. Lifts 6f 100% 0 N/A
DAC Dry Air Coolers 3 100% 0 4h None
a
Equates to the weighting factor (ω) in the summation of deferrable loads (see expression [1]).
b
Need to be shut 5 min after heat pumps to avoid evaporator freezing.
c
See Section 1.1.1.
d
See section 1.1.2
e
N: Normal; DR: Demand response mode; HSE: Health and Safety Executive minimum statutory requirement as outlined in section 1.1.2.
f
Two lifts out of 6 are discounted for fire, emergency and disabled person access.

or heating seasons. Utility providers normally initiate load shed- achieved 14%, 15% and 18% reductions in three trials with no neg-
ding activities in such prolonged overloading periods within which ative feedback reported in either case. A similar study of auto-
building-enabled DR can be one invaluable resource. mated DR field trials in over 1 M square meters of office space
In addition to thermal environment, CO2 concentration is di- reported average lighting energy reductions of 7%−9% over 3 to
rectly affected by an HVAC-based DR event. Acceptance criteria 6 h with maximum reductions of 28% to 56% achieved over shorter
on CO2 concentration is dictated by the length of exposure, and periods of time [36]. Chartered Institute of Building Services Engi-
while levels up to 50 0 0 ppm are tolerated in aviation (aircraft cab- neers (CIBSE) recommends a maintained illuminance of 30 0–50 0
ins) and industrial applications [28], modern building design is ex- Lux in cellular and open plan offices and 200 Lux for circulation
pected to deliver 5 to 10 litres of fresh air per second per person [32]. Minimum allowable maintained illuminances are a function
to maintain CO2 concentration below 10 0 0 ppm [29,30]. The Eu- of whether ‘movement’ or ‘work’ is being catered for. UK Health
ropean standard EN 13,779 [31] defined categories IDA1 to IDA4 and Safety Executive (HSE) sets minimums of 100 Lux (for office
representing high, medium, moderate and low indoor air quali- work with typical admin duties) or 200 Lux (where high percep-
ties. These categories were adapted by the Chartered Institute of tion of details is required such as in electronic component or tex-
Building Services Engineers (CIBSE). CIBSE Guide A (Table 4.5) out- tile design) [37]. HSE also assigns minimums of 50 and 5 lux to
lines how ventilation rates specified within four IDA categories Kitchens and circulation areas respectively. While HSE’s values are
translate to typical total indoor CO2 values of 700–750 ppm, 850– the statutory minimums, this work adopts a maximum illuminance
90 0 ppm, 1150–120 0 ppm and 1550–160 0 ppm respectively [32]. reduction of 24% as at this level of illuminance reduction no ad-
While the latest EU standard on commercial building ventilation verse impact on occupants have been observed [35]. This will en-
(BS EN 16,798–3:2017) makes no direct reference to thresholds of sure that artificial lighting during a DR event will remain well
acceptable CO2 concentration, moderate impairment of cognitive above statutory minimums (see table 1).
performance has been reported at 10 0 0 ppm with significant im- Given their intermittent usage, lifts may be perceived as a poor
pairment at 2500 ppm (from a 2.5 hour exposure (n = 22)) [33]. DR candidate. However given that most lifts have auxiliary (venti-
Therefore in line with best practice guides [29,30], CO2 concentra- lation and lighting) and parasitic and standby electricity consump-
tion of 10 0 0 ppm is defined as the acceptable threshold for IAQ tion, where multiple installations exists a notable load can exist
beyond which DR event shall be abandoned and mechanical venti- in aggregation. Limited studies of lifts as a DR source exists. One
lation activated. such report is an experiment in a 12 storey Danish student dormi-
tory containing 159 small apartments where elevators accounted
1.1.2. DR using lighting and lifts for 5% (i.e. 25 kWh) of average daily building electricity consump-
Automating and optimising lighting control in order to offer DR tion (i.e. 577 kWh). The study utilised published electricity carbon
in buildings have been underway since late 90 s. An office based content to examine if lifts would be abandoned in preference for
experiment with illuminance reduction of 2% per minute and tem- stairs at times of high electricity demand when fossil-fuel derived
perature increase of 1.5 °C over 2.5 hour found 126 participants ratio of electrical supply was at its maximum (indicated by a red
to be more sensitive to (and dissatisfied at) the reduction of il- light in each lift). The study found a short-lived initial change of
luminance than thermal drift [34]. Open plan office spaces in a occupant behaviour which fails to last. Overall the study indicates
governmental building (lit with dimmable T8 lamps) and several no significant reduction of lift energy consumption or pattern of
university campus buildings (lit with dimmable T5 lamps) were use between times indicated by green or red lights [38]. The 6
used to conduct several DR trials while maintaining minimum rec- lifts in operation within the target building in this work are as-
ommended illuminance level for periods of 1 to 30 min [35]. The sumed to be able to partake in DR operation with termination as-
government building trial reported lighting demand reductions of sumed to be triggered in case of an emergency or disabled user
23% and 24% in two DR trials; while university campus buildings request.
4 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794

Table 2
Design guidelines and as built thermo-physical properties of the case-study building.

Illumination level (Lux) Target temp (°C)a

Ventilation
Winter Summer

Offices / labs/meeting rooms 400 22±3 24±3 10 l/s/person

Circulation 150–200 20±3 N/A 2 ACH b
Kitchens 400 22±3 24±3 To suit equipment installed

Fabric: Actual Statutory requirement % improvement

2
Glazing (W/m .K) 1.40 2.20 36%
Walls (W/m2 .K) 0.17 0.35 51%
Floor and roof (W/m2 .K) 0.14 0.25 44%
a
Optimum start HVAC control scheduled between 08:0 0am – 17:0 0pm.
b
Air change per hour.

