Sustainable Cities and Society 35 (2017) 816–829

Contents lists available at ScienceDirect

Sustainable Cities and Society

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

Engineering advance

Upscaling energy control from building to districts: Current limitations and MARK
future perspectives
⁎
Jonathan Reynolds , Yacine Rezgui, Jean-Laurent Hippolyte
BRE Trust Centre for Sustainable Engineering, Cardiff University, Cardiff, UK

A R T I C L E I N F O A B S T R A C T

Keywords: Due to the complexity and increasing decentralisation of the energy infrastructure, as well as growing pene-
Smart controls tration of renewable generation and proliferation of energy prosumers, the way in which energy consumption in
Model predictive control buildings is managed must change. Buildings need to be considered as active participants in a complex and wider
Energy district-level energy landscape. To achieve this, the authors argue the need for a new generation of energy
Smart grid
control systems capable of adapting to near real-time environmental conditions while maximising the use of
HVAC control
renewables and minimising energy demand within a district environment. This will be enabled by cloud-based
Building energy management
District energy management demand-response strategies through advanced data analytics and optimisation, underpinned by semantic data
Microgrid control models as demonstrated by the Computational Urban Sustainability Platform, CUSP, prototype presented in this
paper. The growing popularity of time of use tariffs and smart, IoT connected devices offer opportunities for
Energy Service Companies, ESCo’s, to play a significant role in this new energy landscape. They could provide
energy management and cost savings for adaptable users, while meeting energy and CO2 reduction targets. The
paper provides a critical review and agenda setting perspective for energy management in buildings and beyond.

1. Introduction renewable generation and be closer to the communities that it supplies

allowing local resources to be maximised and efficiency to be improved
Reducing global energy demand and the greenhouse gases that stem (Koirala, Koliou, Friege, Hakvoort, & Herder, 2016). In this new energy
from energy production is one of the most vital technological chal- paradigm, more intelligent control systems are essential to manage the
lenges that the world currently faces. Given that buildings represent a variable, stochastic renewable resources. The old model of demand led
significant proportion of global energy consumption, 40% in the EU energy supply can no longer work in this scenario. Instead we must aim
(European Parliament, 2010), this is a pivotal sector to target for effi- to become more supply led, shifting our energy demands to times when
ciency savings. Key components within buildings such as the HVAC and the supply is available (Ameri & Besharati, 2015; Di Somma et al., 2015;
lighting systems are estimated to account for around 55% and 15% of Ondeck, Edgar, & Baldea, 2015). This can be achieved through demand
all building energy consumption although this is dependent on building response, DR, or demand side management, DSM, techniques. It has
uses and country (Pérez-Lombard, Ortiz, & Pout, 2008). However, re- been estimated that 93 GW of energy consumption can be brought
cent research has shown that a significant proportion of this energy forward and 247 GW delayed across Europe if we have smart enough
consumption is wasted due to poor management by controller systems controllers to exploit this (Gils, 2014). To achieve this vision of a smart
(Hu, Weir, & Wu, 2012). It has been estimated that between 20% and grid, fast, reliable communication and intelligent, autonomous decision
30% of building energy consumption can be saved through optimised making is required between prosumers, storage devices, the central grid
operation and management without changing any hardware within the and end users. If implemented this could reduce electricity demand and
building (Guan, Xu, & Jia, 2010). CO2 emissions by 10% to 15% (Mourshed et al., 2015).
To tackle climate change there has been a recent surge in con- Fortunately, volumes of research is being carried out into intelligent
struction of renewable energy resources such as solar and wind power building controls which have great potential to reduce energy con-
(Lins, Williamson, Leitner, & Teske, 2014). It is predicted that the future sumption and participate in demand response mechanisms. Current
energy infrastructure will be much more decentralised, moving away traditional building controls use Proportional or Proportional-Integral,
from the current fossil fuel, central grid system in place (Kolokotsa, PI, control. Rather than operating fully on or fully off, proportional
2016). This new energy infrastructure will have a greater share of control measures the error between desired value and the measured

⁎
Corresponding author.
E-mail addresses: Reynoldsj8@cardiff.ac.uk (J. Reynolds), RezguiY@cardiff.ac.uk (Y. Rezgui), HippolyteJ@cardiff.ac.uk (J.-L. Hippolyte).

http://dx.doi.org/10.1016/j.scs.2017.05.012
Received 13 February 2017; Received in revised form 16 May 2017; Accepted 17 May 2017
Available online 21 May 2017
2210-6707/ © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/).
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

value and gives a proportional output response signal. PI control adds party with intelligent analytics to save the consumer money. There is
an integral term to remove permanent gains found in proportional only also emphasis that the scheduling strategies need to use demand re-
control and this is the most common system currently used in HVAC sponse strategies. In Zucker, Habib, Blöchle, Wendt, and Schaat (2015),
controllers (CIBSE, 2009). However modern advances in computational a cognitive architecture for building control is introduced. Cognitive
intelligence could allow direct digital control. Microprocessors could be algorithms aim to mimic the way the human mind thinks through
programmed with artificial intelligence to provide advanced, adaptable storing experiences and basing decisions around previous actions.
and more optimal control. Furthermore, given the decentralisation of An intelligent BMS is simulated in Missaoui, Joumaa, Ploix, and
energy supply there is a requirement for building controllers to be more Bacha (2014). The BMS is assumed to control household appliances, the
aware of their environment and adaptable to local circumstances to heating, local PV resources and sense outdoor conditions. A grey box,
make best use of local generation capacity. The increased use of time of resistor-capacitance, RC, thermal model of a small house is created to
use tariffs or real time pricing gives an opportunity for engaged con- simulate the indoor temperature depending on the heating strategy. By
sumers to achieve real energy cost savings and in-turn help the grid shifting the operation of controllable appliances, the smart BMS made
reduce its peak demands. significant cost savings. Smart household scheduling is also addressed
This paper will critically review a number of peer-reviewed research in Chen, Wang, Heo, and Kishore (2013). This also controls the heating
papers on the topic of building energy management (Section 2) and and controllable appliances but uses a Model Predictive Control, MPC,
wider scale district energy management (Section 3). An assessment of technique and utilises electrical battery storage which it theorises could
how integrated district and building level control are will be made as come from a plug-in electrical vehicle. The solution is simulated in
well as a number of proposed future research directions in Section 4. several climates and is shown to save the user up to 20%. However, this
strategy assumes perfect weather forecasting so the saving, in reality, is
2. Building energy management strategies likely to be reduced. In Yuce, Rezgui, and Mourshed (2016), Yuce
combine the use of an Artificial Neural Network, ANN, and a Genetic
Most large complex buildings will be equipped with a Building Algorithm, GA, to schedule domestic appliances to ensure maximum
Management System, BMS. These sense conditions in building zones utilisation of local renewable resources. The use case presented is based
and are programmed with internal logic to take action using various on a small holiday home in Southern England which has onsite PV and
actuators depending on the conditions they perceive and the time of wind production. When grid energy reductions of 10%, 25%, and 40%
day. However, traditional BMS follow fairly static rules without the are imposed on the building, the idle renewable generation is clearly
intelligence to try new, potentially more optimal, strategies. For ex- reduced.
ample instead of turning on the heating at 8am to have the building at
the appropriate temperature by 9am, could the building be pre-heated 2.2. Optimal HVAC control
to avoid a morning spike and possibly reduce overall daily energy
consumption? Can occupancy levels be better predicted and sensed to Given that HVAC systems account for a proportionately large share
ensure that zones are only heated when necessary? Is the optimal of building energy consumption a large amount of the literature focuses
strategy dependant on the outdoor conditions and does that need to be on better management of these systems. These studies often include
sensed more? These are the typical questions facing the development of setting the optimal start and end times for heating or cooling, a focus on
BMS which are vital for reducing energy consumption and improving sensing occupancy to improve efficiency, utilising the building thermal
comfort within buildings. Lee and Cheng (2016) reviewed the impact of mass, and operational control of the components that make up the
BMS over 35 years and found that during this period energy savings HVAC system.
from BMS have increased from 11.39% to 16.22%. However a key
challenge for the future BMS is the availability, cost and quality of 2.2.1. Model predictive control management strategies
sensors as well as the data management problems that arise from the The dominant HVAC control method found within the literature is
increased sensing (Kumar et al., 2016). An excellent review of BMS is Model Predictive Control, MPC. A number of building control techni-
provided by De Paola, Ortolani, Re, Anastasi, and Das (2014), in which ques were assessed in Shaikh, Nor, Nallagownden, Elamvazuthi, and
the author sets out the ideal BMS and how close current technologies Ibrahim (2014) and it concluded that MPC was an excellent strategy for
are to that ideal. managing building systems due to its ability to adapt to disturbances,
exploit thermal mass, take account of price variations and shift load.
2.1. Home energy management systems MPC aims to optimise a decision variable (e.g. temperature set point)
over a time horizon whilst considering outside disturbances such as
Whilst most current BMS are installed in larger commercial build- outdoor temperature. It uses an internal model of the controlled system
ings, smart energy management in residential buildings is also very to be able to predict the outcome of its actions. It then re-evaluates at
important and a growing area of research. We are already seeing ‘smart’ shorter time steps to be able to adapt to incorrect forecasts or unfore-
thermostats like Nest (2016), which aim to learn users preferences and seen disturbances (Afram & Janabi-Sharifi, 2014; Kwadzogah, Zhou,
patterns to save energy as well as linking up with other smart devices Li, & Member, 2013). The model within MPC is obviously essential and
using the Internet of Things, IoT. These products along with smart this remains disputed within the literature. White box models are de-
meters, which are increasingly being rolled out by energy suppliers tails physics based models such as Energy Plus or TRNSYS. These have
(DECC, 2014), can also interact with users smart phones with the hope the advantage of being highly detailed and customisable but have very
of engaging the consumer and encouraging behavioural change. Com- long running times. Grey box models such as RC models are simplified
mercial systems are relatively new and the literature provides a number physics models tuned using real data. Black box models such as ANN
of suggestions for possible system architectures. Capone, Barros, have no knowledge of the physical make-up of the system and are
Hrasnica, and Tompros (2009) presents the AIM gateway which pro- trained purely on available data. Unlike white box models they are
poses a communication architecture for devices and sensors within the quick and not computationally demanding but do require large volumes
home. The authors suggest that this could provide better monitoring of data (Li & Wen, 2014). The complexity of the model is also a key
and prediction for energy suppliers through user profiling. Son, consideration. It is required to accurately simulate the building in
Pulkkinen, Moon, and Kim (2010), Zhou et al. (2016) give a similar question whilst be as simple as possible to reduce computational time.
vision for a Home Energy Management System with a series of devices Privara, Vana, Zacekova, and Cigler (2012) gives a methodology to
connected to a smart meter or controller. They suggest that the man- provide the least complex yet adequately accurate building model for
agement and scheduling of these devices could be outsourced to a third MPC.

