2011 50th IEEE Conference on Decision and Control and

European Control Conference (CDC-ECC)

Orlando, FL, USA, December 12-15, 2011

Energy Management for Buildings and Microgrids

Petr Stluka, Datta Godbole, and Tariq Samad

Abstract — Intelligent consumer energy management systems [2]. Microgrid management can be seen as an optimization
will become important elements at the delivery points of the problem, as shown in Fig. 1, which illustrates an electricity
smart grid inside homes, buildings, and industrial plants. The network interconnecting various energy generation elements
end users will be able to better monitor and manage their (supply side) with energy consumption elements (demand
energy consumption, while utilities will gain more flexible
mechanisms for management of peak demands that will extend
side) and storage devices.
beyond demand response initiatives as they are implemented A. Supply Side
today. With a broader use of distributed generation many
buildings and campuses will become microgrids interconnecting The supply side feeds the microgrid with electricity that is
multiple generation, storage, and consumption devices of one or generated locally by a variety of distributed generation
several end users. We discuss how energy management and elements. These can range from complex cogeneration
control for such facilities can be viewed as a large-scale (CHP) units to stand-alone generation units (wind turbines,
optimization problem. Specific supply-side and demand-side photovoltaics, etc.) that utilize conventional or renewable
aspects include on-site renewable generation, storage
sources of energy. Supply to the microgrid also still comes
technologies, electric cars, dynamic pricing, and load
management. Technical challenges related to the optimization from the electricity distribution network, which is operated
formulation are noted – in general, mixed-integer, nonlinear, by the respective system operator. This connection can also
constrained optimization is needed. We also describe an serve for selling locally generated excess or green energy
implementation of optimization-based energy management back to the main electricity network.
solution for a hospital in the Netherlands, providing economic
details and an analysis of the savings achieved. B. Demand Side
The demand side aggregates all energy-consuming
I. INTRODUCTION devices. These aggregated loads represent various building

W ITH the increased focus on energy efficiency,

deployment of renewable energy sources, and smart
grid technologies, a growing number of buildings and multi-
systems such as lighting or heating, ventilation, and air
conditioning (HVAC). From the demand management
perspective the loads can be categorized as follows:
building facilities will become more active participants in the 1) Critical loads must be met at all times. These are
electricity market. From the system point of view, such next- typically power supplies to essential processes.
generation facilities will be autonomous microgrids with 2) Curtailable loads can be temporarily lowered. Air
capabilities to sell or buy electricity from the power network conditioning systems are a classical example.
and flexibly shift or reduce electrical loads when needed. 3) Reschedulable loads can be flexibly shifted in time to
Modern building systems organized in microgrids will avoid penalties associated with peak demand or overloading
enable improved efficiency of overall energy consumption, of the grid. Rescheduling means that some energy-intensive
reduction of emissions, and integration of green power from activities can be moved forwards or backwards in time. For
renewables. In the context of smart grid initiatives, instance, pre-cooling of a building can be done in the early
microgrids are usually considered as small-scale versions of morning – before actual cooling demand.
the centralized electricity system [1]. However, in its
C. Energy storage and electric cars
essence, a microgrid system of a building or campus can
include any type of local energy generation, distribution, Energy storage is an important element that adds more
consumption, and storage elements. Frequently, a central flexibility to the microgrid, but it also increases the grid’s
combined heat and power (CHP) plant represents a key operational complexity. Energy storage elements can bring
generation element of the microgrid and this means that a significant advantages particularly when the microgrid
heat distribution network – and possibly also thermal storage includes intermittent renewable generation sources. In the
– must be considered in addition to the electricity network future, when electric cars are widely used, their storage
capacity could be exploited in a similar way to smoothen
energy consumption profiles.
P. Stluka is with the Honeywell Prague Laboratory, V Parku 2326/18,
148 00 Prague 4, Czech Republic (phone: +420-234-625-934; fax: +420- D. Supervisory control and optimization for the
234-625-900; e-mail: [email protected]). microgrid
D. Godbole is with Honeywell ACS Labs, Golden Valley, MN 55422
USA. (e-mail: [email protected]). The safe, reliable, and cost-effective operation of the
T. Samad is with Honeywell ACS Labs, Golden Valley, MN 55422 USA microgrid requires coordinating and dispatching these
(e-mail: [email protected]).