1.1.3. DR automation, acceptance and wider impact 3 Data from 130 electrical sub-meters are analysed to [i] illustrate
Clearly given the variety and nature of deferrable electrical the magnitude and duration of deferrable loads and [ii] to draw
loads, DR can only be meaningfully executed by automation. For analogies between these deferrable loads and electrical energy
instance four different machine learning methods were deployed storage technologies in order to to highlight distinct character-
in conjunction with a data-driven technique to determine build- istics of a building as a viable virtual power plant (VPP) [45].
ing thermal response in a study aimed at DR and diagnostics au- 4 Finally the extent to which external temperature can be used as
tomation [39]. A building energy management was developed and a linear predictor of deferrable loads and also the applicability
trailed in this trial (using a 12 storey Danish building) that illus- of our findings to the wider commercial portfolio both in the
trated respective 44%, 8%, 7% and 2% shifting potentials in ventila- UK and across similar climatic conditions are examined.
tion, laundry, heating and lift electricity usage [40]. However while
most of the literature reported here point to DR capabilities of in-
2. Method
dividual buildings, Shen, L. et al. deploy a combination of conven-
tional Genetic Algorithm and HVAC system modelling to respond
2.1. Data collection, imputation and error treatment
to either ‘dynamic pricing’ or ‘time of use’ price signals in cooling
dominated building clusters and concludes that at building-group
Newcastle Urban Observatory (UO) is the largest depository of
levels, DR events can unwittingly produce unwanted peaks [41].
publicly held data in the UK [46] and deploys a completely inde-
These unwanted post DR peaks have opened up research into co-
pendent data collection and back-haul infrastructure that is able
ordinated energy management of multiple buildings that can meet
to work alongside building management system (BMS) and re-
distribution operating constraints [42]. Additionally evidence on
trieve associated data by assigning specific IDs to each data stream.
the extent to which ‘residential consumers’ might be willing to opt
This platform is configured to round incoming values to 3 deci-
into a DR programme was examined and actual participation was
mal places and log a new value only if a change is observed. The
found to be considerably smaller than high levels of DR used in fu-
magnitude of change required to trigger a new log is ±0.0 0 05,
ture energy system scenarios [43]. Therefore both supervision and
which is in excess of the resolution of temperature and CO2 sen-
coordination at building-group level as well as the willingness of
sors (±0.4 °C at 24 °C or ±50 ppm across 400–1250 ppm) or the
residential participants require further clarification and research.
compound instantaneous deviation of electrical sub-meters (man-
ufactured to IEC 62,053–21 with typically ±0.2% accuracy in ac-
1.2. Objectives tive power measurement over operational range). Data backhaul
and compilation is therefore several order of magnitude more pre-
This work attempts to satisfy two objectives. First high- cise than the measurement accuracy of physical sensors and me-
resolution data from 130 electrical sub-meters in a large commer- ters leaving no data type that would have benefited from a more
cial building are used to quantify the magnitude but also impor- accurate data compilation technique. Within the case study build-
tantly the persistence of all deferrable loads across a full annual ing (Urban Science Building or USB, see 2.3) over 15,0 0 0 m, sensors
cycle. Assessing these characteristics not only builds confidence in and HVAC components are detected by UO platform which gener-
determining overall flexibility, but also contributes to de-risking ate around 50bn logged data points per year. The average duration
the use of building-enabled DR in smart grids. The second objec- between raw data points analysed in this work are 4.6 s (AHUs),
tive will concern a closer examination of human comfort under a 5.5 s (incoming mains electrical meters), 8 s (circulating pumps),
DR event which has been highlighted in recent literature as inade- 6.6 s (heat pumps), 4.1 s (lifts) and 11 s (lighting). This resulted
quately understood [24,44]. The following steps are taken to satisfy in approximately 600 m data points that were aggregated to form
these two objectives: half-hourly (HH) electricity consumption. Where missing data ex-
isted, a KNN (K- Nearest Neighbour) approach was found to be the
1 Scientific literature and professional guides reviewed in most appropriate method of data imputation [47]. KNN is a non-
Section 1 are used to define the boundaries of visual, respira- parametric method where the missing value is replaced (imputed)
tory and thermal comfort that have to be maintained in a DR by the average of a discretionary number (K) of neighbouring data
event, as well as acceptable duration of a DR event. points. This work used a k value of 10 to treat missing data points.
Di f f.
2 The case-study building’s response to a DR event is examined At any instance of time (t), the aggregate deferrable load ( p(t ) )
using actual operational data in areas deemed to be impacted calculated in this work refer to metered active power values which
most (referred to as worst case zones) to determine the extent are the summation of all interruptible electrical loads outlined in
to which occupant comfort thresholds might be exceeded. table 1 and subtracted from total active power consumption ( pUSB (t ) )
 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794 5

of the building, in other words: time interval (using kW/7 s unit) to retain the granularity of origi-
⎛ AHU  AHU  HP ⎞ nal data. Finally R2 values in Section 3.4 are driven from expression
ω p(t ) + ωHP p(t ) + 5.
⎜ Lt. α ∈ {A } β ∈ {B }
 ⎟ 2
⎜ ω (t ) + ω
pLt. C.P.
pC. P. ⎟ 
pDi f f.
(t ) = p(t ) − ⎜
USB
γ ∈ { } δ ∈{}
(t ) ⎟ (x − x̄ )(y − ȳ )
⎝  L f.  DAC ⎠
2
R =   (5)
+ωL f. p(t ) + ωDAC p(t ) (x − x̄ )2 (y − ȳ )2

ε ∈ {E } ζ ∈ {Z }
∀α ∈ {A} = 1, . . . , eAHU
, (1) In which the numerator and denominator respectively quantify
∀β ∈ {B} = 1, . . . , f HP
, explained and total variations between dependant and indepen-
dent variables.
∀γ ∈ { } = 1, . . . , gLt. ,

∀δ ∈ {} = 1, . . . , hC.
P. ,
2.2. Characterising deferrable loads using initial power (kW) and
∀ε ∈ {E} = 1, . . . , iL f. ,
stored energy (kWh)
∀ζ ∈ {Z} = 1, . . . , jDAC
As noted earlier deferrable loads refer to the magnitude of the
In which at time t, active power of AHUs ( pAHU (t ) ), heat pumps Di f f.
active power ( p(t ) ) and are characterised based on a 4-hr DR
HP C. P. L f.
( p(t ) ), circulating pumps ( p(t ) ), lifts ( p(t ) ) and Dry air Coolers event assuming that [i] windows can be opened, [ii] examination
(t ) ) have a weighting factor (ω) of 1 (i.e. instantaneous reduc-
( pDAC of deferrable loads over a complete annual cycle involves suffi-
tion of 100%) and lighting circuit active power ( pLt. cient seasonal and operational variation to adequately characterise
(t ) ) has a weight-
a building as a VPP and [iii] that at any point of time if the build-
ing factor of 0.24 (as outlined in table 1).
ing was to undergo a DR trial, the deferrable load at that point
The smallest historical base-load ( pUSB : min
(hist. ) ) observed for the en- of time (and the sum of deferrable loads over the ensuing 4-hour)
tire building is 21 kW (02:00 h on 25 Dec 2018). This load com-
could represent the availability of initial power (and stored energy)
prises of fire and security systems, emergency lighting and comm
in a virtual power plant (or could be equated to the discharge be-
room server duties. In order to ensure that a quality control mech-
haviour of an electrical battery). Clearly in addition to initial power
anism exists, this minimum possible baseload ( pUSB : min
(hist. ) = 21 kW) and stored energies, electrical batteries are characterised by their
is combined with the following two dynamic loads to form an ab- discharge curves, energy and power densities, temperature depen-
solute minimum non-deferrable load for any given time t: dency and service and cycle life. However the aim here is to offer
a measure of a building’s capabilities as an effective component of
1 76% of lighting load which is non-deferrable (i.e. ωLt. = 0.76) as
VPP heterogeneous technologies using actual operational data. As
outlined in table 1.
discussed in Section 1.1, human comfort and HVAC operational in-
2 100% of small power loads (i.e. ωs.p. = 1) serving PCs and office
 tegrities impose the following caveats on a DR event:
equipment (ωs.p. ( ps.p
(t )
)
η ∈{ H }
1 Advance notice: None of the assets considered for DR require
In other words: any advance notice except for the secondary circulating pumps
  that need to remain operational 5 min after heat pumps are
   shut down in order to avoid frost build-up and freezing of evap-
pnon −Di f f.
= pUSB pUSB : min
+ ω Lt. pLt. + ωs.p. ps.p
(t ) − (hi. ) (t )
(t ) γ ∈ { } (t ) orators.