817
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

MPC has been applied to a Czech university building in Prívara, Jan, they are tuned for. For these types of rule based controls to succeed on a
Ferkl, and Cigler (2011), Široký et al. (2011). This MPC strategy took wider scale they must be more intelligent, autonomous and learn from
weather, occupancy, and price prediction into account and modelled the conditions they perceive and the actions that they take. For ex-
the thermal conditions of the building with an RC model. In a 3 month ample, they must predict the conditions to assess whether pre-cooling is
trial the strategy achieved a 17% to 27% reduction in energy con- a suitable strategy for that day. An intelligent, rule based, decision
sumption. Traditional MPC was adapted in Oldewurtel et al. (2012) to support system was created in Doukas, Patlitzianas, Iatropoulos, and
take into account the uncertainties in the forecasts that it uses. The Psarras (2007) to control indoor temperature, humidity, luminosity and
resulting Stochastic MPC reduced energy consumption and gave less air quality. It had a stored knowledge base and a wide array of sensors
comfort violations than traditional rule based control or normal MPC. allowing it to choose the best course of action for the specific condi-
Mahendra, Stéphane, and Frederic (2015), also aims to address the tions. This led to a reduction in energy consumption of around 10% in a
problems that stem from forecasting uncertainties. This solution runs a real trial. A care facility in the Netherlands is modelled and controlled
reactive algorithm in between the MPC time steps that can take swift in Yuce and Rezgui (2015). A validated Energy Plus model is used to
action if the forecasts are clearly incorrect due to an unexpected spike generate vast quantities of training data which is used to train an ANN.
in occupancy for example. A genetic algorithm then chooses the optimal rules through semantic
Aside from the use of RC models, Ma, Qin, Li, and Salsbury (2011) mapping based on environmental conditions, the time and date.
used an Energy Plus model of a fairly simple one storey building which As has already been alluded to throughout this paper so far, there is
was coupled with a MATLAB MPC procedure using the Building Con- a growing use of computational intelligence techniques to tackle the
trols Virtual Test Bed, BCVTB, as a middleware. This strategy found pre- problems of building energy management. Increasingly machine
cooling was effective to shift load off peak to cheaper times hence learning methods such as ANN can be used to predict important vari-
saving money over traditional strategies. Salsbury, Mhaskar, and Qin ables. They can also be tuned to continue learning and adapting after
(2013) also exploited the thermal mass of the building using a MPC their initial creation using sliding window training techniques (Chae,
algorithm that considered price fluctuations throughout the day. A Horesh, Hwang, & Lee, 2016). An excellent review of computational
genetic algorithm is utilised in Molina, Lu, Sherman, and Harley intelligence techniques such as machine learning, fuzzy logic, multi-
(2013), in conjunction with a state space model to optimally control an agent systems and metaheuristic algorithms, applied to HVAC control
ideal heating and cooling system. It aims to balance the economic cost can be found in Ahmad, Mourshed, Yuce, and Rezgui (2016). ANN have
of energy and discomfort measured using Predicted Percentage Dis- been used in Moon and Kim (2010) to maximise occupant comfort by
satisfied, PPD, however fails to find a solution where both criteria are controlling indoor temperature and humidity. The ANN could predict
improved. Ferreira, Silva, and Ruano (2012), aimed to control both future indoor conditions based on current indoor and external condi-
temperature and humidity to minimise the energy consumption and tions and then take appropriate action. Yamada (1999) also used neural
maintain conditions within an acceptable Predicted Mean Vote, PMV, networks to calculate predicted PMV. It then aimed to maintain PMV
range. The strategy was deployed in a classroom and was found to at ± 0.3 rather than 0 and this reduced energy consumption by 18%.
activate the AC unit far fewer times giving a predicted energy saving of Liang and Du (2005), also created a neural network controller that used
over 30%. An explicit MPC strategy was developed in Parisio, Fabietti, PMV as its control objective rather than temperature or humidity. The
Molinari, Varagnolo, and Johansson (2014) to control both CO2 and ANN measures the error between set point and measurement. The error
indoor temperature whilst minimising energy costs. is used as an ANN input and it outputs an appropriate control action. It
In Figueiredo and Sá da Costa (2012) the authors aimed to develop was found to provide more stable control and provide better comfort for
an intelligent MPC layer above the traditional SCADA based building occupants.
controls. Current SCADA systems lack the computational power to carry Papantoniou uses a number of computational intelligence techni-
out any data analytics or advanced control strategies. The paper pro- ques in Papantoniou, Kolokotsa, and Kalaitzakis (2015). ANN are used
poses that the interactive layer holds the intelligence and the legacy to predict the outdoor and indoor air temperature over the next 8 h.
SCADA system is used to sense and actuate. A decentralised approach Then a genetic algorithm and real time fuzzy logic control are used to
has been used in Moroşan, Bourdais, Dumur, and Buisson (2010). This set fan coil operation and speed within a hospital. A Web-based dash-
allows independent MPC controller to be deployed in individual zones board is also set up to display several variables over a given time range
which are not computationally complex or time consuming. The heat for technical hospital staff. Whilst this is only trialled on 3 simulated
transfer that is likely to be transferred from one zone to another is rooms, if extrapolated to the whole hospital energy savings of up to
communicated to the relevant zone controller. This is then used to 35% are predicted. A fuzzy logic controller based around the control of
avoid temperature overshoots compared to independent zone level PMV is developed in Gouda, Danaher, and Underwood (2001). The
controllers. MPC can be formulated to also consider predictions of local authors argue that fuzzy logic controllers are more adaptable to dif-
renewable generation and variable energy prices to allow greater ferent building conditions compared to traditional PID controllers.
saving over traditional rule based control. This was applied in Lee, Whilst PID controllers work well for the conditions they have been
Horesh, and Liberti (2015) and energy savings of 15% to 30% were tuned for, they perform poorly if building characteristics change. The
achieved over basic cooling strategies. controller developed in this paper outperforms the PID controller in
these conditions. Building operation and HVAC control is highly de-
2.2.2. Other optimisation strategies pendent on occupancy patterns. Pisello, Bobker, and Cotana (2012)
Whilst MPC is the dominant control strategy found in the literature, shows that simply by understanding and auditing the occupancy pat-
other smart scheduling or rule based controls can be found. Older terns within the building you can alter the HVAC schedules which
HVAC energy minimisation methods were based on full audits of the provide significant energy savings compared to when the building is
system and then applying different control rules to try and reduce en- first commissioned. An occupancy based optimisation strategy is pro-
ergy consumption. An example of this can be found in Mathews, Arndt, posed in Erickson and Cerpa (2010). This pays particular attention to
Piani, and van Heerden (2000), where a number of additional rules and sporadically occupied zones to achieve energy savings of around 20%.
fine tuning around start and stop times could lead to a 53% reduction in
energy consumption. The building thermal mass is utilised in Lee and 2.2.3. Operational control
Braun (2008) by pre-cooling to avoid on-peak energy consumption. The Slightly beyond the scope of this review, but nevertheless important
building is held at a lower temperature until 1 p.m. and allowed to for the reader to be aware of, is a class of operational level optimisation
linearly increase during the peak energy period. However, these types strategies. These tend to focus of specific HVAC components and ensure
of rule based controls will only work for the particular room or building that they are run at maximum efficiency. HVAC components have very

818
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

complicated interrelationships and balancing the COP (Coefficient of focussed on domestic appliances. Given the increasing electrification of
Performance) of the different components is not necessarily intuitive. heating in residential properties, through rising popularity of devices
The chilled water and supply air temperature set points are optimised like heat pumps, thermal control and appliance control will need to
using evolutionary algorithms in Fong, Hanby, and Chow (2006), Fong, become more integrated in the future. Even with gas based heating
Hanby, and Chow (2009). MPC is used to control set point of shared, systems, HVAC systems consume electricity through fans and pumps so
University campus, cooling towers, chillers and thermal storage in Ma holistic management of electricity is likely to be advantageous in most
et al. (2012) achieving around a 20% increase in overall COP. A genetic future buildings. The increasing interconnection of devices through IoT
algorithm is used in Lu, Cai, Xie, Li, and Soh (2005) to minimise the technology is greatly encouraging in this field. However, users’ privacy
overall energy consumption from fans, chillers, and pumps by con- concerns must be addressed and consumers must be better educated in
trolling the chilled water temperature, cooling coil and chilled water their energy choices and feel as though they are kept informed.
pump pressure, and the sequences of pumps and chillers used. This kind In the authors’ opinion, operational optimisation is something that
of system efficiency optimisation is important in reducing overall will continue to develop alongside the improvement of smart building
building energy consumption and can be improved in conjunction with control strategies to provide efficiency savings. It is also encouraging
intelligent, wider scale, building control strategy found in the other that the vast majority of reviewed work considers outside influences
sections of this paper. Interested readers can find more detail in the such as the weather when optimising building energy consumption.
above references and in Komareji et al. (2008), Rehrl and Horn (2011), External conditions have a large influence on the ideal building control
Xu and Li (2007), Yuan and Perez (2006), Yang and Wang (2012). strategy and therefore systems must move away from static, rule-based,
strategies that carry out the same actions each day regardless of con-
2.3. Multi-variable zone level optimisation ditions.

A more detailed class of building optimisation can also be found in 3. District level energy management
the literature that aims to not only to control heating or cooling devices
but also ventilation and lighting systems. Due to the complexity that As alluded to in the Section 1, the energy landscape and infra-
comes from managing these highly coupled systems, they tend to be structure is changing. Energy generation is becoming increasingly de-
focussed at a room or zone level rather than an entire building level. An centralised to take advantage of local energy resources. This is partly
example can be found in Ahmad, Hippolyte, Reynolds, Mourshed, and due to the concept of the ‘prosumer’, one who both produces and
Rezgui (2016), which aims to minimise energy consumption of a UK consumes energy, gaining popularity. Furthermore, district heating
classroom whilst maintaining good indoor air quality, thermal and vi- systems are gaining popularity partly due to improved efficiencies that
sual comfort using a genetic algorithm. The decision variables included come through use of combined heat and power, CHP, units. These
the operation of a window blind, window opening and ventilation unit generate electricity but can utilise the heat produced in a district
operation. Mossolly, Ghali, and Ghaddar (2009), aimed to control the heating network. Waste heat from industrial processes and waste in-
fresh air flow rate depending on the CO2 levels within the room. A GA cineration plants could also be exploited with these systems. Managing
was used to minimise a cost function comprising three components, the energy resources at a wider district level rather than a building level
thermal comfort, the energy consumption and the indoor air quality. A could provide greater flexibility, energy and cost savings to all users.
study by Kim, Jeon, and Kim (2016) combines the use of Energy Plus Buildings of different uses, e.g. residential and commercial, could have
simulations, artificial neural networks (ANN) and a GA to control a mutually beneficial demand profiles and could therefore be better
blind slat angle and the HVAC operation to satisfy thermal and visual managed in a ‘microgrid’ setting. This does encourage the development
comfort. A Langrangian relaxation – Dynamic programming technique of district level controllers to optimally manage the energy flows and
was used in Sun et al. (2013). The study aimed to control the HVAC, does require building level controllers to be more perceptive of their
lighting, shading and natural ventilation systems by reducing the pro- neighbours. Clusters of buildings also have more chance to collectively
blem and solving sub problems. Their solution was found to save up to participate in grid DR events. Single buildings can rarely shift enough
9.3% of energy cost when compared to conventional strategies. A fur- load to qualify for these types of events but a group of collectively
ther paper, (Yan et al., 2014), builds on this work by considering a managed buildings may have more bargaining power with the grid
combined cooling, heat and power unit, CCHP, and energy storage (Aduda, Labeodan, Zeiler, Boxem, & Zhao, 2016).
capacity. Yang, Wang, and Wang (2011) presents a GUI platform to
manage lighting, CO2 and temperature within a building whilst mini- 3.1. Optimal operation of a microgrid
mising energy cost using a particle swarm optimisation, PSO, algo-
rithm. Much of the district level energy optimisation is focussed around the
supply side. Many studies consider a cluster of buildings with shared
2.4. Discussion energy production units, some of which may be uncontrollable re-
newable sources like solar or wind, energy storage capacity and a de-
Many of the optimisation papers discussed so far are summarised in mand source. Staino, Nagpal, and Basu (2016), demonstrated a co-
Table 1. It is immediately clear that there is no real consensus around operative model predictive control, MPC, framework to manage the use
which type of building model to use in building control optimisation. of a shared heat pump between a cluster of buildings. A cooperative
RC and ANN appear to be the leading candidates and the best choice is algorithm takes advantage of greater flexibility in the larger buildings
likely to be dependent on a number of case by case factors such as the to reduce overall district cost. However, despite an overall reduction in
availability of data and the optimisation method used. Table 1 also cost the smaller buildings face an energy cost increase raising questions
shows the need for advanced building control strategies to be tested in of fairness and billing. A microgrid equipped with solar PV, batteries, a
the field and implemented in real buildings. Only after this stage can CHP and grid backup is considered in Guan et al. (2010). The optimi-
this work become truly validated and begin to be accepted by the wider sation strategy is based on Mixed Integer Linear Programming, MILP,
public. Furthermore, most of the optimisation strategies reviewed this and maximises the use of solar energy to avoid energy consumption
section do not explicitly consider DR controls. This is a vital research during peak price periods. An ‘islanded’ microgrid is considered in
gap that needs to be filled given global energy trends towards more Marzband, Sumper, Ruiz-Alvarez, Luis Dominguez-Garcia, and
decentralised use of uncontrollable renewable resources. Tomoiaga (2013) with wind power, PV, a CHP and battery storage. The
There is a clear difference between the detailed HVAC control and proposed optimisation reduces energy cost by 15%. MPC is applied to a
the home energy management systems which seem to be particularly smart residential microgrid in Zhang, Zhang, Wang, Liu, and Guo

819
 J. Reynolds et al.

Table 1
Summary of Building Optimisation Literature.