978-1-61284-801-3/11/$26.00 ©2011 IEEE 5150

 Fig. 1. Schematic representation of the microgrid

multiple generation, consumption, and storage devices aspects of microgrid energy management, highlighting the
connected to the grid. The associated optimization problem several complexities that must be considered in optimization
is a difficult one for several reasons. First, the optimizer and control schemes. Next, in Sections III and IV, we
needs to collect and process information from individual present optimization formulations for supply-side and
components about their past, current, and predicted future demand-side problems respectively. Section V discusses a
states. Second, a significant portion of the optimized microgrid optimization solution and its application to a
variables are symbolic, indicating operational modes for hospital facility in the Netherlands. We conclude by noting
equipment (including binary variables indicating whether a some areas for future work.
specific piece of equipment should be running or not). Third,
the optimization task has to span a reasonably long future II. ASPECTS OF MICROGRID ENERGY MANAGEMENT
time interval (e.g., 24 hours) to assure consistency of Depending on the types of devices connected to the
operation [3]. Finally, the context within which the microgrid microgrid, the energy management problem can cover
operates is uncertain because of the limited predictability of optimization of the microgrid’s supply side or demand side,
renewable generation, future energy demands, and dynamic or optimization of the whole system.
prices. The optimization objective is to minimize the total Supply side optimization. The basic supply side
energy bill within the operational, safety, reliability, and configuration includes CHP and other conventional
environmental constraints of the microgrid. generation devices. The system is connected to a public
We note that microgrid operation also involves fast electricity grid with two-way energy flows. All demand side
electrical control of the phase, frequency, and voltage of loads are considered critical and must always be met. The
individual elements. A special functionality that integrates energy management decisions then consist in the dispatching
both electrical control and system coordination relates to of individual energy supplies, switching among various types
management of emergency situations in the grid [4]. There of fuel, and optimizing the time for purchasing electricity or
can be a need to isolate feeder faults and enable healthy selling excess electricity back to the grid.
sections of the microgrid to continue operating during a Integration of renewable generation sources. Adding
fault, or transition to an islanding mode where the microgrid renewable generation sources makes the energy management
can operate independently of the main grid. In this paper we tasks more complicated. Renewable sources bring more
assume that these high-speed controllers are available and we uncertainty as their operation depends on hardly predictable
focus on system-level optimization of the microgrid. environmental conditions such as wind speed or sun
In the rest of this paper, we first discuss a number of irradiation. Then the objective is to balance different forms