η ∈ {H }
2 Duration: If it was possible for windows to be opened, the only
∀γ ∈ { } = 1, . . . , k Lt.
time limit enshrined in best practice guides for a DR event is
∀ζ ∈ {Z} = {1, . . . , l s.p} 4 h (see 1.1.1). If extreme external temperature made window
(2) opening impractical, the only limiting factor is the CO2 con-
centration that for the worst case zone (a high density teach-
Clearly a direct relationship should exists between expression 1
ing area) was 30 min (the time taken for CO2 concentration to
and 2 as follows:
reach 10 0 0 ppm).
non−Di f f.
pDi f f. USB
(t ) = p(t ) − p(t ) (3) 3 Recovery time: [i] thermal: actual temperatures for 2 worst
case zones showed that thermal recovery time was twice as
Instances of error in this work are defined if for any instance fast as thermal drift (Fig. 3–4). ASHREA maximum permissi-
of half hourly data, Eq. (3) is violated. With respect to half- ble hourly thermal drifts of 2.2 °C, 2.8 °C and 3.3 °C in suc-
hourly data in 2018 (i.e. 17,520 instances of time), in 139 instances cessive hours post DR (as outlined in 1.1.1) would not be vio-
Di f f.
the aggregate deferrable load ( p(t ) ) exceeded the difference be- lated post DR in the case study building as demonstrated using
non−Di f f. worst case zones. [ii] IAQ: in the absence of window opening,
tween total and non-deferrable USB load ( pUSB (t ) − p(t ) ). This
CO2 restoration time to pre-DR event in the worst case scenario
equates to 0.8% error in measured data that can result from (i)
was 45 min (Fig. 2), while if window opening is undertaken the
floating-point and truncation errors (these occur when averaging
maximum CO2 threshold (10 0 0 ppm) won’t be reached.
the dataset to HH time steps), (ii) the impact of imputation and
(iii) the measurement errors of metering infrastructure (i.e. sensor
non−Di f f. 2.3. Case study building
downtime). In these instances pUSB (t ) − p(t ) is deemed to be a
more accurate representation and is treated as the deferrable load
The case study building (Urban Science Building (USB) lo-
at time t.
cated at 54°58 N, 1°36 W) provides 12,500m2 of laboratory and
In sub-Section 3.3, active power ramp rate at any instance of
office space of which 91.3% are treated. The building was de-
time (RRUSB
(t ) ) are derived using expression [4]. signed and built with advanced metering infrastructure (AMI) in
RRUSB USB USB order to allow detailed energy audits for research projects. The re-
(t ) = p(t ) − p(t−1 ) (4)
sults reported in this work belong to 2018 when the min, aver-
Given that original active power data was recorded at an aver- age and Max site-specific temperatures were −4.5 °C, 10.5 °C and
age frequency of 7 s, active power ramp rate is driven for the same 29 °C.
6 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794

Fig. 1. The case study building has 24 plantrooms distributed across 3 vertical service cores.

With an overall envelope thermal resistance of 0.15 W/m2 .K, 3. Results

the building’s opaque fabric exceeds UK’s legal limits [48] by
44%. Internally large areas of exposed concrete ceilings and walls This section first addresses the impact of a DR event on oc-
mean that the building’s mass and zone air are closely coupled. cupant comfort (3.1) before outlining the magnitude of deferrable
The building’s total weight (excluding live loads) is 27,0 0 0 tonnes loads, addressing the possibilities of post DR power surge and the
and high exposed internal mass within a well-insulated fabric re- degree to which external temperature can be a predictor of de-
sults in very slow internal temperature drifts. Heating, cooling ferrable loads within a linear model (3.2).
and domestic hot water (DHW) are provided by water to water
heat pumps utilising R410a refrigerant that evaporate and con- 3.1. DR impact on human comfort
dense between 5 °C-47 °C (three heat pumps designated to DHW
production condense at 57°). Natural gas boilers boost DHW to The following 3 sub-sections use actual building climate data
60 °C and provide space heating top-up duty during peak winter to establish acceptable duration of a DR event during which HVAC
demand. equipment and lighting will have to be shut (or turned) down us-
The bivalent nature of the heat pumps requires heat rejec- ing specific building zones that represent worst case scenarios.
tion and extraction which is facilitated via 3 heating and 3
cooling water loops that are housed in 3 main service cores 3.1.1. IAQ
(Fig. 1). These loops are maintained at relatively constant tem- Although CO2 is not strictly an indoor contaminant, its concen-
peratures of 40 °C (heating) and 15 °C (cooling) and act as tration is an indication of indoor air quality and increasingly values
primary heat source/sink for 22 distributed heat pumps that above 10 0 0 ppm are discouraged. Using actual internal CO2 levels
in turn maintain secondary heating (flow/return: 47 °C/40 °C) and occupant numbers as starting points, DR-driven CO2 accumu-
and cooling (flow/return: 10 °C/15 °C) circuits that supply ac- lation is simulated using a validated EnergyPlus model which has
tive ceiling beams. This cascaded design, together with the large ±2% area and ±5% volumetric accuracy against the actual building
building size and the distributed nature of HVAC services re- [49]. Fig. 2 outlines actual CO2 concentrations (solid black lines) in
quire 53 circulating pumps to distribute thermal energy around a typical office space (representative of 53% of net internal area) as
the building. These circulating pumps are also considered as DR well as a high density space (representative of 19.8% of net internal
assets. area that also form worst case scenario). Both spaces are ventilated
 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794 7

Fig. 2. CO2 build-up in typical and worst case zones following two DR events @ 1100 and 1800 hrs.