Reference Model Type Aims to Control Objective Disturbances Included Real Case Load Smoothing
Study? Considered?

Missaoui et al. (2014) RC Indoor Temperature, Domestic Appliances Minimise Energy Cost Weather, Occupancy, Energy Prices No Yes
Chen et al. (2013) Linear Equations Indoor Temperature, Domestic Appliances Minimise Energy Cost Weather, Energy Prices No No
Yuce et al. (2016) ANN Domestic Appliances Minimise Grid Reliance Weather, Renewable Output No No
Široký et al. (2011) RC Indoor Temperature Minimise Energy Consumption Weather, Occupancy, Energy Prices Yes No
Oldewurtel et al. (2012) RC Indoor Temperature Minimise Energy Consumption Weather No No
Ma et al. (2011) Energy Plus Indoor Temperature Minimise Energy Cost Weather No Yes
Salsbury et al. (2013) System Identification Indoor Temperature Minimise Energy Cost Weather, Occupancy No Yes
Molina et al. (2013) State Space Indoor Temperature Minimise Energy Cost and Discomfort External Temperature, Solar Irradiance No No
Ferreira et al. (2012) ANN Indoor Temperature and Humidity Minimise Energy Consumption External Temperature, Irradiance, Humidity Yes No
Parisio et al. (2014) Explicit Model Indoor Temperature and CO2 Minimise Energy Cost Weather, Occupancy Yes No
Figueiredo and Sá da Costa RC Indoor Temperature and Luminosity Minimise Energy Cost External Temperature and Luminosity No No
(2012)
Moroşan et al. (2010) State Space Indoor Temperature Minimise Energy Consumption Occupancy No No

820
Lee et al. (2015) ANN Indoor Temperature Minimise Energy Cost DR signals, Energy Storage, Energy No Yes
Generation
Lee and Braun (2008) RC Indoor Temperature Minimise Peak Energy Consumption External Temperature, Irradiance No Yes
Doukas et al. (2007) – Air Quality, Luminosity, Temperature and Minimise Energy Consumption External Temperature, Humidity, Air Yes No
Humidity Quality, Luminosity
Yuce and Rezgui (2015) ANN Indoor Temperature Minimise Energy Consumption Weather, Occupancy Yes No
Moon and Kim (2010) ANN Indoor Temperature and Humidity Maximise Comfort External Temperature and Humidity No No
Yamada (1999) ANN PMV Minimum Deviation from PMV Set External Temperature, Irradiance, Yes No
Point Occupancy
Papantoniou et al. (2015) ANN Indoor Temperature Minimise Energy Consumption External Temperature No No
Erickson and Cerpa (2010) State Space Indoor Temperature, CO2 Minimise Energy Consumption External Temperature, Occupancy No No
Ahmad et al. (2016b) Energy Plus Indoor Temperature, Luminosity, CO2 Minimise Energy Consumption and Weather, Occupancy No No
Discomfort
Mossolly et al. (2009) RC Indoor Temperature, CO2 Minimise Energy Consumption Weather No No
Kim et al. (2016) ANN and Regression Indoor Temperature, Luminosity Minimise Energy Consumption External Temperature and Luminosity No No
Sun et al. (2013) State Space Indoor Temperature, Luminosity, CO2 Minimise Energy Consumption Weather, Occupancy No No
Sustainable Cities and Society 35 (2017) 816–829
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

(2015). When compared to a static day ahead scheduling strategy the demand. However, the profit achievable is highly dependent on the
MPC gives a 64% reduction in cost and manages uncertainties in pre- accuracy of the forecast as inability to meet the pre-agreed energy
diction better. A MPC framework for managing multiple residential supply would lead to penalties from the transmission system operator.
buildings in a microgrid is developed in Parisio, Wiezorek, Kyntäjä, Elo, Gruber, Huerta, Matatagui, and Prodanovic (2015) aims to improve
and Johansson (2015). Shared energy generation and storage are best on the typical MPC strategy to create a 2 stage MPC process. A short-
utilised to flatten peak loads and hence reduce the cost of energy for the term optimisation takes place every 5 min in between hourly optimi-
district. The demand response potential of a power to heat microgrid is sations to react and adjust to real time information and any errors in
also assessed in Rodriguez, Hinker, and Myrzik (2016). In Yan et al. initial predictions. MPC is used for a different purpose in Hu, Zhu, and
(2013) branch and cut methodology is used to optimise the operation of Guerrero (2014), the controllers’ objective is to maintain power quality
a small eco-district in China with PV, a CHP, waste to energy and sto- within the microgrid despite the variable renewable supply. Electric
rage capacity. vehicles are specifically exploited in Battistelli, Baringo, and Conejo
The growth in smart meters which can in greater detail inform the (2012). This study develops an aggregation and optimisation model for
consumer of their energy consumption as well as provide the utility the inclusion of vehicle to grid battery storage in a local microgrid.
company with finer granularity consumption data has enabled an in- Whilst one electric vehicle could only provide a small storage capacity,
creased use of time of use or dynamic energy pricing tariffs. However, if aggregated it could be substantial. However, it is questionable as to
whilst consumers may have more information at hand it is impractical whether users would accept their vehicle batteries being used like this
to expect them to constantly check the energy prices for the next day and some form of financial incentive would be necessary. A multi-ob-
and continue to adjust the scheduling of their household appliances to jective GA is used to minimise energy consumption and emissions from
minimise their energy costs. This must be automated and controlled to a 3 potential heat production units supplying a district heating system in
large degree by a bi-directional communicating smart controller that Jayan, Li, Rezgui, Hippolyte, and Howell (2016). The site is equipped
can communicate with the energy supplier and control household de- with a biomass boiler, a gas boiler and a gas CHP. The optimisation
vices such as washing machines, dishwashers and possibly electric ve- algorithm can produce a 24-h schedule of which production units to use
hicle charging points (Hatami & Pedram, 2010). This gives consumers based on demand. A memetic algorithm, an extension of a genetic al-
an opportunity for reductions in energy costs and aids the energy net- gorithm, is used in Hu et al. (2012) to control the set point of two
work due to a flatter load profile with less extreme peaks. Barbato and buildings that share an ice storage system. Each building is responsible
Capone (2014) gives a generic guide of the procedure for mathemati- for solving its own sub problem and the solutions are brought together
cally modelling the components of a smart grid or smart house for the by an aggregator to eventually converge on an overall cluster solution.
purposes of optimisation using linear, quadratic or dynamic program- The proposed decentralised optimisation provides a better solution that
ming. Energy consuming devices, energy storage, energy generation, ‘greedy’ solutions dominated by one building.
user comfort, and interactions with the energy market or supplier must
be modelled including all the associated constraints. An objective 3.2. Inter-district energy trading and aggregation
function could relate to minimisation of cost, discomfort, emissions, or
maximisation of local resources. Due to increasing grid decentralisation, some authors have devel-
The control logic behind a proposed smart scheduler is presented in oped business models for inter-district energy trading and bidding. This
Mohsenian-Rad and Leon-Garcia (2010). It provides a linear program- moves beyond time-of-use tariffs, which set fairly static pricing condi-
ming, appliance control optimisation method that considers predicted tions each day, to a more real-time energy market. The most developed
energy pricing fluctuations from the energy supplier. It schedules ap- standard on an integrated energy future is given in the Universal Smart
pliances to minimise the cost to the consumer and the waiting time for Energy Framework by the USEF Foundation, (USEF, 2015). It clearly
the appliances to complete the user requested task. They achieve a cost defines several stakeholders including the prosumer, the balance re-
reduction of around 25% and reduce the peak to average ratio by 38% sponsible party (BRP), the distribution system operator (DSO) and the
meaning the control strategy benefits both the consumer and the energy transmission system operator (TSO). It outlines the interactions the
supply network. This method is simulated on several networked re- stakeholders’ have with each other (Fig. 1) and the role an energy ag-
sidential buildings and cost savings by all households are reported al- gregator can play to provide flexibility in the system. The grid can re-
though the level of saving is determined by the load flexibility of each quest flexibility at specific times from a series of aggregators which in
consumer. A similar appliance scheduling optimisation strategy is de- turn manage a portfolio of prosumers from which it can leverage flex-
veloped in Deng, Yang, and Chen (2014) with specific consideration for ibility. Once agreement is reached and the decisions have been actuated
electricity price uncertainty in the real time market. The user can re- the grid must financially compensate the prosumer for their flexibility
serve energy a day in advance at wholesale cost or can purchase from service according to pre-agreed conditions.
the real-time energy market. The overall problem is decomposed into a Fanti, Member, Mangini, Roccotelli, and Ukovich (2015), develops a
set of solvable sub problems and the future electricity price uncertainty district energy management system based on day-ahead pricing
is managed using a stochastic gradient approach. schemes and real-time power monitoring. In this model buildings are
Barbato et al. (2011) also uses linear programming to schedule required to submit day ahead energy consumption predictions. Then
household appliances but also considers the effect PV panels and bat- the actual consumption of the buildings is monitored and compared to
tery storage could have in this optimisation. They consider both single the estimations to determine rewards or penalties. A DR aggregator for
house optimisation and a group of houses working cooperatively. Sig- a group of residential buildings is presented in Siano and Sarno (2016).
nificant cost savings can be achieved by consumers if they are more The controller bids for energy based on real-time pricing fluctuations
flexible with their device usage and have PV and battery storage. Fur- set by the DSO. This allows empowered consumers to shift their load,
thermore, if the district works cooperatively it can achieve significant avoiding peak prices, to achieve cost savings. This is also greatly ben-
reductions in the peak load and peak to average ratio which con- eficial to the DSO as overall peak demands on the system will be re-
siderably removes stress on the energy supply network. Local, small duced. In Anees and Chen (2016) a community controller acts as a
scale, residential renewable energy resources could be used to provide virtual DSO to implement real time price variation to a group of smart
ancillary or balancing services to the local energy network provided homes. Domestic appliances operation times are shifted to reduce the
they can accurately forecast the electricity they will provide a day peak energy demand. However, in their case study some residences
ahead. Clastres, Ha Pham, Wurtz, and Bacha (2010) presents a MILP receive increased costs even though the overall cost for the district is
optimisation strategy to maximise the profit from selling surplus elec- reduced. This raises crucial issues of potential unfairness that could
tricity through optimal management of storage capacity and shifting arise.