5151
 sources on the supply side [7],[8], as well as forecasts of
future energy consumption on the demand side. Both
forecasts are primarily influenced by future environmental
conditions that can be available in the form of weather
forecast. Energy load forecasts also strongly depend on
behavioral patterns, which are tightly linked to calendar data.
The other inputs to the system optimization algorithm
include pricing data, equipment models, and characteristics
of the curtailable or reschedulable loads.
Load forecasting. Modeling and forecasting of the energy
consumed on the demand side of the microgrid usually leads
to implementation of forecasting models for commercial or
residential buildings and other types of facilities. Each
building has a specific energy consumption pattern that is
composed from daily, weekly, and seasonal cycles [9].
Models are run to predict future demand for the interval of
interest, typically one day ahead.
Renewable generation forecasting. Forecasting of
renewable power production is a significantly more difficult
task than energy load forecasting due to the immediate
impact of fluctuating primary factors – wind or cloudiness.
a) Wind power forecasting techniques range from quite
simple ones to rather complex [10]. The simplest methods
Fig. 2. Input data and preparation steps for microgrid optimization
are based on climatology or averages of past production
values. The advanced approaches for short-term (4-6 hours)
wind power forecasting require predictions of wind speed,
of supplies: renewables, conventional sources and direct which allows prediction of the produced electricity through
imports from the main grid. the so-called power curve, provided by the manufacturer or
Optimal use of energy storage. Energy storage devices are inferred from historical production data.
used to improve the generation – consumption balance of the b) Solar power forecasting is mostly based on short-term
microgrid by temporarily storing the surplus energy. This forecasting of cloudiness [11]. While motion patterns for
may be particularly advantageous for systems with renewable “stable” clouds are relatively predictable, a major challenge
energy sources, which are highly intermittent. To optimally is with convective events of “unstable” clouds. The
use the storage capacity it is necessary to determine optimal forecasting methods work with digitized cloud masks whose
charging and discharging schemes for each device that will motion vectors are identified and projected to the future.
take into account the dynamic environment. In both cases the forecast accuracy is inevitably limited
Demand-side optimization. When all loads are considered and prediction errors achieved by the best forecasting
to be critical, there is no opportunity to optimize the demand solutions are usually in the range of 5-15%.
side of the microgrid. Otherwise, the optimization will Equipment models should be reasonably simple but
include lowering or rescheduling some of the loads. In realistic to assure the optimization results are meaningful.
current practice, the optimal control of building systems, Typically, each piece of energy generation equipment can be
such as HVAC, is frequently addressed as a stand-alone characterized by an efficiency curve, which can be obtained
problem, not necessarily connected to load management. from the equipment manufacturer or estimated from
Model predictive control (MPC) techniques have been historical data. Other characteristics can include ramp-up and
recently studied in this context [5],[6]. ramp-down rates, start-up costs, normal running costs, or
The optimal operation of the microgrid as a whole can be minimal/maximal required up and down times. Energy
achieved when both supply and demand sides are optimized storage devices are usually characterized by maximum
simultaneously, which however also increases complexity of storage capacity and by charging and discharging rates.
the energy management problem in mathematical terms. Pricing data is needed to allow comparison between
Before starting any system optimization task, specific input energy costs from self-generation and utility purchase. In
data needs to be collected and several pre-processing steps addition to electricity, cost details of natural gas, heating oil,
executed as illustrated in the workflow in Fig. 2. For and other fuels must be known as well. Electricity prices can
reaching the optimum or near-optimum solution over a given follow either the traditional static tariffs, or dynamic pricing
horizon of interest (e.g., 1 day) it is important to consider mechanisms such as real-time pricing (RTP).
forecasts of future electricity production from renewable Load characteristics come into play in cases of demand

5152
side optimization, known also as load management. region is closed, bounded and convex, the problem has a
Curtailment characteristics can include the percentage of global minimum (possibly non-unique) solution and can be
load that can be curtailed, the cost of curtailment, the length solved using a suitable numerical method of convex
of time for which the load can be curtailed, and the mathematical programming.
maximum frequency of curtailments. Reschedulable loads
B. Unit Commitment
can be characterized by maximum acceptable time to
reschedule, cost of rescheduling, lead time needed before The economic dispatch problem presented above for one
rescheduling can take effect, etc. instant of time is based on the assumption that all parallel
generation units i = 1, ..., N are running. Let us assume that
III. FORMULATING THE SUPPLY-SIDE PROBLEM the system consists of N generating units working in a
parallel configuration. Decisions on which of them to run are
Microgrid supply side optimization problems are quite
the outcomes of the unit commitment problem, which
similar to the traditional problems known from the domain of
introduces binary decision variables X t ,i for each unit i.
bulk power generation. This is true primarily for the
problems of unit commitment and economic dispatch – two X t ,i = 1 if unit i is ON at time t, and X t ,i = 0 if unit i is
short-term optimization tasks that need to be solved to OFF. Then a combined formulation of both the economic
minimize the total fuel cost or to maximize the total profit. dispatch and the unit commitment problems is as follows:
These problems are typically considered from a utility Minimize (4)
perspective but they apply also to microgrid configurations. 
T N