using AHUs that operate at setback level (07:0 0–18:0 0 hrs) and are slower rate of heat loss in particular during office hours when
boosted when CO2 concentration exceeds 700 ppm (all zones have internal heat gains are much greater than after office hours. The
openable windows). Using actual CO2 concentration as the start- worst case zone thermal recovery rate was recorded as 1.23 °C/hr
ing points and a constant outdoor CO2 concentration of 400 ppm; on 8 Jan 2018, when against an external temperature of −1 °C, the
two DR trials are simulated in EnergyPlus. The first DR event rep- heating system lifted zone air temperature from 16.9 °C (at 7pm)
resents peak office occupancy (i.e. 11am) and the second peak elec- to 18.8 °C (at 8pm). Wintertime thermal recovery for the rest of
trical grid congestion time (i.e. 6pm). At peak occupancy (i.e. 11am) conditioned zones are therefore assumed to be equal to or greater
without opening windows and upon turning off AHUs, the typi- than 1.23 °C/hr.
cal office space can accommodate a DR event for 3.5 hrs before The space within the building with the highest recorded sum-
10 0 0 ppm CO2 threshold is reached, while the high density teach- mertime temperatures (i.e. summer-time worst case zone) is a
ing space reaches 10 0 0 ppm in 30 min. After the DR event (and 166m2 high density teaching space on the 4th floor. This zone ab-
with AHU at boost mode), CO2 concentration is restored to below sorbs solar irradiance throughout the day as it is partially glazed
threshold levels in less than 15 min and approaches initial concen- on east, south and west aspects (overall 33% glazed) and given its
trations in an hour (Fig. 2- footnotes). Note that while the dense high occupant thermal gains is prone to rapid temperature rises.
teaching space can only accommodate an AHU-based DR event for Fig. 4 illustrates hourly instances of thermal drift after HVAC shut-
half an hour, if window opening is undertaken the CO2 concen- down in summertime for this zone. While summertime thermal
tration remains below the threshold for the proposed maximum drifts have a small average value close to zero (0.0 0 02 °C/hr), in-
DR duration of 4 h (windows are simulated to be 15% openable). stances above 0.5 °C/hr account for 14.5% of temperature drifts
Therefore AHUs can be interrupted at minimum for a DR duration during summer evenings. Therefore when compared to wintertime
of 30 min (windows closed) and at maximum for 4 h (windows worst case scenario, a DR event in a well-insulated building dur-
opened). ing peak summer months require closer administration of human
comfort. Under clear skies and with 20 occupants, a trial on 25
3.1.2. Thermal comfort Jun 2018 (with outside temperature at 24 °C) demonstrated this
Heating and cooling are provided in the case-study building by zone to have a cooling rate of 1.8 °C/hr (from 23 °C at midday to
22 distributed heat pumps which are configured for optimum start 21.2 °C at 1pm). Summertime thermal recovery rate for the rest of
to achieve comfort temperatures by 08:00am and are turned off at the building is therefore assumed to be equal to or greater than
17:00pm. A combination of adequate insulation, high thermal mass 1.8 °C/hr.
and internal gains mean that in the absence of winter-time heat
injection, the temperature drift in the building is notably slow.
To demonstrate this a north-facing top-floor office space which is 3.1.3. Visual comfort
adjacent to 3 unheated zones is identified as winter-time worst The building has surface mounted LED luminaires that are
case zone. This space is 33% glazed and has the lowest recorded scheduled to operate between 080 0–170 0 h and are controlled
wintertime temperatures in the building. Fig. 3 illustrates hourly via Digital Addressable Lighting Interface (DALI) that incorporates
instances of thermal drift across a full winter season after heat- presence detection and photocell daylight compensation sensors.
ing is shut down at 17:00pm. In the absence of HVAC input the A proprietary combination of spot and tube LED lights are used
average hourly thermal drift is 0.0 0 04 °C which is considerably within the luminaires that offer continuous dimming capability
smaller than the comfort threshold of 0.5 °C/hr (outlined in sub- down to 20% of rated output via digital LED drivers. While vari-
Section 1.1.1 and table 1). The 0.5 °C/hr threshold is exceeded only ations are observed in relative consumed power versus light out-
in 1.2% of instances after HVAC operation is ceased. It could there- puts, LEDs have been reported to typically have a 1:1 light output
fore be reasonably assumed that intermediate floors will have a to power input ratios [50]. This 1:1 ratio is used to quantify the
8 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794

Fig. 3. Boxplot of measured hourly thermal drifts after HVAC shut-down in wintertime worst case zone. Zone description: 33m2 below the roof cellular office, 4 occupants,
adjacent to 3 heated (and 3 unheated) zones.

Fig. 4. Boxplot of measured hourly thermal drifts after HVAC shut-down in summertime worst case zone. Zone description: 166m2 teaching space, 120 seats, adjacent to 1
heated (and 5 unheated) zones.
 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794 9

Fig. 5. Annual profiles of active power for building assets considered for DR (data aggregated at hourly and intervals). Key: C. Pumps: circulating pumps; DAC: Dry Air Cooler.

Table 3
Descriptive statistics of active electricity consumption for assets considered for DR.
a b c d
Lighting HP AHU H Pumps C Pumps DAC & C.L. pumps Lifts Building total

ie (K) 421 456 421 473 650 150 19 170

Av (kW) 22.5 58.9 16.4 11.6 2.1 12.9 1.8 238.6
Max (kW) 62.6 255 39.6 24.9 26.6 24.1 8.3 497.6
Min (kW) 2.2 3.6 1.6 1.6 0.0 0.0 −2.7 98.4
 f
(kWh) 197,357 516,292 143,561 101,797 18,069 113,046 15,763 2090,281
Share of annual total (%) 9.4 24.7 6.9 4.9 0.9 5.4 0.8 100

Notes:.
a
Bivalent Heat pumps.
b
Primary and secondary circulating pumps (heating).
c
Cooling secondary circulating pumps (the only asset that requires 5 min advance notice before undertaking DR).
d
Dry Air Cooler fans and cooling loop circulating pumps.
e
i: The count of all instances of original data points examined (numbers in Thousands).
f
Annual sum of active power consumption for each category.

24% active power reduction in lighting circuits for a DR event as pre- and post-Christmas periods. Only 0.9% of the annual aggregate
listed in table 1 and justified in sub-Section 1.1.2. deferrable loads require a 5-minute advanced notice (Table 3, foot-
note c), with 99.1% of annual deferrable loads available instantly
3.2. Deferrable load characterisation to partake in a DR event. Fig. 7 arranges all aggregated HH de-
ferrable loads in cumulative distribution function (CDF) and his-
Fig. 5 outlines seasonal variation in the active power of six sep- togram forms. A lognormal distribution fit best describes the gen-
arate sub-categories of electrical loads which are considered for DR eral trend of both the histogram (Fig. 7[a]) and the CDF graphs
in this work. Despite being daylight-linked the amplitude and pro- (Fig. 7[b]). These cumulative illustrations characterise the value of
file of lighting electricity consumption remain constant across 2018 a large commercial building as a VPP, since the building never fails
annual cycle. AHU and lift electrical loads also remain constant. to offer electrical demand reduction (the minimum building DR ca-
Heat pumps to a larger extent and associated circulating pumps to pability during 2018 was 28.4 kW). A more representative figure
a lesser extent display seasonal signatures, with surprisingly per- could be derived from half hourly CDF datasets (imbedded table
sistent cooling and heating circulating pump loads throughout the in Fig. 7[b]) where for 50%, 90% and 95% of time during 2018 the
year. This persistence is due to the main cooling and heating loops DR capability of the building was equal to or larger than 99 kW,
that have to be maintained all year round (see 2.3) and the per- 61 kW and 54 kW respectively.
sistence of secondary cooling circulating pumps all throughout the Fig. 8 illustrates how a notable rise in DR capability of the
year is primarily due to cooling duties for 13 server rooms (serving building emerges during office hours. Most building services assets
around 2400 networked devices of which around 10 0 0 are PCs and are scheduled to operate between 0800–1700 hrs, during which
laptops). Table 3 outlines statistical description of active powers a consistently high proportion of building electrical load remain
from nominated DR loads, which in aggregation account for 53% available for DR. The average magnitude of office hour deferrable
of the building’s annual electricity consumption. load is 141 kW. In 2018, the office-hour DR capability of the build-
Given that original data had an average frequency of about ing fell below 50 kW for only 11 h (or 0.3% of time) when the
6.6 s, a longer time step needs to be specified to enable syn- building was still able to offer on average 46 kW of demand re-
chronised data assessment and more importantly data aggregation duction. A VPP analogy (as outlined in 2.2) can better illustrate
across a full year. To that end half hourly intervals are chosen to and summarise the 4-hour DR capability of the building whereby
aggregate different data streams. Fig. 6[a] illustrates magnitude of at any point of time the prevailing half hourly deferrable load is
HH deferrable electrical loads that could be shed in a DR event regarded as the initial power (kW) and the sum of subsequent 4-
across 2018 annual cycle in the form of monthly boxplots. When hour deferrable loads is regarded as the stored energy (kWh) of
examined at HH intervals, the average magnitude of deferrable a VPP component. This allows the annual instances of deferrable
loads during 2018 was 109 kW. Expressed as a percentage of total loads to be more concisely expressed with minimum, mean and
electrical demand, on average deferrable loads form 46.4% of to- upper quartile instances of initial power being used to cluster all
tal building electricity consumption (Fig. 6[b]). Whether expressed 2018 deferrable values. As illustrated in table 4, at minimum the
in units of original data (kW) or as a percentage of building to- building has the characteristics of a 50 kW (564±86 kWh) VPP as-
tal demand, a perceptible rise exists in peak summer months and set for 99.7% of office hours. Similarly for 54.2% of time in 2018 the
10 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794