821
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

Fig. 1. USEF Interaction Model (USEF, 2015).

Agent-based, real-time price variation controllers can also be found A two-level demand side management game is developed in Chai,
in the literature. The PowerMatcher software is detailed in De Ridder, Chen, Yang, and Zhang (2014). The lower level evolutionary game
Hommelberg, and Peeters (2009). In this study the author proposes that composes of a population of residential, household consumers choosing
consuming and producing appliances are represented by intelligent how much energy to purchase at specific hours from different utility
agents. These agents submit the price that they are willing to pay or companies based on their prices. The upper game is a non-cooperative
receive for their energy. Once all bids are assembled, the market game between the utility companies where they determine their gen-
clearing price is calculated. If this price is higher than the consumer eration amount and future energy price. Both games are proven to
agent is willing to pay, then it does not consume energy and waits for converge quickly, the method is shown to be scalable and results in a
the next round. In a case study the rate of over or underproduction from lower average price for the consumers and a lower peak to average
a wind farm is reduced by 50% and peak load is reduced. Power- ratio. Gkatzikis, Koutsopoulos, and Salonidis (2013) investigates the
Matcher is also used in De Ridder et al. (2009). It finds that if you role an aggregator can play in a future smart grid setting. A three-level
increase the percentage of intelligent loads within a large district you scenario involving 10000 households, several aggregators and a single
get an almost linear decrease in peak power up to 20%. PowerMatcher utility is investigated. A day ahead the utility advertises a demand
is enhanced in Booij and Kamphuis (2013) to consider both electricity shifting target and a price they are willing to pay for this. The ag-
and heat in an integrated way which is important considering the in- gregator then bids a certain level of demand shifting on behalf of their
creasing electrification of heat through devices like heat pumps. portfolio of households whom they compensate for their flexibility. The
system is shown to be highly dependent on the reward the utility is
likely to offer and the level of flexibility shown by the residential
3.2.1. Game theoretical approaches consumer. However, the study did show potential for a 15% reduction
Traditional optimisation focussed at a district level could lead to in operating costs where all three parties gain compared to a baseline,
overall system optimal e.g. minimum total cost of district energy but flat price scenario. A combined MILP and game theoretical approach is
could lead to cost rises for specific individuals within the district. These used in Zhu et al. (2011) to optimise the scheduling of controllable
issues of unfairness could potentially be resolved by instead using a appliance to minimise the cost to a group of residential consumers
game theory approach to solving district energy management problems. working cooperatively. Wu, Mohsenian-Rad, Huang, and Wang (2011)
Game theory approaches can more fairly model individual “players” uses game theory to optimally control household appliances of several
rational desire to minimise their own energy costs. Saad, Han, Poor, and residential consumers in an islanded microgrid with wind and gas
Basar (2011) provides an excellent review of the game theory appli- generation. Using this method reduces the community energy bill by
cations in a smart grid environment. The autonomous, distributed, and 38% even with imperfect, Markov chain, wind generation forecasts. If
heterogeneous nature of the smart grid make game theory well suited to the forecasts are improved a further 21% saving could be achieved.
smart grid problems. The review argues that interactions and energy Rather than considering appliance scheduling, Atzeni, Ordóñez,
trading between microgrids and the wider network can be modelled as Scutari, Palomar, and Fonollosa (2013a), Atzeni, Ordonez, Scutari,
well as interactions between the consumer and utility company re- Palomar, and Fonollosa (2013b) use a game theoretical approach to
garding demand side management and load shifting.

822
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

optimise a smart grid in which a small percentage of users have dis- contained in different data silos. This paper uses Semantic Web tech-
patchable electricity generation and/or storage capacity. It assumes day nology to extract relevant information to the cloud and then proposes to
ahead knowledge of user demands from which a pricing tariff is set. The offer users cloud based building data services.
active users then use their flexibility to minimise their own electricity Wagner, Speiser, and Harth (2010) argues that sematic web tech-
bills which results in a flatter demand profile and hence lower prices. nologies are well suited to be applied to the smart grid given that it can
The users with greater flexibility (generation and storage) achieve very support heterogeneous data, it is flexible, it is extendable to new ad-
high savings around 80% but even the passive users see a reduction in ditions and the smart grid is distributed in its nature. Also within the
cost around 15% simply due to the reduction in peak prices. field of smart grids and urban planning Keirstead, Samsatli, and Shah
Mohsenian-Rad, Wong, Jatskevich, Schober, and Leon-Garcia (2010) (2010), created SynCity which contains a generic ontology describing
suggests that dynamic pricing set by the utility encourages each in- an urban energy systems. This ontology could easily be adapted and
dividual user reduce their energy cost by reducing their peak to average applied to different urban case studies. The paper then demonstrates
ratio. However, the author argues that this is not necessary providing a the usefulness of a shared information model by using it to optimise the
district works cooperatively to ensure their collective peak to average layout of a proposed eco-town in the UK. DR events can also be mod-
ratio is small. To achieve this the authors’ develop a distributed, game elled and represented in an ontology as demonstrated in Zhou,
theoretical approach to minimise a collective district energy bill by Natarajan, Simmhan, and Prasanna (2012). The authors also argue that
scheduling their appliances iteratively and broadcasting their fore- ontologies have the advantages that they can be re-used and they can be
casted energy consumption to their neighbours. This results in a 17% extended. Different ontologies can be defined separately for different
reduction in peak to average ratio and 18% reduction in cost. domains of the built environment. They can then be linked together by
an upper level ontology that map the concepts between domains re-
3.3. Communication infrastructure and data organisation sulting in a holistic urban knowledge model.
Ontologies have been demonstrated in two FP7 EU projects,
All proposed intelligent building control systems at both building RESILIENT and WISDOM. In RESILIENT a district energy ontology was
and wider district level must share large quantities of data. This comes produced to represent the electricity system, shared district heating
from multiple different sources and sensors in different locations pos- network, production units, stakeholders and the interactions between
sibly with different owners. These data streams are often produced in them (RESILENT Project, 2013). WISDOM applied smart ICT and se-
incompatible formats especially if systems have been retrofitted at later mantic modelling to aid the management of the water network (Duce
dates. Furthermore, this data must be kept secure and private to et al., 2014). An ontology for use in multi-agent systems which aligns
maintain consumer trust. Therefore, robust communication archi- and expands on existing standards and the USEF framework is detailed
tectures are essential in district energy management. in Hippolyte, Howell, Yuce, and Mourshed (2016). Smart grid roles
A cluster energy management system is developed in Kamiyoshi, such as the customer energy management system, the aggregator, the
Mine, and Nishi (2010). This system has a collective energy data col- distribution system operator are defined as well as the interactions
lection system through connected terminals deployed around a uni- between these autonomous stakeholders. A district energy ontology is
versity. The advantage of collecting such volumes of data is to allow used as the backbone for a web-based, urban, decisions support system
higher level data analytics to take place. For example, in this study the in Howell, Hippolyte, Jayan, Reynolds, and Rezgui (2016). The devel-
collated data is used to generate energy demand predictions. A split oped platform aims to engagingly and visually display buildings uti-
HVAC system across multiple university buildings was also considered lising existing building information models, apply background data
in Escrivá-Escrivá, Segura-Heras, and Alcázar-Ortega (2010). The sys- analytics and machine learning, and employ optimisation through web
tems are linked using the common Ethernet network and TCP-IP pro- services.
tocols. The information is gathered in a central server where the control
centre and SQL database is kept. This system then allows data to be 3.3.2. Multi-agent systems
accessed through a simple web interface to provide support for tech- Multi-agent systems, MAS, have achieved significant attention and
nical staff and facility managers. A similar communication network is growth in recent years. MAS involves several distributed, intelligent
utilised in Kolokotsa et al. (2016) taking advantage of the common agents that can implement bottom-up control. These agents are pro-
university internet structure and replacing all sensors so they can use grammed with internal logic or intelligence to manage the operation of
Internet Protocols to communicate with a central university controller. selected components such as a generation unit, a HVAC component, or
In this study, an ANN is trained to predict building energy consumption. a storage device. As well as aiming to achieve their set objective they
Using this a GA optimises the building’s lighting and HVAC strategies. are also perceptive of their environment and able to communicate and
Barbato et al. (2016) is also focussed on university campuses. This study cooperate with neighbouring agents. Advantages to MAS include a
presents a data collection and communication architecture to capture completely scalable computing architecture, resilience to failures in
building energy characteristics. From this base, prediction models and communication, and potentially increased security as no agent will
optimal schedulers can be created and implemented to save energy. have access to every piece of information.
Iwamura creates an information platform that gathers data from smart Lagorse, Paire, and Miraoui (2010), applies MAS to a hybrid re-
communities from the user and utility side. From this they can imple- newable system. Each device has internal rules and a ‘token’ is passed
ment data analytics to monitor trends and produce predictions. This between devices to indicate which agent is in control of the system
could allow implementation of DR measures in response to natural based on their individual internal states. Ramchurn, Vytelingum,
disasters or faults. Rogers, and Jennings (2011) studies agent based control of domestic
appliances on a wide scale. An agent is based in the smart meter of each
3.3.1. Sematic interoperability home representing the aggregation of several controllable and un-
Interoperability and communication between different hetero- controllable household appliances. A connected system of 5000 homes
geneous data sources could be addressed utilising Semantic Web tech- is theorised and if the developed agent based system is implemented
nology. An ontology can allow the formation of a machine readable, energy peaks could be decreased by up to 17%. Similarly, Joumaa,
multi domain knowledge base. They could potentially provide the Ploix, Abras, and De Oliveira (2011), creates an agent-based archi-
means to facilitate data retrieval, interoperability and decision support tecture for controlling household appliances. Services are categorised as
in the built environment (Hu et al., 2014). Curry, Battistelli et al. either energy supply, storage or end use and end use services were
(2012), argues that ‘linked data’ techniques are needed within buildings further broken down into temporary, controllable or permanent con-
as they are producing more data than ever and these are often sumers. The GRENAD, MAS framework is detailed in Ductor, Gil-