In either case, the responsible party must ensure that the total ∑ ∑ [X ⋅ ( f (P ) + C
t =1 i =1
t ,i i t ,i i
fixed
)+ C start
t ,i ]
max( X t , i − X t −1, i ,0) + Pt ,u Rtsell 

generation must equal the forecasted demand. Unit subject to
commitment is the process of deciding when and which N

generating assets to start-up and shut-down. Economic ∑P

i =1
t ,i + Pt ,u = Dt (5)
dispatch is the process of deciding what the individual power
outputs of the scheduled generating assets should be at each Pi ,min X t ,i ≤ Pt ,i ≤ Pi ,max X t ,i (6)
time-point. Both unit commitment and economic dispatch Pu ,min ≤ Pt ,u ≤ Dt (7)
problems are interdependent and should be solved X t ,i ∈ {0,1} (8)
simultaneously, although a possible solution strategy may of
course apply decomposition ideas [12][13]. Note that X 0,i = X 0,i (9)
renewable generation sources are not considered in the The problem is now formulated over multiple periods of
following problem formulations because, in principle, the time t = 1,…,T. The objective is to determine for each time
renewable power production cannot be scheduled. interval t optimum values of Xi, Pi, and Pu that will minimize
the function (4) over the entire time interval 1,…,T. Note
A. Economic Dispatch that solving the optimization problem requires forecasting of
The basic problem of economic dispatch of the simplest future load demands D1, …, DT.
configuration of N parallel energy generation units can be The objective function (4) consists of, term by term: the
described mathematically as a cost minimization problem: operating costs (we have included fixed running costs Ci fixed
N
Minimize F = ∑ f i ( Pi ) (1) here), start-up costs Ctstart
,i
, and also costs associated with
i =1
purchasing electricity Pt,u from the utility at the sell
The minimization is over the energy generated in the units,
price Rtsell in interval t. Equation (5) represents the demand
P1 ... PN. F is the total cost and f i ( Pi ) is the cost function of
constraints – i.e. the demand D must be met by a
the i-th unit. The function characterizes the dependence of
combination of locally produced energy Pi and energy
operation costs ($) on the generated energy P (kW or MW)
supplied from the main electricity grid Pu. Inequalities (6)
and includes energy conversion efficiency. The minimization
and (7) correspond to restrictions on production limits and
is subject to the following constraints:
N
the amount of purchased electricity that must not exceed the
∑P = D i
(2) respective demand Dt. Expression (9) specifies initial states
i =1 of generation units, which can be 0 (unit is OFF) or 1 (unit is
Pi ,min ≤ Pi ≤ Pi ,max (3) ON) as defined in (8). The formulation can be further
Where D is the energy demand that must be met. As we extended by adding constraints for minimal up and down
are focused on local generation and microgrids, transmission times, detailing the start-up costs, or including ramp rates.
losses are not considered in equation (2). The optimization Differences between optimization of microgrids and bulk
problem is formulated for one instant of time. For many power generation. Microgrids are more complex and more
applications it is reasonable to assume that the cost functions flexible in configuration than a simple parallel ordering,
are convex. In this case, the objective function F is also which is typical for large power plants. The microgrid
convex. Since all constraints are linear and the feasible frequently requires joint optimization of both electrical and