Fig. 6. Boxplots of aggregated [a] and percentage of total [b] deferrable electrical loads. Source data: 2018 metered values averaged at half hourly intervals (connect lines
and data labels: mean values).

Fig. 7. Histogram [a] and CDF [b] of half hourly instances of aggregated deferrable loads.

Fig. 8. Boxplots of HH deferrable loads in magnitude [a] and percentage of total [b] grouped across 24hr daily cycle.

building’s average office-hour DR capability was 141 kW with 640± In summary, lighting, AHUs, heat pumps, dry air cooler fans
68 kWh of stored energy available across the ensuing 4 h. For all together with all circulating pumps and lifts (accounting for 53%
instances of time in 2018, the building could be characterised by of annual electrical load in our target building) can form a non-
a minimum of 28.4 kW of initial power and 401±117 kWh of en- critical DR capability that can be shed using a ‘machine to ma-
suing 4-hr stored energy. For a reduced 24.6% of annual time, the chine’ operational supervision (automated for instance via BMS).
building could offer 138 kW of initial power and 625±18 kWh of The advantage of using building loads to build up DR capabilities
4-hr stored energy. (in addition to raising awareness of building energy consumption)
is also that unlike utility level renewable generation and storage
 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794 11

Table 4
Initial power (IP) and subsequent 4-hr stored energy (SE) of the case study building characterised as a virtual power
plant using 2018 metered data (figures round to whole numbers for clarity).
a b
Office hours Out of office hours Full annual cycle
c
IP (kW)SE (KWh) Persistence IP (kW)SE (KWh) Persistence IP (kW)SE (KWh) Persistence

min 50(564 ± 86) 99.7% 28(344 ± 13) 100% 28(401 ± 117) 100%
Mean 141 (640 ± 68) 54.2% 86(407 ± 50) 40.40% 109(571 ± 82) 41.5%
Q3 d 168(684± 64) 27.7% 100(445 ± 53) 23% 138(625 ± 18) 24.6%
a
08:00 – 17:00.
b
17:00 – 08:00.
c
Percentage of office/non-office/annual time that the stated VPP capabilities could be offered in 2018.
d
Upper Quartile (75% percentile) of 2018 deferrable loads.

Fig. 9. Boxplot of active power ramp rates across weekdays at averaged HH [a] and 7 second [b] intervals (Source data: 2017–2019 metered active power).

(i.e. wind turbines, PV arrays or utility level batteries), no phase gregated levels can be detected (Fig. 9[a]). Fig. 10 is a histogram
imbalance follows when a building plant is switched on and off. of early morning ramp rates during 06:30–07:30 hrs with plot-
ted data limited to 3 standard deviation of all original data points
3.3. Post-DR power surge (i.e. ±41.58 kW/7 s). While this is the most likely period of time
when HVAC assets and their auxiliaries with 578.69 kW of nomi-
As discussed in 1.1.3, on termination of a DR event, in particular nal capacity are started up, in practice for 99.7% of instances (or
if a cluster of buildings on the same network are utilised, sudden the aforementioned 3 St. Dev.) the ramp rate of the building is
power surges can occur that jeopardise system reliability. This for within ±41.58 kW/7 s. It is therefore reasonable to conclude that
instance can be caused by all HVAC equipment in one or a cluster sequential delays in plant start up after a DR event can also avoid
of buildings returning on boost mode to restore IAQ and thermal excessive power surges on the supply network resulting from a
comfort. However well-insulated nature of modern buildings such single building. Clusters of buildings powered by the same net-
as this case-study and its thermal mass results in extremely slow work clearly need coordination at district level which is beyond
thermal drifts in winter, and acceptable drifts in summer (even in the scope of this work.
most vulnerable zones). Therefore the restoration of HVAC plants
would not trigger a boost operation as building thermal environ- 3.4. External temperature as DR predictor
ment is almost guaranteed to have stayed within the comfort band
during a DR event. Table A1 outlines all the assets considered for Any independent variable that can be used to predict the DR
DR in this work, which collectively have a combined rated power capability of the building is beneficial in understanding building’s
of 701.7 kW. This table also illustrates how a 30 second time de- energy behaviour and assisting the management of DR activities.
lay is deployed on all HVAC items to prevent excessive plant start- Most building energy forecast methods rely on external tempera-
up currents. While all HVAC plants are set to start at 08:00 hrs, ture (To ) for load projections, and a strong correlation between To
heat pumps use an optimum start algorithm to get corresponding and building energy consumption exists in all climatic conditions
zones to target temperature by 8am. Fig. 9 illustrates actual Sep [51] . With To as an independent variable, it is possible to see how
2017 to Dec 2019 early morning building active power ramp rates closely building deferrable loads can be assumed to be a tempera-
(derived from Eq. (4)). Fig. 9[a] is a boxplot of ramp rates with data ture dependant variable by constructing R2 values within a linear
aggregated at half hourly intervals that shows the greatest rise in regression model. A strong dependency between these two can en-
power demand occurring during 06:30–07:30 hrs. Fig. 9[b] focuses able aggregators to more confidently predict the capability of VPP
more closely on 06:30–07:30 hrs by producing ramp rates from resources at their disposal. Using 2018 half hourly data, on the
data recorded every 7 second (plotted into boxplots at 2-minutes whole To is a very weak linear predictor of the case study build-
frequencies). As evident, during 28 months of building operation, ing’s DR capability (explaining only 1.96% of variations (i.e. R2 =
instances of active power ramp rate greater than 100 kW in HH 1.96%) as in Fig. 11[a]) and a weak linear predictor of the percent-
intervals were limited to 13 (Fig. 9[a]) and 18 at 7 second inter- age of deferrable loads (explaining only 6.9% of variations (i.e. R2 =
vals (Fig. 9[b]). The early morning sequential plant start-up results 6.9%) as in Fig. 11[b]). Fig. 11[a] shows a slight rise in DR capability
in a very gradual rise in power demand which only at HH ag- of the building at the extremities of external temperatures, when
12 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794

Fig. 10. Histogram of early morning building ramp rates across 3 standard deviations from the mean (Source data: 2017–2019 metered active power).