823
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

Quijano, Stefanovitch, and Mele (2015). This aims to simulate and schemes possibly facilitated by an energy aggregator. However, the
control smart power grids using MAS through a generic, modular and applicability of these types of aggregators is likely to be highly de-
flexible platform. Jun, Jie, Jun-feng, La-mei, and Min (2008) defines a pendent on the type of buildings that make up the district in question. A
MAS architecture for an eco-house that could operate on or off grid due district of public sector office buildings with single ownership will have
to a number of agent controlled renewable resources. very different privacy and engagement challenges than a residential
district with hundreds to thousands of different owners and stake-
3.3.3. Peer to peer networks holders. In the authors’ opinion government policy and regulation in
Lasseter (2011), argues that microgrid components should conform this area is lacking. Strict controls are required to ensure that any en-
to peer-to-peer concepts. As opposed to centralised control schemes, a ergy trading system works fairly for all consumers. Furthermore, sig-
peer-to-peer based system can be highly scalable, extensible, resilient to nificant effort will be required to persuade the public to participate in
failure of single components and more secure. (Dorri, Kanhere, Jurdak, such programmes and clear information on the trade-off between
and Gauravaram (2017) demonstrates the use of Blockchain tech- sharing private information and possible financial gains must be well
nology, a key underpinning component of the Bitcoin cybercurrency, to communicated. Some consumers may be reluctant to cede any control
enhance the security and privacy of a IoT based smart home system. In over their personal appliances and accept any level of discomfort that is
Rusitschka, Gerdes, and Eger (2009), the authors propose that in- required for financial gain. A strong period of demonstration and per-
stallation of intelligent smart meters would be very expensive and this suasion will be required before these strategies can become main-
fixed infrastructure is not necessary or adaptable for the future. Instead stream.
they propose a peer to peer network of homes system containing digital
electricity meters. These homes would form self-organising, scalable, 4. Future research directions
networks large enough to be able to provide DR, load balancing, or
power quality services to the grid operator and hence benefit from the This paper has reviewed a wide breadth of peer-reviewed research
financial rewards currently only available to large scale consumers. The papers on the topic of building energy management at both a building
equipment needed to construct such a network is far cheaper than and district level. It has emphasised the need for smarter energy con-
current smart meter models and the operator has no installation, trol, improving on current static, rule-based management systems.
maintenance or operating costs due to the self-organising nature of Given the evolution of energy infrastructure there is an urgent need for
peer-to-peer networks. Due to the increasing number of prosumers in predictive control and implementation of demand response measures to
the future smart grid, the availability and opportunity for transactions match fluctuating demand with fluctuating supply introduced by re-
between peers and utilities is increasing (Ipakchi, 2011). Mature tech- newable energy sources. However, this review has identified a number
nologies such as Blockchain could provide a solution to ensuring se- of gaps in the current body of research that give rise to future research
curity, fairness and consumer confidence in a proposed energy and directions.
money trading system.
4.1. Holistic energy management
3.4. Discussion
The district level microgrid control, examples of which are shown in
Table 2 summarises the literature reviewed in Section 3. As de- Section 3.1, are generally much more aware of demand response and
monstrated in Table 2, optimisation from the point of view of the include it as a stated objective of their control schemes. However, they
supply side and the demand side is considered. Many of these papers generally consider the building demand as perfectly predicted and
consider assumed knowledge of a fixed heat or electric demand profile uncontrollable. They often model building profiles as a constraint
over the optimisation time horizon. Many also assume perfect knowl- which their optimisations must meet. There is a real need for there to be
edge of future renewable energy generation. In reality, these profiles a bi-directional link between a building level controller that is aware of
can be difficult to accurately predict and are highly dependent on dis- demand response and a district level controller that actively views
turbances such as weather conditions and building occupancy. New buildings as potential demand response sources. This could be a two-
prediction methodologies are being applied to overcome this challenge tiered optimisation model. At a building level, it could optimise the
such as the use of ANN or random forest algorithms (Ahmad, control of HVAC systems or domestic appliances similar to many of the
Mourshed, & Rezgui, 2017). To be implemented in real case studies, studies in Section 2. It would send the predicted schedule to a district
prediction of these variables should be further improved and the control level controller but crucially include a degree of flexibility in the energy
algorithms themselves should take into account the inherent un- profile. The district level would aggregate a number of demand profiles
certainty within the prediction. If this is not the case, frequent breaches from several buildings and optimally manage any shared generation
of comfort constraints will arise and contingency control algorithms resources and storage capacity. It could then engage in negotiations to
will need to take over. seek further flexibility if necessary. This sort of bi-directional, iterative
The management of highly heterogeneous data and the sheer vo- approach would keep the level of detail at a building level but also have
lume of data required for district level control must be addressed. the wider awareness of the districts energy needs, providing a more
Whilst this is manageable within an environment with shared owner- holistic approach to energy management.
ship and ICT infrastructure it remains a problem for multi-use, multi-
owner districts. These more diverse districts are likely to have a wider 4.2. Semantic representation of the district energy environment
number of systems, which use different communication protocols, are
different ages, and have dissimilar privacy issues. From the reviewed To manage and control a modern, complex district, information and
literature, the use of semantic web technologies such as ontologies characteristics from various heterogeneous, multi domain, data sources
appears to provide an adequate solution to ensure the interoperability must be linked. This can be achieved through semantic web technology
of this diverse data. These can then be used as a base from which in- which allow machine interpretable descriptions of the district to be
telligent data analytics and control platforms can be developed. Indeed, defined. This can allow domains which have been considered in isola-
they have already been used as the platform to build MAS which is a tion to each other to be optimised in a holistic way. Leveraging se-
rapidly increasing field of knowledge. MAS seem particularly applicable mantic modelling could provide the tool to enable the exchange of data
to the smart grid architecture due to their inherently decentralised between different data models and software’s which may not use the
structures and scalability. same communication protocols to allow truly holistic energy manage-
Finally, interesting work is taking place on virtual energy trading ment approaches. As shown in Section 3.2.1 they can also provide the

824
 J. Reynolds et al.

Table 2
Summary of District Optimisation Literature.

Reference Optimisation Method Aims to Control Objective Predictions Included Real Case Load Smoothing
Study? Considered?

Guan et al. (2010) MILP 1 Scheduling Generation Minimise Energy Cost Weather, Renewable Supply, Demand No No
Staino et al. (2016) Economic MPC Shared Heat Pump Operation Minimise Energy Cost External Temperature, Electricity Prices No No
Marzband et al. (2013) MINLP 2 Scheduling Generation Minimise Energy Cost Weather, Renewable Supply, Demand No No
Zhang et al. (2015) MILP Scheduling Generation and Energy Minimise Energy Cost Renewable Supply, Demand, Electricity No Yes
Storage Price
Parisio et al. (2015) MILP Demand Scheduling and Energy Minimise Energy Cost Renewable Supply, Demand, Electricity No Yes
Storage Prices
Rodriguez et al. (2016) Linear Programming Heat Supply and Thermal Storage Minimise Temperature Set Point Error and Indoor Temperature, Thermal Comfort No Yes
Operation Peak Power Flow
Yan et al. (2013) Branch and Cut Scheduling Generation and Building Minimise Energy Cost Indoor Temperature, Energy Prices No No
Demand
Hatami and Pedram (2010) Dynamic Programming Demand Scheduling Minimise Energy Cost Electricity Prices, Demand No Yes
Mohsenian-Rad and Leon-Garcia Linear Programming Appliance Scheduling Minimise Energy Cost and User Electricity Prices, Demand No Yes
(2010) Dissatisfaction
Deng et al. (2014) Stochastic Gradient Appliance Scheduling Minimise Energy Cost Electricity Prices, Demand No Yes
Barbato et al., (2011) Linear Programming Appliance Scheduling Minimise Energy Cost Electricity Prices, Demand, Solar Energy No Yes
Supply
Clastres et al. (2010) MILP Scheduling Generation and Energy Minimise Energy Cost Electricity Prices, Demand, Solar Energy No Yes
Storage Supply
Hu et al. (2012) Memetic Algorithm Building Temperature Set Point Minimise Energy Cost Weather, Demand, Renewable Supply, No No

825
Electricity Prices
3
Gruber et al. (2015) MILP Scheduling Generation, Demand and Minimise Energy Cost Renewable Supply, Demand, Electricity Yes Yes
Energy Storage Prices
Battistelli et al. (2012) Robust Optimisation Scheduling Generation and Energy Minimise Energy Cost Renewable Supply, Demand, Electricity No No
Storage Prices
Jayan et al. (2016) Genetic Algorithm Scheduling Generation Minimise Energy cost and Emissions Energy Prices, Demand No No
Fanti et al. (2015) Linear Programming Demand Scheduling Minimise Energy Cost Electricity Prices, Weather, Demand No Yes
Siano and Sarno (2016) D-LMP 4 Demand Scheduling Minimise Energy Cost Electricity Prices, Demand, Building No Yes
Temperature
Anees and Chen (2016) Linear Programming Demand Scheduling and Real Time Minimise Energy Cost Electricity Prices, Demand No Yes
Pricing
De Ridder et al. (2009) Agent Based Appliance Scheduling Minimise Energy Cost Renewable Supply, Demand, Electricity No Yes
Prices
Booij and Kamphuis (2013) Agent Based Scheduling Generation Minimise Energy Cost Renewable Supply, Demand, Electricity No Yes
Prices
Chai et al. (2014) Game Theory Demand Scheduling Minimise Energy Cost, Maximise Profit Electricity Prices, Demand No Yes
Gkatzikis et al. (2013) Game Theory Demand Scheduling Minimise Energy Cost, Maximise Profit Electricity Prices, Demand No Yes
Zhu et al. (2011) Game Theory Demand Scheduling Minimise Energy Cost, Electricity Prices, Demand No Yes
Wu et al. (2011) Game Theory Appliance Scheduling Minimise Energy Cost, Renewable Supply, Demand, Electricity No Yes
Prices
Atzeni et al. (2013a) Game Theory Demand and Generation Scheduling Minimise Energy Cost, Minimise Peak to Renewable Supply, Demand, Electricity No Yes
Average Ratio Prices
Mohsenian-Rad et al. (2010) Game Theory Appliance Scheduling Minimise Energy Cost, Electricity Prices, Demand No Yes
Sustainable Cities and Society 35 (2017) 816–829
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

platform from which district level optimisation can be implemented. visualisation and decision support is provided in the form of a web-
Furthermore, utilising semantics and ontologies allows a scalable and based application, which uses the Unity game engine. Note the Unity
flexible approach to district modelling as additions can be made to engine only displays data and visuals to the user, the data processing is
adapt to unforeseen future changes to the district. performed in the HPC, cloud-based infrastructure. The CUSP platform
Future research in semantic-based modelling could provide the link couples data analytics and prediction services with BIM models. A key,
between the currently available Building Information Modelling, BIM, underpinning, feature of the CUSP architecture is the ontology server
models and real time, operational data collected by sensors embedded which effectively can link the previously isolated information. Ontol-
within the building. A centralised ontology that has knowledge of the ogies bring context, meaning and provenance to the data. For instance,
buildings physical components and characteristics as well as access to external weather data or internal sensor readings through the BMS can
BEMS sensory information would allow truly powerful and useful data be gathered and stored in the time series server. The CUSP platform
analytics for a facility manager. This can provide the platform to allow aims to provide intuitive controls to access the district or building level
prediction of future energy consumption, behaviour patterns and oc- data and decision support from a selected date range, including future
cupancy. Smart control algorithms could also be built on top of the dates. A simulation server utilising Energy Plus models or machine
central ontology allowing resulting schedules and instructions to be learning, surrogate models can produce day ahead forecasting which
sent for the BEMS to action. The base provided by the semantic mod- may influence facility managers’ decision making. Given the collection
elling of a district could lead to a 3D visualisation of the district for of large quantities of useful information, optimisation procedures must
facility managers, local authorities, or urban planners. The link with the be deployed utilising powerful, cloud-based, high performance com-
sensory information of the district could allow relevant data, depending puting which could provide energy and cost savings to the consumer.
on the user, to be displayed in a more dynamic, clear and useful manner CUSP also aims to simply display various key performance indicators,
compared to current BEMS interfaces. KPI’s, such as energy consumption per unit area, CO2 emissions, or
economic running costs. This, alongside automatic anomaly detection
alerts can quickly inform a facility manager if the district is performing
4.3. Computation urban sustainability platform
as expected.

The Computational Urban Sustainability Platform, CUSP, is cur-

rently under development by the authors’ research group. Its interface 4.4. Business models and energy policy
and architecture is displayed in Fig. 2. This project aims to implement
the vision set out by this paper. CUSP presents real time, actionable Given the rise in implementation of smart metering devices, time of
information in a web based platform for facility managers or urban use or real time energy pricing tariffs are likely to become more
planners (Howell et al., 2016). An engaging, 3D user interface of district available and popular with consumers. This is possible as smart meters

Fig. 2. CUSP Interface and Architecture.