5153
thermal energy. A significant difference between power determining optimal values Lctrl
t
over the entire time interval
plants and microgrids is also in how the demand side is 1,...,T. The formulation above is simplified as each of the
handled. While the meeting of demand requirements is a base and controllable loads is actually composed of multiple
strong constraint for electricity generating plants, in individual device loads. Also it is important to note some of
microgrids the demand is more flexible and may represent the other neglected aspects. Firstly, certainty in the load
another degree of freedom for optimization. In the following profile is assumed. The controllable part of the load is
section, we will formulate the demand side optimization assumed to be completely fragmentable (over the time
problem including integration of distributed renewable horizon), and no part of the load is discretionary. The
generation sources and storage devices. formulation also assumes a “planning” approach when the
prices are communicated in advance and the consumer
IV. DEMAND-SIDE FORMULATION determines the load profile over a future time window.
Active participation of buildings, campuses and other Distributed renewable sources. With the ability to
facilities in the smart grid environment assumes a scenario in generate power locally and to sell any excess production
which the utility provides dynamic prices to these consumers from renewables to the utility, consumers have additional
who can then decide how to adapt their usage to minimize decisions to make. In the absence of storage capability, on-
their consumption cost. We can start to formulate this site generation at any time must either be used to power
decision problem as follows: existing loads or sold back to the utility. The updated
T
formulation is as follows.
Minimize ∑R t
sell
Lt (10)
T

subject to
t =1 Minimize ∑ (R
t =1
t
sell
Pt , u − Rtbuy Pt ,excess
LG ) (10a)

Lt = Lbase
t + Lctrl
t
(11) subject to
Lctrl ≥0 (12) Lt = Lbase
t + Lctrl
t
(11)
t
T

∑L
T ctrl
= Lctrl (13)
∑L
t =1
ctrl
t = Lctrl (13)
t =1
t

Here Rtsell is the electricity price charged for time interval t Pt ,u + Pt cons
, LG = Lt
(14)
and Lt is the consumer’s load at t. This load is composed of a Pt , LG = Pt cons excess
, LG + Pt , LG
(15)
fixed base amount for each time unit, Lbase t
, and a non- ctrl
L ,P cons
,P excess
≥0 (16)
t t , LG t , LG
ctrl
negative controllable (reschedulable) – amount L t
(11). Here we distinguish between the electricity price charged
ctrl
The total controllable amount L over the day is fixed (13). by the utility Rtsell and the utility’s payment for consumer-
In this case, the optimization problem is to decide how to supplied power Rtbuy . The load demand Lt is now covered by
spread and/or reduce the load over a planning horizon (e.g. 1
day) so as to minimize the total cost. This requires a combination of utility-supplied power Pt,u and locally

Fig. 3. Model parameters of a boiler

5154
 Fig. 4. Building’s microgrid system configured in VERA

generated and consumed power Pt cons

, LG
(14). In addition to Pt ,u + Pt cons cons
, LG + Pt , s = Lt (17)
load management decisions on Lctrl
t
(11), other decisions (
St = St −1 +η c Pt stored stored
,LG + Pt ,u ) (
− Pt cons grid
,s + Pt ,s )
/η d (18)
cons
must be made about what proportion P t , LG
of the on-site S min ≤ S t ≤ S max (19)
generation Pt , LG should be consumed and what proportion (P stored
t , LG + Pt ,stored
u ) (
= 0 ∨ Pt cons
,s + Pt ,grid
s =0 ) (20)
Pt excess
, LG
should be sold to the grid (15). Lctrl cons excess stored cons grid
t , Pt , LG , Pt , LG , Pt , LG , Pt , s , Pt , s ≥0 (21)
Integration of storage. Another important enhancement is Respective charging and discharging efficiencies ηc and ηd
the integration of storage devices. This extends the demand- are considered in (18). The state of charge St must always be
side problem formulation in a way that the load within its lower and upper bounds (19), while constraint (20)
demand Lt can be now covered by a combination of utility- ensures that the storage device is not simultaneously being
supplied power Pt,u, locally generated power Pt cons , and charged and discharged. Additional decision variables are
, LG
introduced: the amount of local generation Pt stored
, LG
that should
power provided by the storage device Pt cons
,s
(17). We model
be used to charge storage and the amount of stored power
a storage device as a state of charge St (18) that can be
Pt cons
,s
that should be used to service loads. Simplifications
affected positively by some amount Pt ,stored
LG
of the locally
here include omission of constraints related to maximum and
generated power and some amount of utility supplied minimum charging and discharging rates and cycling
power Pt stored
,u
and negatively by discharging the storage between charging to discharging modes.
device to service loads Pt cons
,s
and to provide grid power Pt ,grid
s
.
This leads to the following problem formulation.
T
V. VERSATILE ENERGY RESOURCE ALLOCATION
Minimize ∑ (R
t =1
t
sell
Pt ,u − Rtbuy Pt excess
, LG )
− Rtbuy Pt ,grid
s
(10b) Versatile Energy Resource Allocation (VERA) is an
energy management software solution that is focused on the
subject to optimization of microgrids with renewable generation and
Lt = Lbase
t + Lctrl
t
(11) storage. It is also applicable to other integrated systems that
T combine various forms of cooling, heating, and power
∑L
t =1
ctrl
t = Lctrl (13)
generation, such as CHP plants. Target markets include