Fig. 11. External air temperature as a predictor of aggregated [a] and percentage of total [b] deferrable loads. (Data resolution: half hourly).

heat pumps and associated auxiliary components need to work ing mode, heat pumps are able to be used as a DR resource,
harder to maintain an adequate indoor climate. This specifically making their availability independent from To within a linear
explains why a linear regression model (that utilises the mean of model.
temperature) cannot predict DR magnitude since at high and low It is important to note that the relationship between response
external temperatures DR capability is boosted. Closer examination (i.e. DR magnitude) and predictor (i.e. To ) was examined using a
is possible by developing R2 values for all sub-categories of de- linear regression model. A linear model seeks to explain the vari-
ferrable loads with To acting as the predictor (Table 5). To is the ability of the response by the ‘mean’ of the predictor. Future work
best predictor for deferrable loads from dry air cooler (DAC) and can examine none-linear regression methods that may more suc-
its auxiliary circulating pumps (R2 = 30.4%), followed by secondary cessfully link the case-study building’s DR capabilities with exter-
circulating pumps in cooling (R2 = 12.8%) and heating (R2 = 10.8%) nal temperature.
circuits. Electrical consumption of lifts as a deferrable load pre-
dictably shows near-complete independence from external temper- 3.5. Generalisation and applicability
ature (R2 = 1.11%). But most notably heat pumps (which at 24.7%
have the largest share of total annual electrical demand) are very Case study reports tend to be limited in that their findings
poorly predicted as a DR source by external temperature (R2 = might not be applicable to the wider pool of which they are one
0.05%). This is because whether the building is in heating or cool- specific example. To examine the broader application of results of
 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794 13

Table 5
R2 value of subcategories of deferrable loads (as response) versus external temperature (as predictor).

Lighting HPa AHU H. Pumpsb C. Pumpsc DAC & C.L. pumpsd Lifts

R2 (%) e 4.7 0.05 8.4 10.8 12.8 30.4 1.1

Share of annual total power demand (%) 9.4 24.7 6.9 4.9 0.9 5.4 0.8

Notes:.
a
Bivalent Heat pumps.
b
Primary and secondary circulating pumps (heating).
c
Cooling secondary circulating pumps (the only DR load that requires 5 min advance notice).
d
Dry Air Cooler fans and cooling loop circulating pumps.
e
Constructed using half hourly data points (i.e. n = 17,520) with external temperature as predictor.

Table 6
Energy consumption benchmarks for existing UK buildings for treated floor area (kWh/m2 /yr.).
a
Air conditioned (standard) Air conditioned (prestige) Average USB

Good practice 128 234 181 183.2

Typical practice 226 358 292
a
Total electricity consumption of 2,090,281 kWh (2018) across 11,412 of treated floor area.

this work, the electricity consumption of the case-study building that building designers can achieve substantial resilience in large,
was benchmarked against representative UK figures (source: Table well-insulated buildings with exposed thermal mass and openable
20.1 of [52]). Table 6 illustrates that, if normalised to unit area of windows, allowing comfort criteria to be maintained in free-float
treated space, the case study building in 2018 had similar energy mode and enabling buildings to operate as a VPP asset.
consumption as the average UK air-conditioned building in ‘good Against thermo-visual comfort constraints, 2018 metered data
practice’ category. 2018 data also demonstrates that the building’s showed that the case study building was able to shed its load
electrical consumption was 37.3% lower than that in ‘typical prac- by an average of 109 kW (corresponding to 46.4% of building to-
tice’ category. It could therefore be reasonably assumed that USB tal). Viewed as a VPP deployable over 4-hour per cycle, the build-
represents lower to medium categories of UK commercial buildings ing could offer an initial power (and stored energy) of 28 kW
with air-conditioning facilities. (401 ± 117 kWh) for 100% of time, 109 kW (571 ± 82 kWh)
While this work derived its results from a case-specific exam- for 41.5% of time and 138 kW (625 ± 18 kWh) for 24.6% of the
ination, the findings can more generally describe the substantial time. The minimum initial power available for DR in 2018 was
flexibilities that exists in commercial office buildings. This will en- 28 kW. However the value of a building as a VPP is less evident
able buildings to be viewed as distributed energy resources that in its initial power and more tangible in its stored energy, since
by virtue of their controllable loads and substantial climatic inertia throughout the year the magnitude of 4-hour stored energy (i.e.
can be aggregated to act as heterogeneous yet reliable components 401 ± 117 kWh) remained substantial with the lowest value be-
of a virtual power plant. ing 284 kWh. While electrical storage devices are more suitable for
frequency response that demand instantaneous large initial pow-
ers, buildings (in particular in aggregation) can more confidently
Conclusions
be characterised as sources of stored energy suited to DR and
peak shaving on heavily loaded local distribution networks. Actual
Network operators worldwide and more specifically in devel-
electrical consumption showed that the case study building only
oped economies are seeking to operate their assets in a more dy-
experienced 18 instances of early morning power surge that ex-
namic manner in order to deal with anticipated strains that can
ceeded a ramp rate of 100 kW/7 s. This confirmed that at build-
result from rapid electrification of heating and transport. Using a
ing level a 30 s delay in the start-up of multiple electrical plants
12,500m2 commercial building, this work identified 100 separate
(as currently deployed in the building) is sufficient to avoid post-
electrically powered HVAC assets, 4 lifts and 22 lighting circuits
DR power surges that may jeopardise network integrity. Finally the
that formed deferrable load resources capable of participating in
persistent nature of deferrable loads across all external air tem-
a DR event. The aggregated magnitude of these deferrable loads
peratures meant that a linear regression model remains a very
were assessed using high resolution data across 2018 full annual
weak predictor of aggregate deferrable loads in an air-conditioned
cycle in order to quantify full capability of the case study build-
building, explaining only 1.9% of observed deferrable load
ing when operating as a virtual power plant. The duration of a
variations.
DR event was limited to 4 h as imposed by ASHRAE 55 in or-
By virtue of their controllable loads, distributed nature and in-
der to limit excessive thermal drifts. In the absence of HVAC op-
ternal climate inertia, buildings have substantial operational flexi-
eration and providing windows can be opened, simulation results
bilities that can be exploited to relieve grid constraints and reduce
showed that CO2 concentration would stay below 10 0 0 ppm across
distribution network congestions. Given that well-coordinated DR
a 4 hour DR event in areas with the highest occupant density. The
events have intangible occupant comfort implications, commercial
installed mechanical ventilation was able to restore CO2 levels to
buildings are well-placed to be operated as stationary electrical en-
below 10 0 0 ppm in 15 min and to pre-DR levels in 45 min. An-
ergy storage assets within a dynamic and interactive energy sys-
nual room temperatures showed that with no HVAC input, thermal
tem.
drifts would exceed the threshold of imperceptibility (0.5 °C/hr)
for only 1.2% of wintertime and 14.5% of summertime instances
in the worst case zones. While thermal drifts in the absence of Declaration of Competing Interest
HVAC input were much lower than 0.5 °C/hr, the zone tempera-
tures could be restored at a rate of +1.23 °C/hr (peak winter) and We wish to confirm that there are no known conflicts of inter-
−1.8 °C/hr (peak summer) in worst case zones. This demonstrates est associated with this publication and there has been no signifi-
14 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794