826
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

will give utility companies greater detail on the quantity and the time of building level controls needs a mechanism to adjust and respond to
energy consumption. These tariffs will allow engaged and empowered demand response instructions from a district level controller. The vast
users to gain substantial cost savings over fixed rate energy tariffs by majority of the literature reviewed is simulation based. Whilst this is an
intelligently shifting their consumption to advantageous times. excellent method to prove concepts there is now a desire to demonstrate
However, we cannot expect the average consumer to constantly that these strategies can be applied in real time and do provide their
monitor or understand energy price fluctuations and manually reset estimated savings. The review has also found a demand for the use of
many of their devices to consume or stop consuming. This leads to semantic web technology to effectively manage the heterogeneous data
opportunities for so called Energy Service Companies, ESCo’s. streams that modern, complex, building control requires. This has been
Consumers could effectively outsource their energy management and demonstrated by the Computational Urban Sustainability Platform,
relevant data to 3rd party companies which would aim to provide en- CUSP, which aims to provide 3D visualisation, simulation and KPI in-
ergy cost reductions for the consumer and in return take a proportion of dication for facility managers all underpinned by an ontology. Finally,
that saving. ESCo’s would have to gain access to large amounts of user there is a need for adjustment in energy policy through the increased
data and could provide virtual, cloud-based energy management. The use of real time pricing or time of use tariffs. These will allow engaged
recent introduction of commercially available, on demand, high per- users to save money on their energy bills and will help the grid to lower
formance computing, HPC, from cloud services companies such as IBM, peak demand. In the authors opinion, this will lead to increased op-
Google, Amazon and Penguin could revolutionise what is achievable in portunity for ESCo’s to manage households (including prosumers) en-
building energy management. It could allow greater levels of data ergy in a number of business models outlined in Section 4.3.
analytics, improved prediction models or even the use of detailed,
white box energy models in real-time optimisation (Petri et al., 2014; Acknowledgments
Yang et al., 2014). Several different control approaches could be im-
plemented by ESCo’s: The authors would like to acknowledge the financial support of
EPSRC (Engineering and Physical Sciences Research Council) and BRE
• Internal Energy Markets – Similar to some of the studies in Section (Building Research Establishment) as well as the European Commission
3.2, an internal energy market between the ESCo’s clients could be as part of the Horizon2020 THERMOSS (Project Id: 723562) and
formed. Users with complimentary load profiles or excess renewable PENTAGON (Project Id: 731125) projects.
generation could combine to form virtual energy sharing partner-
ships facilitated by the central grid and the ESCo. These consumers References
could be in different geographic locations and provide mutual sav-
ings for all parties and achieve greater prices for excess energy ra- Aduda, K. O., Labeodan, T., Zeiler, W., Boxem, G., & Zhao, Y. (2016). Demand side
ther than selling back to the grid. flexibility: Potentials and building performance implications. Sustainable Cities and

•
Society, 22, 146–163. http://dx.doi.org/10.1016/j.scs.2016.02.011.
Centralised Control – This control architecture would be more ap- Afram, A., & Janabi-Sharifi, F. (2014). Theory and applications of HVAC control
plicable for, natural, self-contained, districts with a single owner systems—A review of model predictive control (MPC). Building and Environment, 72,
and a single utility bill such as University campuses, industrial es- 343–355. http://dx.doi.org/10.1016/j.buildenv.2013.11.016.
Ahmad, M. W., Mourshed, M., & Rezgui, Y. (2017). Trees vs neurons: Comparison be-
tates or public sector buildings. In this situation, direct, advanced, tween random forest and ANN for high-resolution prediction of building energy
MPC could be applied utilising the existing SCADA based system for consumption. Energy and Buildings, 147, 77–89. http://dx.doi.org/10.1016/j.enbuild.
data collection and actuation. The intelligence and decision making 2017.04.038.
Ahmad, M. W., Hippolyte, J., Reynolds, J., Mourshed, M., & Rezgui, Y. (2016). Optimal
would be held in a control layer above the SCADA system.
•
scheduling strategy for enhancing IAQ, thermal comfort and visual using a genetic
Intelligence Update – A more passive approach that ensures the user algorithm. Iaq 2016 Defin. Indoor Air Qual. Policy, Stand. Pract., Ashrae.
feels in control of their systems. The ESCo could carry out data Ahmad, M. W., Mourshed, M., Yuce, B., & Rezgui, Y. (2016). Computational intelligence
techniques for HVAC systems: A review. Building Simulation, 9, 359–398. http://dx.
analytics and feedback simple suggestions to the user based on the
doi.org/10.1007/s12273-016-0285-4.
data available. These could be slight adjustments to the current Ameri, M., & Besharati, Z. (2015). Optimal design and operation of district heating and
rules, for example to turn off the HVAC system an hour earlier. The cooling networks with CCHP systems in a residential complex. Energy and Buildings,
user would then decide whether to implement this. 110, 135–148. http://dx.doi.org/10.1016/j.enbuild.2015.10.050.

•
Anees, A., & Chen, Y. P. (2016). True real time pricing and combined power scheduling of
Demand Response Coordinator – This architecture is much closer to electric appliances in residential energy management system. Applied Energy, 165,
USEF. Initially each BMS would locally optimise their own day 592–600. http://dx.doi.org/10.1016/j.apenergy.2015.12.103.
ahead demand. This would be fed to a district level controller to Atzeni, I., Ordóñez, L. G., Scutari, G., Palomar, D. P., & Fonollosa, J. R. (2013a).
Noncooperative and cooperative optimization of distributed energy generation and
build a district demand profile. Using this knowledge and predic- storage in the demand-side of the smart grid. IEEE Transactions on Signal Processing,
tions of generation capacity, it could make decisions on how to 61, 2454–2472. http://dx.doi.org/10.1109/tsp.2013.2248002.
flatten the overall demand profile. An iterative, negotiation based Atzeni, I., Ordonez, L. G., Scutari, G., Palomar, D. P., & Fonollosa, J. R. (2013b). Demand-
side management via distributed energy generation and storage optimization. IEEE
arrangement would take place to lead to a more optimal district Transactions on Smart Grid, 4, 866–876. http://dx.doi.org/10.1109/tsg.2012.
demand profile. 2206060.
Barbato, A., & Capone, A. (2014). Optimization models and methods for demand-side
management of residential users: A survey. Energies, 7, 5787–5824. http://dx.doi.
5. Conclusion
org/10.3390/en7095787.
Barbato, A., Capone, A., Carello, G., Delfanti, M., Merlo, M., & Zaminga, A. (2011). House
This paper has reviewed a large number of publications on the topic energy demand optimization in single and multi-user scenarios. 2011 IEEE Int. Conf.
Smart Grid Commun. SmartGridComm 2011, 345–350. http://dx.doi.org/10.1109/
of building energy management. Crucially, it has also assessed building
SmartGridComm.2011.6102345.
energy management from a district level. This is necessary due to the Barbato, A., Bolchini, C., Geronazzo, A., Quintarelli, E., Palamarciuc, A., Pitì, A., et al.
increasingly distributed and localised nature of the energy infra- (2016). Energy optimization and management of demand response interactions in a
structure as well as the changing mix of supply sources. The review has smart campus. Energies, 9, 398. http://dx.doi.org/10.3390/en9060398.
Battistelli, C., Baringo, L., & Conejo, A. J. (2012). Optimal energy management of small
found that most district level optimisation procedures consider the electric energy systems including V2G facilities and renewable energy sources.
optimal supply of a district. They generally consider the demand profile Electric Power Systems Research, 92, 50–59. http://dx.doi.org/10.1016/j.epsr.2012.
from the buildings as known, perfectly predicted and unalterable. 06.002.
Booij, P., & Kamphuis, V. (2013). Multi-agent control for integrated heat and electricity
Therefore, there is a requirement for a holistic district energy man- management in residential districts. 12th Int. Conf. Auton. Agents Multiagent
agement controller that finds the right balance between the type of Systhttp://www.ates2013. org/papers/paper5.pdf.
papers found in Sections 2 and 3. The district level management needs CIBSE (2009). Chartered institute of building services engineers. CIBSE Guide H.
Capone, A., Barros, M., Hrasnica, H., & Tompros, S. (2009). A new architecture for
to consider buildings as an active demand response participant and the