5155
campuses, military bases, hospitals, office buildings and
residential neighborhoods.
VERA can solve a combination of problems including unit
commitment, economic dispatch, fuel switching, balancing
of local generation with utility purchases, and optimal
utilization of the capacity of storage devices. The underlying
optimization task is currently formulated as the supply-side
problem (Section III) with integrated aspects of renewable
generation and storage. VERA finds the most cost-effective
setpoint schedules for generation equipment, while
attempting to achieve the lowest operation costs. The
resulting solution complies with constraints such as meeting
all energy demands, equipment capacities, influence of
changing weather on equipment efficiency, variable
availability of units, and maintenance schedules. VERA also Fig. 5. The relation between monthly savings and outdoor
takes into account principal economic aspects: time temperature
dependent costs of purchased energy and fuels, penalties for
emissions, earnings from selling energy, start-up and shut- load forecast. This can be partially overcome by running
down costs of equipment, and fixed costs of operation. Fig. 3 more model scenarios.
provides an illustration of such parameters for a boiler. B. Application – hospital utility optimization
A. Solution elements The optimization technology of VERA was implemented
Two advanced modules – load forecasting and optimal in a project for the Atrium Hospital in Heerlen, the
resource allocation – represent the core of VERA. Netherlands. It is a 850-bed hospital, which, at the time of
Load forecasting. VERA uses methodology known from project implementation, had an average annual energy
non-parametric statistics (as locally weighted regression) and consumption of approximately 3.7 million m3 of natural gas
machine learning (as memory-based learning). The and imported 2.5 GWh of electrical energy from the public
differentiating feature is that the forecasting algorithm runs grid. This represented a variable energy cost of
on top of a history database, and the local regression models approximately €1,200,000 per year.
are built on the fly using only a fraction of the most relevant This hospital built a modern and state-of-the-art
past data points. The regression model y = f (x1, x2, …, xM) trigeneration utility system in the late 1990s to feed steam,
represents the correlation between energy consumption y and heat, electricity, and cooling to the hospital’s utility
a number of independent influencing factors x1, …, xM – networks. The energy generating equipment consists of a hot
mainly weather conditions, calendar-based variables, and water boiler, two steam boilers, an electrical and absorption
seasonal effects. Ambient temperature usually has the key chiller, and two gas engine CHP units. A schematic layout of
impact, while the other environmental factors like humidity, the hospital utility is shown in Fig. 4. Electricity can be
wind speed, cloud cover, or sun irradiation can be used for generated using own electricity generators or imported from
better interpretation and finer modeling of the demand data. the public grid, absorption or electrical chillers can be used
Calendar-based variables can help with capturing behavioral for cooling purposes, and finally, heat can be generated in
patterns. These variables include time of day, which is the CHP gas turbines or using the steam or hot water boilers.
defined on the closed interval <0h;24h>, and also categorical The operating strategy in the past was based on the fact
variables like the day of week and holiday and special day that fuel prices were low and electricity prices were high.
indicators, which enable clustering of similar days into However, later due to various reasons, economical energy
coherent groups. Details of the data-centric forecasting management has turned out to be a very difficult task. These
approach used in VERA can be found in [15]. reasons include the liberalization and liquidity of the gas and
Optimal resource allocation. VERA uses a Sequential electricity market, environmental taxes, CHP subsidies, the
Quadratic Programming (SQP) solver for finding a solution flexibility of configuration of the hospital utility, economic
of the nonlinear optimization task. The nonlinearity is given pressures, and the continuously varying energy demands.
by the cost functions fi of individual units, which usually To improve the quality of energy management decisions,
correspond to the nonlinear efficiency curves (as in Fig. 3). first generations of load forecasting and optimal resource
The optimization problem is defined as a dynamic allocation technologies were installed and configured to
programming problem over the time interval of 24 hours and optimize the operation within a two-days-ahead period.
solved with a solution step ranging typically from 15 to 60 Evaluation of the new energy management concept was done
minutes. Obviously, the accuracy and reliability of the primarily in years 2003 and 2004 (see [14] for details). The
optimal solution relies on the accuracy and reliability of the application is operating for eight years and the results are