cant financial support for this work that could have influenced its CRediT authorship contribution statement
outcome. We confirm that the manuscript has been read and ap-
proved by all named authors and that there are no other persons Mohammad Royapoor: Writing - review & editing, Formal
who satisfied the criteria for authorship but are not listed. We fur- analysis, Visualization, Conceptualization, Methodology. Mehdi
ther confirm that the order of authors listed in the manuscript has Pazhoohesh: Data curation. Peter J. Davison: Data curation.
been approved by all of us. We confirm that we have given due Charalampos Patsios: Conceptualization, Supervision, Funding ac-
consideration to the protection of intellectual property associated quisition. Sara Walker: Conceptualization, Funding acquisition.
with this work and that there are no impediments to publication,
including the timing of publication, with respect to intellectual
property. In so doing we confirm that we have followed the regu-
lations of our institutions concerning intellectual property. We fur- Acknowledgement
ther confirm that no aspect of the work covered in this manuscript
has involved either experimental animals or human patients that Funded by Engineering and Physical Sciences Research Council
would have triggered ethical approval of all relevant bodies. We projects EP/P034241/1 and EP/P001173/1. Original energy and envi-
understand that the Corresponding Author is the sole contact for ronmental data courtesy of Urban Observatory (UO) team at New-
the Editorial process (including Editorial Manager and direct com- castle University who developed and maintain UO data backhaul
munications with the office). He is responsible for communicat- and repository platform.
ing with the other authors about progress, submissions of revisions
and final approval of proofs.
We confirm that we have provided a current, correct email
address which is accessible by the Corresponding Author and Appendix
which has been configured to accept email from: Moham-
[email protected]. Table A1

Table A1
Nominal ratings (kW) and historically recorded maximum active power (kW) of all assets considered for DR .

HPs AHUs Lighting circuits Circulating pumps Dry Air Coolers Row Total
Pn 1 PMax,h 2 Pn PMax,h Pn PMax,h Pn PMax,h Pn PMax,h Pn PMax,h Sequential optimum
(kW) (kW) (kW) (kW) (kW) (kW) (kW) (kW) (kW) (kW) (kW) (kW) start-up time

Ground Zone 1 23.9 26.0 3.2 1.68 5 3.75 0.53 0.6 32.6 32 top
Zone 2 23.9 25.3 3.2 2.43 5 5.64 0.60 1.4 – – 32.7 35 

Zone 3 15.9 13.9 3.2 1.68 3 2.43 2.63 1.4 – – 24.7 19 

1st Zone 1 15.9 19.6 3.2 2.30 8 4.82 0.32 0.6 – – 27.4 27 top + 30 s
Floor Zone 2 15.9 14.6 8 2.20 5 2.02 0.24 0.6 – – 29.1 19 

Zone 3 13.3 13.6 3.2 2.05 5 1.68 0.23 0.6 – – 21.7 18 

Zone 4 19.9 22.9 3.2 2.47 8 4.96 0.52 0.6 – – 31.6 31 

2nd Zone 1 15.9 18.9 3.2 2.26 5 3.21 0.42 0.6 – – 24.5 25 top + 60 s
Zone 2 13.3 16.7 3.2 2.50 3 1.21 0.25 0.6 – – 19.7 21 

Zone 3 19.9 24.3 3.2 2.24 3 0.78 0.49 0.6 – – 26.6 28 

3rd Zone 1 15.9 18.4 2.6 2.53 13 11.30 0.51 0.6 – – 32.0 33 top + 90 s
Zone 2 23.9 28.6 4.8 2.65 13 8.69 0.38 0.6 – – 42.1 41 

Zone 3 17.8 21.7 3.2 1.89 5 5.04 0.56 0.6 – – 26.6 29 

Zone 4 17.8 22.2 3.2 2.68 3 2.06 0.49 0.6 – – 24.5 28 

4th Zone 1 13.3 13.9 3.2 1.85 13 9.08 0.26 0.6 – – 29.8 25 top + 120 s
Zone 2 15.9 19.1 3.2 1.79 8 6.54 0.44 0.6 – – 27.5 28 

Zone 3 15.9 21.2 3.2 2.11 3 2.53 0.40 0.6 – – 22.5 26 

Zone 4 17.8 20.9 3.2 1.69 3 0.32 0.40 0.6 – – 24.4 24 

5th Zone 1 17.8 43.1 3.2 3.49 3 3.24 0.70 0.6 – – 24.7 50 top + 150 s
Zone 2 19.9 23.8 3.2 2.19 3 2.94 0.55 0.6 – – 26.6 29 

6th Zone 1 17.8 19.1 3.2 2.48 3 1.88 1.09 0.6 – – 25.1 24 top + 180 s
Zone 2 19.9 24.8 3.2 1.93 3 2.59 1.91 0.6 – – 28.0 30 

Roof – – – – – – 15 14.70 9.3 7.00 24.3 15 top

Plantroom
1
– – – – – – 15 16.33 12.4 13.38 27.4 16 

Plantroom
2
– – – – – – 33 25.45 12.4 13.45 45.4 25 

Plantroom
3
391.5 472.5 76.2 49.11 123 86.70 76.89 71.39 34.1 34
Category
total
Building total 701.7 679.73

Notes:.
1
Pn : Nominal or rated power, equal to the work of compressor and controls (for HPs), supply and extract fan duties and controls (for AHU), rated lighting circuit
power, etc.
2
PMax,h : Maximum instantaneous active power that has been observed historically during Nov 2017 – Dec 2019 data collection period.
3
Observed maximum active power demands of different loads are non-simultaneous and are summed-up for illustration of magnitude only.
 M. Royapoor, M. Pazhoohesh and P.J. Davison et al. / Energy & Buildings 213 (2020) 109794 15