827
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

reduction of energy consumption of home appliances. TOWARDS eENVIRONMENT, Urban Decision Support through Intelligent and Interoperable Services. Second Int.
European conference of the Czech Presidency of the Council of the EU 2009. Smart Cities Conf. IEEE.
Chae, Y. T., Horesh, R., Hwang, Y., & Lee, Y. M. (2016). Artificial neural network model Hu, M., Weir, J. D., & Wu, T. (2012). Decentralized operation strategies for an integrated
for forecasting sub-hourly electricity usage in commercial buildings. Energy and building energy system using a memetic algorithm. European Journal of Operational
Buildings, 111, 184–194. http://dx.doi.org/10.1016/j.enbuild.2015.11.045. Research, 217, 185–197. http://dx.doi.org/10.1016/j.ejor.2011.09.008.
Chai, B., Chen, J., Yang, Z., & Zhang, Y. (2014). Demand response management with Hu, J., Zhu, J., & Guerrero, J. M. (2014). Model predictive control of smart microgrids.
multiple utility companies: A two-level game approach. IEEE Transactions on Smart 17th Int. Conf. Electr. Mach. Syst. 2815–2820.
Grid, 5, 722–731. http://dx.doi.org/10.1109/tsg.2013.2295024. Ipakchi, A. (2011). Demand side and distributed resource management—A transactive
Chen, C., Wang, J., Heo, Y., & Kishore, S. (2013). MPC-based appliance scheduling for solution. IEEE Power & Energy Society General Meeting, 1–8. http://dx.doi.org/10.
residential building energy management controller. IEEE Transactions on Smart Grid, 1109/PES.2011.6039272.
4, 1401–1410. Jayan, B., Li, H., Rezgui, Y., Hippolyte, J.-L., & Howell, S. (2016). An analytical opti-
Clastres, C., Ha Pham, T. T., Wurtz, F., & Bacha, S. (2010). Ancillary services and optimal mization model for holistic multiobjective district energy management—A case study
household energy management with photovoltaic production. Energy, 35, 55–64. approach. International Journal of Modeling and Optimization, 6, 156–165. http://dx.
http://dx.doi.org/10.1016/j.energy.2009.08.025. doi.org/10.7763/IJMO.2016.V6.521.
DECC (2014). Smart metering implementation programme: Third annual report on the roll-out Joumaa, H., Ploix, S., Abras, S., & De Oliveira, G. (2011). A MAS integrated into home
of smart meters. automation system, for the resolution of power management problem in smart
De Paola, A., Ortolani, M., Re, G. L. O., Anastasi, G., & Das, S. K. (2014). Intelligent homes. Energy Procedia, 6, 786–794. http://dx.doi.org/10.1016/j.egypro.2011.05.
management systems for energy efficiency in buildings: A survey. ACM Computing 089.
Surveys, 47, 1–38. Jun, Z. J. Z., Jie, W. J. W., Jun-feng, L. J. L., La-mei, G. L. G., & Min, L. M. L. (2008). An
De Ridder, F., Hommelberg, M., & Peeters, E. (2009). Four potential business cases for agent-based approach to renewable energy management in eco-building. IEEE
demand side integration. 6th Int. Conf. Eur. Energy Mark., IEEE (pp. 1–6). International Conference on Sustainable Energy Technologies, 46–50. http://dx.doi.org/
Deng, R., Yang, Z., & Chen, J. (2014). Load scheduling with price uncertainty and tem- 10.1109/ICSET. 2008.4746970.
porally-coupled constraints in smart grids. IEEE Transactions on Power Systems, 29, Kamiyoshi, Y., Mine, G., & Nishi, H. (2010). Construction of energy measurement system
2823–2834. in a university for cluster energy management. IECON – 36th Annu Conf. IEEE Ind.
Di Somma, M., Yan, B., Bianco, N., Graditi, G., Luh, P. B., Mongibello, L., et al. (2015). Electron. Soc. (pp. 2423–2429).
Operation optimization of a distributed energy system considering energy costs and Keirstead, J., Samsatli, N., & Shah, N. (2010). Syncity: An integrated tool kit for urban
exergy efficiency. Energy Conversion and Management, 103, 739–751. http://dx.doi. energy systems modelling. Energy Efficient Cities: Assessment Tools and Benchmarking
org/10.1016/j.enconman.2015.07.009. Practices, 21. http://dx.doi.org/10.1596/978-0-8213-8104-5.
Dorri, A., Kanhere, S. S., Jurdak, R., & Gauravaram, P. (2017). Blockchain for IoT security Kim, W., Jeon, Y., & Kim, Y. (2016). Simulation-based optimization of an integrated
and privacy: The case study of a smart home. 2nd IEEE Work. Secur. Priv. Trust Internet daylighting and HVAC system using the design of experiments method. Applied
Things. Energy, 162, 666–674. http://dx.doi.org/10.1016/j.apenergy.2015.10.153.
Doukas, H., Patlitzianas, K. D., Iatropoulos, K., & Psarras, J. (2007). Intelligent building Koirala, B. P., Koliou, E., Friege, J., Hakvoort, R. A., & Herder, P. M. (2016). Energetic
energy management system using rule sets. Building and Environment, 42, 3562–3569. communities for community energy: A review of key issues and trends shaping in-
http://dx.doi.org/10.1016/j.buildenv.2006.10.024. tegrated community energy systems. Renewable and Sustainable Energy Reviews, 56,
Duce, E., Musetti, A., Belziti, D., Beach, T., Gluhak, A. D., Nalesso, M., et al. (2014). WThe 722–744. http://dx.doi.org/10.1016/j.rser.2015.11.080.
ISDOM project for smart water monitoring using advanced ICT equipment, data and Kolokotsa, D., Gobakis, K., Papantoniou, S., Georgatou, C., Kampelis, N., Kalaitzakis, K.,
tools. Water Ideas 22–24 October (pp. 1–10). et al. (2016). Development of a web based energy management system for University
Ductor, S., Gil-Quijano, J.-J., Stefanovitch, N., & Mele, P. R. (2015). GRENAD, a modular Campuses: The CAMP-IT platform. Energy and Buildings, 123, 119–135. http://dx.doi.
and generic smart-grid framework. In: Proc. Fed. Conf. Comput. Sci. Inf. Syst, org/10.1016/j.enbuild.2016.04.038.
1781–1792. http://dx.doi.org/10.15439/2015F310. Kolokotsa, D. (2016). The role of smart grids in the building sector. Energy and Buildings,
Erickson, V. L., & Cerpa, A. E. (2010). Occupancy based demand response HVAC control 116, 703–708. http://dx.doi.org/10.1016/j.enbuild.2015.12.033.
strategy. In: Proc. 2nd ACM Work. Embed. Sens. Syst. Energy-Efficiency Komareji, M., Stoustrup, J., Rasmussen, H., Bidstrup, N., Svendsen, P., & Nielsen, F.
Build.—BuildSys’ 10, 7. http://dx.doi.org/10.1145/1878431.1878434. (2008). Optimal model-based control in hvac systems. Am. Control Conf. (pp. 1443–
Escrivá-Escrivá, G., Segura-Heras, I., & Alcázar-Ortega, M. (2010). Application of an 1448). . http://dx.doi.org/10.1109/CCA.2009.5280724.
energy management and control system to assess the potential of different control Kumar, P., Martani, C., Morawska, L., Norford, L., Choudhary, R., Bell, M., et al. (2016).
strategies in HVAC systems. Energy and Buildings, 42, 2258–2267. http://dx.doi.org/ Indoor air quality and energy management through real-time sensing in commercial
10.1016/j.enbuild.2010.07.023. buildings. Energy and Buildings, 111, 145–153. http://dx.doi.org/10.1016/j.enbuild.
European Parliament (2010). Directive 2010/31/EU of the European Parliament and of 2015.11.037.
the council on the energy performance of buildings. Official Journal of the European Kwadzogah, R., Zhou, M., Li, S., & Member, S. (2013). Model predictive control for HVAC
Union, 53, 40. systems—A review. IEEE Int. Conf. Autom. Sci. Eng. 442–447. http://dx.doi.org/10.
Fanti, M. P., Member, S., Mangini, A. M., Roccotelli, M., & Ukovich, W. (2015). A district 1109/CoASE.2013.6654072.
energy management based on thermal comfort satisfaction and real-time power Lagorse, J., Paire, D., & Miraoui, A. (2010). A multi-agent system for energy management
balancing. IEEE Transactions on Automation Science and Engineering, 12, 1271–1284. of distributed power sources. Renewable Energy, 35, 174–182. http://dx.doi.org/10.
Ferreira, P. M., Silva, S. M., & Ruano, A. E. (2012). Energy savings in HVAC systems using 1016/j.renene.2009.02.029.
discrete model-based predictive control. IEEE World Congr. Comput. Intell. (pp. 1–8). . Lasseter, R. H. (2011). Smart distribution: Coupled microgrids. In: Proc. IEEE. 99,
http://dx.doi.org/10.1109/IJCNN.2012.6252538. 1074–1082. http://dx.doi.org/10.1109/JPROC.2011.2114630.
Figueiredo, J., & Sá da Costa, J. (2012). A SCADA system for energy management in Lee, K. H., & Braun, J. E. (2008). Model-based demand-limiting control of building
intelligent buildings. Energy and Buildings, 49, 85–98. http://dx.doi.org/10.1016/j. thermal mass. Building and Environment, 43, 1633–1646. http://dx.doi.org/10.1016/
enbuild.2012.01.041. j.buildenv.2007.10.009.
Fong, K. F., Hanby, V. I., & Chow, T. T. (2006). HVAC system optimization for energy Lee, D., & Cheng, C.-C. (2016). Energy savings by energy management systems: A review.
management by evolutionary programming. Energy and Buildings, 38, 220–231. Renewable and Sustainable Energy Reviews, 56, 760–777. http://dx.doi.org/10.1016/j.
http://dx.doi.org/10.1016/j.enbuild.2005.05.008. rser.2015.11.067.
Fong, K. F., Hanby, V. I., & Chow, T. T. (2009). System optimization for HVAC energy Lee, Y. M., Horesh, R., & Liberti, L. (2015). Optimal HVAC control as demand response
management using the robust evolutionary algorithm. Applied Thermal Engineering, with on-site energy storage and generation system. Energy Procedia, 78, 2106–2111.
29, 2327–2334. http://dx.doi.org/10.1016/j.applthermaleng.2008.11.019. http://dx.doi.org/10.1016/j.egypro.2015.11.253.
Gils, H. C. (2014). Assessment of the theoretical demand response potential in Europe. Li, X., & Wen, J. (2014). Review of building energy modeling for control and operation.
Energy, 67, 1–18. http://dx.doi.org/10.1016/j.energy.2014.02.019. Renewable and Sustainable Energy Reviews, 37, 517–537. http://dx.doi.org/10.1016/j.
Gkatzikis, L., Koutsopoulos, I., & Salonidis, T. (2013). The role of aggregators in smart rser.2014.05.056.
grid demand response markets. IEEE Journal on Selected Areas in Communications, 31, Liang, J., & Du, R. (2005). Thermal comfort control based on neural network for hvac
1247–1257. http://dx.doi.org/10.1109/JSAC.2013130708. application. IEEE Conf. Control Appl. (pp. 819–824).
Gouda, M. M., Danaher, S., & Underwood, C. P. (2001). Thermal comfort based fuzzy Lins, C., Williamson, L., Leitner, S., & Teske, S. (2014). The first decade: 2004–2014, 10
logic controller. Building Services Engineering Research and Technology, 22, 237–253. years of renewable energy progress. 9783981593440.
http://dx.doi.org/10.1177/014362440102200403. Lu, L., Cai, W., Xie, L., Li, S., & Soh, Y. C. (2005). HVAC system optimization – In-building
Gruber, J. K., Huerta, F., Matatagui, P., & Prodanovic, M. (2015). Advanced building section. Energy and Buildings, 37, 11–22. http://dx.doi.org/10.1016/j.enbuild.2003.
energy management based on a two-stage receding horizon optimization. Applied 12.007.
Energy, 160, 194–205. http://dx.doi.org/10.1016/j.apenergy.2015.09.049. Ma, J., Qin, S. J., Li, B., & Salsbury, T. (2011). Economic model predictive control for
Guan, X., Xu, Z., & Jia, Q. (2010). Energy-efficient buildings facilitated by microgrid. IEEE building energy systems. ISGT 2011, IEEE (pp. 1–6). . http://dx.doi.org/10.1109/
Transactions on Smart Grid, 1, 243–252. http://dx.doi.org/10.1109/tsg.2010. ISGT.2011.5759140.
2083705. Ma, Y., Borrelli, F., Hencey, B., Coffey, B., Bengea, S., & Haves, P. (2012). Model pre-
Hatami, S., & Pedram, M. (2010). Minimizing the electricity bill of cooperative users dictive control for the operation of building cooling systems. IEEE Transactions on
under a quasi-dynamic pricing model. 2010 First IEEE Int. Conf. Smart Grid Commun., Control Systems Technology, 20, 5106–5111. http://dx.doi.org/10.1109/ACC.2010.
IEEE, 421–426. http://dx.doi.org/10.1109/SMARTGRID.2010.5622080. 5530468.
Hippolyte, J. L., Howell, S., Yuce, B., & Mourshed, M. (2016). Ontology-based demand- Mahendra, S., Stéphane, P., & Frederic, W. (2015). Modeling for Reactive Building Energy
side flexibility management in smart grids using a multi-agent system. Second Int. Management. Energy Procedia, 83, 207–215. http://dx.doi.org/10.1016/j.egypro.
Smart Cities Conf. IEEE. 2015.12.175.
Howell, S., Hippolyte, J., Jayan, B., Reynolds, J., & Rezgui, Y. (2016). Web-based 3D Marzband, M., Sumper, A., Ruiz-Alvarez, A., Luis Dominguez-Garcia, J., & Tomoiaga, B.