5156
very encouraging. In short, the total gas consumption has 3. Interactions between system-level energy management
decreased and the operation of the cogeneration units and the and power flow control. The objective of power flow control
absorption chiller has become more economical. The units is to transform the raw quality power generated by each
electrical import and the use of the electrical chillers have local source to grid quality power. This is essential for
increased as well as the use of the emergency cooler, which ensuring smooth transitions among different modes of
was sometimes active during low heat and cooling demand at operation. Although power flow control is a different type of
a high electricity price. problem than presented in this paper, it makes sense to study
The correlation between monthly savings and average possible interactions between both.
outdoor temperature (Fig. 5) shows that the greatest savings
were achieved during summer months when the operation Acknowledgements: We thank the reviewers of the
can be improved by optimal coordination of absorption and original version of this paper for several helpful comments.
compressor chillers and gas engines. As prices of electricity
and fuels are growing, the savings are more and more REFERENCES
significant. The annual savings achieved every year until [1] R. Galvin, K. Yeager, J. Stuller, “Perfect Power – How the Microgrid
2010 varied between 6% and 12% of utility costs. The Revolution Will Unleash Cleaner, Greener, and More Abundant
Energy”, McGraw-Hill 2009, ISBN 978-0-07-154882-3.
achieved return of investment was shorter than 1 year.
[2] M. Arnold, R. R. Negenborn, G. Andersson, and B. De Schutter,
“Distributed predictive control for energy hub coordination in coupled
VI. CONCLUSIONS AND FUTURE WORK electricity and gas networks,” in Intelligent Infrastructures, R. R.
Negenborn, Z. Lukszo, and H. Hellendoorn, Eds. Dordrecht, The
In this paper, an optimization approach is presented to Netherlands: Springer, 2010, pp. 235–273
support energy management decisions on both supply and [3] V. M. Zavala, M. Anitescu, E. Constantinescu, S. Leyffer, J. Wang,
demand sides of building microgrid systems. The G. Conzelmann, “Proactive Energy Management for Next-Generation
Building Systems”, in: Proc. IBPSA-USA SimBuild 2010, New York,
documented results of the optimization-based application for NJ, USA, August 11-13, 2010.
the hospital in the Netherlands indicate the promise of this [4] R. Firestone, C. Marnay, "Energy Manager Design for Microgrids",
approach. For our future work we are considering the Reserach Report LBNL-54447, January 2005.
[5] F. Oldewurtel, A. Parisio, C. N. Jones, M. Morari, D. Gyalistras, M.
following main directions. Gwerder, V. Stauch, B. Lehmann, K. Wirth, “Energy Efficient
1. Full integration of the supply side and demand side Building Climate Control using Stochastic Model Predictive Control
formulations. This is the natural next step, which however and Weather Predictions”, in: Proc. 2010 American Control
Conference, Baltimore, MD, USA, June 30 – July 2, 2010.
may bring new challenges related to the overall solution
[6] Y. Ma, F. Borrelli, B. Hencey, B. Coffey, S. Bengea, A. Packard, M.