References [27] M. Frontczak, P. Wargocki, Literature survey on how different factors influ-
ence human comfort in indoor environments, Building and Environment 46
[1] U.S. Energy Information Administration (EIA), in: International Energy Outlook, (4) (2011) 922–937.
Washington, DC, 2016, p. 20585. [28] American Society of Heating, R.a.A.-C.E., 2017 ASHRAE Handbook—Fundamentals,
[2] D. Ürge-Vorsatz, et al., Energy end-use: buildings, in: Global Energy Assess- Atlanta, GA, 2017.
ment: Toward a Sustainable Future, T. Global Energy Assessment Writing, Edi- [29] ASHRAE 62.1-2016, Ventilation For Acceptable Indoor Air Quality, Atlanta, USA,
tor, Cambridge University Press, Cambridge, 2012, pp. 649–760. 2016.
[3] International Energy Agency, in: Modernising Building Energy Codes, New [30] Chartered institution of building services engineers, KS17: Indoor Air Quality
York, NYUSA, 2013, p. 10017. & Ventilation (Oct 2011) London.
[4] R. M, et al., Carbon mitigation unit costs of building retrofits and the scope for [31] European committee for standardization, ventilation for non-residential build-
carbon tax, a case study, Energy Build (2019) 109415. ings; performance requirements for ventilation and room-conditioning systems, EN
[5] H.T. Haider, O.H. See, W. Elmenreich, A review of residential demand response 13779, 2008.
of smart grid, Renewable and Sustainable Energy Reviews 59 (2016) 166–178. [32] CIBSE, Guide A; Environmental Design, The Chartered Institution of Building
[6] M. Royapoor, A. Antony, T. Roskilly, A review of building climate and plant Services Engineers, London, 2015.
controls, and a survey of industry perspectives, Energy Build 158 (2018) [33] U. Satish, et al., Is CO(2) an indoor pollutant? direct effects of low-to-moderate
453–465. CO(2) concentrations on human decision-making performance, Environ. Health
[7] M. Gabriel, J. Nathwani, Meeting the texas electricity peak demand conundrum: Perspect. 120 (12) (2012) 1671–1677.
a case for wind and solar, The Electricity Journal 31 (1) (2018) 57–64. [34] G. Newsham, et al., Control strategies for lighting and ventilation in offices: ef-
[8] S. Kiliccote, M.A. Piette, D. Hansen, Advanced Controls and Communica- fects on energy and occupants, Intelligent Buildings International 1 (2) (2009)
tions For Demand Response and Energy Efficiency in Commercial Buildings, 101–121.
Lawrence Berkeley National Laboratory, 2006. [35] G. Newsham, B. Birt, Demand-responsive lighting - A field study, LEUKOS -
[9] E.T. Hazel, et al., The relationship between wind power, electricity demand and Journal of Illuminating Engineering Society of North America 6 (3) (2010)
winter weather patterns in great britain, Environmental Research Letters 12 (6) 203–226.
(2017) 064017. [36] S.W. GDavid, et al., Strategies for demand response in commercial buildings,
[10] European network of transmission system operators, Nordic Winter Power Bal- in 2006 ACEEE Summer Study on Energy Efficiency in Buildings. Pacific Grove,
ance Forecast 2016-2017, ENTSOE, 2016. CA (2006) 13–18.
[11] F.M. Andersen, et al., Households’ hourly electricity consumption and peak de- [37] Health and Safety Executive, Lighting At Work, HSE publications, Richmond,
mand in denmark, Appl Energy 208 (2017) 607–619. Surrey, 1997 TW9 4DU.
[12] The UK National Grid. Frequency response services. 2017 [cited 2018, [38] S. Rotger-Griful, et al., Green lift: exploring the demand response potential of ele-
3 Jan 2018]; Available from: https://www.nationalgrid.com/uk/electricity/ vators in danish buildings, Energy Research & Social Science 32 (2017) 55–64.
balancing-services/frequency-response-services. [39] Y. Guo, et al., Machine learning-based thermal response time ahead energy de-
[13] J.H. Yoon, R. Baldick, A. Novoselac, Dynamic demand response controller based mand prediction for building heating systems, Appl Energy 221 (2018) 16–27.
on real-time retail price for residential buildings, IEEE Trans Smart Grid 5 (1) [40] S. Rotger-Griful, U. Welling, R.H. Jacobsen, Implementation of a building energy
(2014) 121–129. management system for residential demand response, Microprocess Microsyst
[14] J. Qi, et al., Demand response and smart buildings: a survey of control, commu- 55 (2017) 100–110.
nication, and cyber-physical security, ACM Trans. Cyber-Phys. Syst. 1 (4) (2017) [41] L. Shen, Z. Li, Y. Sun, Performance evaluation of conventional demand response
1–25. at building-group-level under different electricity pricings, Energy Build 128
[15] P. Cappers, C. Goldman, D. Kathan, Demand response in U.S. electricity markets: (2016) 143–154.
empirical evidence, Energy 35 (4) (2010) 1526–1535. [42] Y. Liu, et al., Coordinating the operations of smart buildings in smart grids,
[16] Commission of the European Union, Directive 2012/27/EU of the european parli- Appl Energy 228 (2018) 2510–2525.
ment and of the council of of 25 october 2012 on energy efficiency. 2012. [43] B. Parrish, R. Gross, P. Heptonstall, On demand: can demand response live up to
[17] Department for Business, E.I.S., Realising the potential of demand-side response expectations in managing electricity systems?, Energy Research & Social Science
to 2025: a focus on small energy users. Nov 2017. 51 (2019) 107–118.
[18] S. Rotger-Griful, et al., Demand response potential of ventilation systems in [44] H. Wang, S. Wang, R. Tang, Development of grid-responsive buildings: opportu-
residential buildings, Energy Build 121 (2016) 1–10. nities, challenges, capabilities and applications of hvac systems in non-residential
[19] X. Xue, et al., A fast chiller power demand response control strategy for build- buildings in providing ancillary services by fast demand responses to smart grids,
ings connected to smart grid, Appl Energy 137 (2015) 77–87. Appl Energy 250 (2019) 697–712.
[20] D. Christantoni, et al., Implementation of demand response strategies in a mul- [45] K.O. Adu-Kankam, L.M. Camarinha-Matos, Towards collaborative virtual power
ti-purpose commercial building using a whole-building simulation model ap- plants: trends and convergence, Sustainable Energy, Grids and Networks 16
proach, Energy Build 131 (2016) 76–86. (2018) 217–230.
[21] M.A. Piette, S. Kiliccote, J.H. Dudley, Field demonstration of automated demand [46] Newcastle University. Urban observatory. 2019 [cited Jan- Dec 2018; Available
response for both winter and summer events in large buildings in the pacific from: http://www.urbanobservatory.ac.uk/.
northwest, Energy Efficiency 6 (4) (2013) 671–684. [47] T. Cover, P. Hart, Nearest neighbor pattern classification, IEEE Trans. Inf. Theor.
[22] Price, P.N., Nathan J Addy, and Sila Kiliccote„ Predictability and persistance of 13 (1) (2006) 21–27.
demand response load shed in buildings. 2015: LBNL-187399. [48] HM Government, Approved Document L2A, Conservation of Fuel and Power in
[23] R. Yin, et al., Study on auto-dr and pre-cooling of commercial buildings with New Buildings and Other Dwellings, Edition, NBS, London, 2013.
thermal mass in california, Energy Build 42 (7) (2010) 967–975. [49] M.R. Maryam Zirak, Thomas Gilbert, Cross-platform energy modeling for scal-
[24] S. Aghniaey, T.M. Lawrence, The impact of increased cooling setpoint tempera- able urban energy simulation: a case-study, IAPE, 19, 2019.
ture during demand response events on occupant thermal comfort in commercial [50] L.T. Doulos, et al., Harvesting daylight with led or T5 fluorescent lamps? the
buildings: a review, Energy Build 173 (2018) 19–27. role of dimming, Energy Build 140 (2017) 336–347.
[25] J. Kolarik, et al., Occupant responses and office work performance in environ- [51] D.H.W. Li, L. Yang, J.C. Lam, Impact of climate change on energy use in the
ments with moderately drifting operative temperatures (RP-1269), HVAC and built environment in different climate zones – A review, Energy 42 (1) (2012)
R Research 15 (5) (2009) 931–960. 103–112.
[26] ANSI/ASHRAE Standard 55, Thermal Environmental Conditions For Human Oc- [52] CIBSE, Guide F; energy efficiency in buildings. 2012, The Chartered Institution of
cupancy, Atlanta, Georgia, 2010. Building Services Engineers: London.