828
J. Reynolds et al. Sustainable Cities and Society 35 (2017) 816–829

(2013). Experimental validation of a real time energy management system for mi- predictive control for an energy efficient building heating system. Applied Energy, 88,
crogrids in islanded mode using a local day-ahead electricty market and MINLP. 3079–3087. http://dx.doi.org/10.1016/j.apenergy.2011.03.009.
Energy Conversion and Management, 76, 365–376. http://dx.doi.org/10.1016/j. Saad, W., Han, Z., Poor, H. V., & Basar, T. (2011). Game-theoretic methods for the smart
apenergy.2013.02.018. grid: An overview of microgrid systems, demand-side management, and smart grid
Mathews, E., Arndt, D., Piani, C., & van Heerden, E. (2000). Developing cost efficient communications. IEEE Transactions on Power Delivery, 26, 86–105. http://dx.doi.org/
control strategies to ensure optimal energy use and sufficient indoor comfort. Applied 10.1109/msp.2012.2186410.
Energy, 66, 135–159. http://dx.doi.org/10.1016/S0306-2619(99)00035-5. Salsbury, T., Mhaskar, P., & Qin, S. J. (2013). Predictive control methods to improve
Missaoui, R., Joumaa, H., Ploix, S., & Bacha, S. (2014). Managing energy smart homes energy efficiency and reduce demand in buildings. Computers and Chemical
according to energy prices: Analysis of a building energy management system. Energy Engineering, 51, 77–85. http://dx.doi.org/10.1016/j.compchemeng.2012.08.003.
and Buildings, 71, 155–167. http://dx.doi.org/10.1016/j.enbuild.2013.12.018. Shaikh, P. H., Nor, N. B. M., Nallagownden, P., Elamvazuthi, I., & Ibrahim, T. (2014). A
Mohsenian-Rad, A.-H., & Leon-Garcia, A. (2010). Optimal residential load control with review on optimized control systems for building energy and comfort management of
price prediction in real-time electricity pricing environments. IEEE Transactions on smart sustainable buildings. Renewable and Sustainable Energy Reviews, 34, 409–429.
Smart Grid, 1, 120–133. http://dx.doi.org/10.1109/TSG. 2010.2055903. http://dx.doi.org/10.1016/j.rser.2014.03.027.
Mohsenian-Rad, A.-H., Wong, V. W. S., Jatskevich, J., Schober, R., & Leon-Garcia, A. Siano, P., & Sarno, D. (2016). Assessing the benefits of residential demand response in a
(2010). Autonomous demand-side management based on game-theoretic energy real time distribution energy market. Applied Energy, 161, 533–551. http://dx.doi.
consumption scheduling for the future smart grid. IEEE Transactions on Smart Grid, 1, org/10.1016/j.apenergy.2015.10.017.
320–331. http://dx.doi.org/10.1109/tsg.2010.2089069. Son, Y. S., Pulkkinen, T., Moon, K. D., & Kim, C. (2010). Home energy management
Molina, D., Lu, C., Sherman, V., & Harley, R. (2013). Model predictive and genetic al- system based on power line communication. IEEE Transactions on Consumer
gorithm based optimization of residential temperature control in the presence of Electronics, 56, 1380–1386.
time-varying electricity prices. IEEE Transactions on Industry Applications, 49, 7–13. Staino, A., Nagpal, H., & Basu, B. (2016). Cooperative optimizationof building energy
Moon, J. W., & Kim, J.-J. (2010). ANN-based thermal control models for residential systems in an economic model predictive control framework. Energy and Buildings,
buildings. Building and Environment, 45, 1612–1625. http://dx.doi.org/10.1016/j. 128, 713–722. http://dx.doi.org/10.1016/j.enbuild.2016.07.009.
buildenv.2010.01.009. Sun, B., Luh, P. B., Jia, Q., Jiang, Z., Wang, F., & Song, C. (2013). Building energy
Moroşan, P.-D., Bourdais, R., Dumur, D., & Buisson, J. (2010). Building temperature management: Integrated control of active and passive heating, cooling, lighting,
regulation using a distributed model predictive control. Energy and Buildings, 42, shading, and ventilation systems. IEEE Transactions on Automation Science and
1445–1452. http://dx.doi.org/10.1016/j.enbuild.2010.03.014. Engineering, 10, 588–602. http://dx.doi.org/10.1109/tase.2012.2205567.
Mossolly, M., Ghali, K., & Ghaddar, N. (2009). Optimal control strategy for a multi-zone USEF Foundation, USEF (2015). The framework explained. 1–55 http://www.usef.info/
air conditioning system using a genetic algorithm. Energy, 34, 58–66. http://dx.doi. Framework/Download-the-Framework.aspx (Accessed 1 June 2016).
org/10.1016/j.energy.2008.10.001. Wagner, A., Speiser, S., & Harth, A. (2010). Semantic web technologies for a smart energy
Mourshed, M., Robert, S., Ranalli, A., Messervey, T., Reforgiato, D., Contreau, R., et al. grid: Requirements and challenges? In: CEUR Workshop Proc. 33–36.
(2015). Smart grid futures: Perspectives on the integration of energy and ICT services. Wu, C., Mohsenian-Rad, H., Huang, J., & Wang, A. Y. (2011). Demand side management
Energy Procedia, 75, 1132–1137. http://dx.doi.org/10.1016/j.egypro.2015.07.531. for wind power integration in microgrid using dynamic potential game theory. 2011
Nest Labs, Nest (2016). https://nest.com/uk/ (Accessed 6 January 2017). IEEE GLOBECOM Work. (GC Wkshps) (pp. 1199–1204). . http://dx.doi.org/10.1109/
Oldewurtel, F., Parisio, A., Jones, C. N., Gyalistras, D., Gwerder, M., Stauch, V., et al. GLOCOMW.2011.6162371.
(2012). Use of model predictive control and weather forecasts for energy efficient Xu, M., & Li, S. (2007). Practical generalized predictive control with decentralized
building climate control. Energy and Buildings, 45, 15–27. http://dx.doi.org/10.1016/ identification approach to HVAC systems. Energy Conversion and Management, 48,
j.enbuild.2011.09.022. 292–299. http://dx.doi.org/10.1016/j.enconman.2006.04.012.
Ondeck, A. D., Edgar, T. F., & Baldea, M. (2015). Optimal operation of a residential Yamada, F. (1999). Development of air-conditioning control algorithm for building en-
district-level combined photovoltaic/natural gas power and cooling system. Applied ergy-saving. In: Proc. 1999 IEEE Int. Conf. Control Appl. (pp. 1579–1584).
Energy, 156, 593–606. http://dx.doi.org/10.1016/j.apenergy.2015.06.045. Yan, B., Luh, P. B., Sun, B., Song, C., Dong, C., Gan, Z., et al. (2013). Energy-efficient
Petri, I., Li, H., Rezgui, Y., Chunfeng, Y., Yuce, B., & Jayan, B. (2014). A HPC based cloud management of eco-communities. IEEE Int. Conf. Autom. Sci. Eng. (pp. 106–111).
model for real-time energy optimisation. Enterprise Information Systems, 10, 108–128. Yan, B., Luh, P. B., Bragin, M. A., Song, C., Dong, C., & Gan, Z. (2014). Energy-efficient
http://dx.doi.org/10.1080/17517575.2014.919053. building clusters. IEEE Int. Conf. Autom. Sci. Eng..
Pérez-Lombard, L., Ortiz, J., & Pout, C. (2008). A review on buildings energy consump- Yang, R., & Wang, L. (2012). Optimal control strategy for HVAC system in building energy
tion information. Energy and Buildings, 40, 394–398. http://dx.doi.org/10.1016/j. management. Transm. Distrib. Conf. Expo. IEEE PES (pp. 1–8). . http://dx.doi.org/10.
enbuild.2007.03.007. 1109/TDC.2012.6281687.
Papantoniou, S., Kolokotsa, D., & Kalaitzakis, K. (2015). Building optimization and con- Yang, R., Wang, Z., & Wang, L. (2011). A GUI-based simulation platform for energy and
trol algorithms implemented in existing BEMS using a web based energy management comfort management in zero-energy buildings. 2011 North Am. Power Symp., IEEE,
and control system. Energy and Buildings, 98, 45–55. http://dx.doi.org/10.1016/j. 1–7. http://dx.doi.org/10.1109/NAPS.2011.6024869.
enbuild.2014.10.083. Yang, C., Li, H., Rezgui, Y., Petri, I., Yuce, B., Chen, B., et al. (2014). High throughput
Parisio, A., Fabietti, L., Molinari, M., Varagnolo, D., & Johansson, K. H. (2014). Control of computing based distributed genetic algorithm for building energy consumption
HVAC systems via scenario-based explicit MPC. IEEE Conf. Decis. Control (pp. 5201– optimization. Energy and Buildings, 76, 92–101. http://dx.doi.org/10.1016/j.enbuild.
5207). . http://dx.doi.org/10.1109/CDC.2014.7040202. 2014.02.053.
Parisio, A., Wiezorek, C., Kyntäjä, T., Elo, J., & Johansson, K. H. (2015). An MPC-based Yuan, S., & Perez, R. (2006). Multiple-zone ventilation and temperature control of a
energy management system for multiple residential microgrids. IEEE Int. Conf. Autom. single-duct VAV system using model predictive strategy. Energy and Buildings, 38,
Sci. Eng.http://dx.doi.org/10.1109/coase.2015.7294033. 1248–1261. http://dx.doi.org/10.1016/j.enbuild.2006.03.007.
Pisello, A., Bobker, M., & Cotana, F. (2012). A building energy efficiency optimization Yuce, B., & Rezgui, Y. (2015). An ANN-GA semantic rule-based system to reduce the gap
method by evaluating the effective thermal zones occupancy. Energies, 5, 5257–5278. between predicted and actual energy consumption in buildings. IEEE Transactions on
http://dx.doi.org/10.3390/en5125257. Automation Science and Engineering, 1–13. http://dx.doi.org/10.1109/tase.2015.
Prívara, S., Jan, S., Ferkl, L., & Cigler, J. (2011). Model predictive control of a building 2490141.
heating system: The first experience. Energy and Buildings, 43, 564–572. http://dx. Yuce, B., Rezgui, Y., & Mourshed, M. (2016). ANN-GA smart appliance scheduling for
doi.org/10.1016/j.enbuild.2010.10.022. optimised energy management in the domestic sector. Energy and Buildings, 111,
Privara, S., Vana, Z., Zacekova, E., & Cigler, J. (2012). Building modeling: Selection of the 311–325. http://dx.doi.org/10.1016/j.enbuild.2015.11.017.
most appropriate model for predictive control. Energy and Buildings, 55, 341–350. Zhang, Y., Zhang, T., Wang, R., Liu, Y., & Guo, B. (2015). Optimal operation of a smart
http://dx.doi.org/10.1016/j.enbuild.2012.08.040. residential microgrid based on model predictive control by considering uncertainties
RESILENT Project (2013). D2.2-District information model. and storage impacts. Solar Energy, 122, 1052–1065. http://dx.doi.org/10.1016/j.
Ramchurn, S., Vytelingum, P., Rogers, A., & Jennings, N. (2011). Agent-based control for solener.2015.10.027.
decentralised demand side management in the smart grid. AAMAS ‘11 (pp. 5–12). . Zhou, Q., Natarajan, S., Simmhan, Y., & Prasanna, V. (2012). Semantic information
http://dx.doi.org/10.1016/j.renene.2009.02.031. modeling for emerging applications in smart grid. 2012 Ninth Int. Conf. Inf. Technol. ?
Rehrl, J., & Horn, M. (2011). Temperature control for HVAC systems based on exact New Gener. 775–782. http://dx.doi.org/10.1109/ITNG. 2012.150.
linearization and model predictive control. IEEE Int. Conf. Control Appl. (pp. 1119– Zhou, B., Li, W., Chan, K. W., Cao, Y., Kuang, Y., Liu, X., et al. (2016). Smart home energy
1124). management systems: Concept, configurations, and scheduling strategies. Renewable
Rodriguez, D. I. H., Hinker, J., & Myrzik, J. M. A. (2016). On the problem formulation of and Sustainable Energy Reviews, 61, 30–40. http://dx.doi.org/10.1016/j.rser.2016.03.
model predictive control for demand response of a power-to-heat home microgrid. 047.
Power Syst. Comput. Conf. (pp. 1–8). . http://dx.doi.org/10.1109/PSCC.2016. Zhu, Z., Tang, J., Lambotharan, S., Member, S., Chin, W. H., & Fan, Z. (2011). An integer
7541024. linear programming based optimization for home demand-side management in smart grid,
Rusitschka, S., Gerdes, C., & Eger, K. (2009). A low-cost alternative to smart metering 1205–1210.
infrastructure based on peer-to-peer technologies. 2009 6th Int. Conf. Eur. Energy Zucker, G., Habib, U., Blöchle, M., Wendt, A., & Schaat, S. (2015). Building energy
Mark. (pp. 1–6). . http://dx.doi.org/10.1109/EEM.2009.5311431. management and data analytics. Int. Symp. Smart Electr. Distrib. Technol. (pp. 462–
Široký, J., Oldewurtel, F., Cigler, J., & Prívara, S. (2011). Experimental analysis of model 467).

829