complexity and dealing with the inherent uncertainty. Wetter, P. Haves, “Model Predictive Control for the Operation of
a) Need for a cost-effective approach. The microgrid Building Cooling Systems”, in Proc. of IEEE American Control
optimization is a complex mathematical problem whose Conference, Baltimore, Maryland, 2010.
[7] A. Uehara, T. Senjyu, Y. Kikunaga, A. Yona, N. Urasaki, “Study on
formal description belongs to the category of MINLP Optimum Operation Planning of Wind Farm / Battery System using
problems. For real applications it is necessary to determine Forecasted Power Data”, in Proceedings of PEDS 2009 conference,
the right balance between the level of detail of the 907 – 912, 2009.
[8] M. K. C. Marwali, H. Ma, S. M. Shahidehpour, K. H. Abdul-Rahman,
mathematical model, time and resources available for “Short-Term Generation Scheduling in Photovoltaic-Utility Grid with
calculation, and the complexity of optimization method. At Battery Storage”, IEEE Transactions on Power Systems, Vol. 13, No.
any rate, the total cost of the solution (configuration effort, 3, August 1998.
[9] K. Marik, Z. Schindler, P. Stluka, “Decision support tools for
cost of commercial software and its maintenance, upgrades
advanced energy management”, Energy, Volume 33, Issue 6, 858-
of communication infrastructure and hardware) must be in 873, Elsevier 2008.
relation to its potential benefits (cost savings). [10] A. Botterud, J. Wang, “Wind Power Forecasting and Electricity
b) Dealing with uncertainty. Uncertainty relates to several Market Operations”, in Proceedings of 32nd IAEE International
Conference, June 21-24, San Francisco, California, 2009.
operational aspects, including uncertainty of future energy [11] M. Ahlstrom, J. A. Kankiewicz, “Perspectives and Understanding on
demands, renewable generation, and dynamic prices. To Solar Power Forecasting”, Utility-Scale PV Variability Workshop,
correctly address the uncertainty may require adoption of Cedar Rapids, Iowa, October 2009.
[12] F. Gao, G. B. Sheble, “Economic Dispatch Algorithms for Thermal
stochastic optimization methods, such as stochastic SQP, as Unit System Involving Combined Cycle Units”, 15th PSCC
well as higher-level probabilistic decision-making concepts, conference, Liege, Belgium, August 2005.
such as those based on Bayesian theory. [13] E. Dotzauer, K. Holström, H. F. Ravn, “Optimal Unit Commitment
and Economic Dispatch of Cogeneration Systems with a Storage”, in
2. Synchronization of system-level energy management Proceedings of 13th Power Systems Computation Conference, pp. 738
with building controls. The formulations presented in this – 744, Trondheim, Norway, 1999.
paper are primarily addressing the problem of optimal [14] Z. Schindler, “Efficient Load Forecasting and Optimal Resource
dispatching of all building devices. There is a natural Allocation”, in Proceedings of AICARR International Conference,
Milan, 2004.
opportunity for synchronization between this dispatching [15] Z. Beran, K. Marik, P. Stluka, “Data-Centric Demand Forecasting for
problem and the operation of building controls, including Utilities”, Computer Aided Chemical Engineering, 2006, vol 21, pp
thermostats and unitary and plant-level controllers that 1809 - 1814.
ensure smooth operation of the respective building systems.

5157