MONITORING AND MODELING BUILDING ENERGY

EXPENDITURE WITH SENSOR NETWORKS

Erwing R. Sanchez, Bartolomeo Montrucchio and Maurizio Rebaudengo

Politecnico di Torino, Corso Duca degli Abruzzi, 24, Torino, Italy

Keywords:

Energy efﬁciency, Wireless sensor networks, Sustainable technologies.

Abstract:

Residential and commercial buildings constitute one of the largest energy consumption sectors in industrial-

ized countries. This paper introduces a ﬂexible commissioning system for maximizing energy usage efﬁciency

in large and complex buildings. The system focuses on energy saving models to efﬁciently acquire and process

sensor network data, thus reducing energy consumption and costs through the whole process. Novel proto-

col mechanisms are included in the implementation of the network to reduce its energy expenditure while

maintaining a reliable communication. Special attention is placed in developing the performance monitoring

interface, speciﬁcally designed for increasing personal energy consciousness of end-users. The overall system

has been implemented at the Politecnico di Torino main site.

1 INTRODUCTION

Nowadays sustainability and green policies have be-

come real drivers for industrial and economic projects

in every modern country. Financial crisis, effects of

pollution on human health and climate changes are

only some of the biggest threats related to energy pro-

duction, uses and wastes. For these reasons, the devel-

opment of innovative concepts and systems concern-

ing energy management has achieved increasing in-

terest during last years. Furthermore, since residential

and commercial buildings represent 20-40% of the to-

tal energy demand (Perez-Lombard et al., 2008), it be-

comes evident the importance of optimizing resources

in buildings and the signiﬁcant possible savings.

The aim of this paper is to describe the creation of

an energy management system in large and complex

structures, which represents one of the most interest-

ing challenges concerning energy efﬁciency and ICT.

A non-intrusive, embedded technology was con-

sidered: Wireless Sensor Netowrks (WSN). WSNs

constitute a pervasive and ubiquitous technology

which may be deployed in the environment in or-

der to gather information about a physical phenom-

ena. Due to a combination of recent technological ad-

vances in electronics, nanotechnology, wireless com-

munications, computing and networking, it is possible

to design tiny, low-cost and low-power sensors.

The main contributions of this paper are described

in the following. Mainly, a commissioning system

based on WSN was implemented in a large building

structure. Particular concern was devoted to the efﬁ-

cient design and implementation of the WSN archi-

tecture. The system includes novel mechanisms to

tackle down the energy expenditure problem of the

network; improvements to traditional Medium Ac-

cess Control (MAC) and routing protocols are intro-

duced which leads to an energy-efﬁcient system with

a several-month lifespan. Furthermore, analysis of

captured environmental data is presented, and a model

is introduced based on cyclostationary processes.

The paper is organized as follows: Section 2 is

devoted to the description of the data management by

using wireless sensors networks and their optimized

schemes for acquisition and data transfer. Section 3

presents a data analysis framework providing energy

saving models for sensor networks, while Section 4 is

devoted to the description of the envisaged humans-

data interface, developed with the objective of creat-

ing awareness and a real perception of energy usage

to the user. Results obtained from a sensor network

deployed at Politecnico di Torino are presented and

discussed in Section 5 and Section 6 concludes the

paper.

283

R. Sanchez E., Montrucchio B. and Rebaudengo M. (2011).

MONITORING AND MODELING BUILDING ENERGY EXPENDITURE WITH SENSOR NETWORKS.

In Proceedings of the 1st International Conference on Pervasive and Embedded Computing and Communication Systems, pages 283-287

DOI: 10.5220/0003372602830287

 SciTePress

2 DATA MANAGEMENT SYSTEM

Wireless sensor networks are exploited in rather di-

verse applications to continuously monitor a given

environment. To this aim, the sensor network is

frequently queried, i.e., acquisition from all sensors

of measurements describing the state of the moni-

tored environment (Gehrke and Madden, 2004; Mad-

den et al., 2003; Yao and Gehrke, 2003) is per-

formed. However, this approach is characterized by

high energy consumption. Since main contributors to

sensor energy consumption are communication and

data acquisition (Deshpande et al., 2004), novel in-

telligent techniques for sensor network querying are

needed. Hence, devising power-efﬁcient architectures

and models for energy saving during data collection is

desired.

Raw measurements from sensor nodes are trans-

ferred to a central processing unit via the base sta-

tions, which constitute intermediate entities in mesh

networking topology. This architecture increases net-

work reliability, avoiding information losses, shirink-

ing delays due to distances in communication and en-

abling fast response to events detected.

Raw data collected is stored in the DataBase Man-

agement System (DBMS), which in turn elaborates

information to provide energy indicators for different

networks areas. Moreover, information from different

base stations are processed simultaneously, increas-

ing the overall interoperability of the system. Results

from the commissioning system are reported com-

pletely to trained staff responsible for pursuing build-

ings energy policy. In addition, information provided

to non-technical users allows to increase their per-

sonal energy consciousness, possibly modifying some

of their bad habits.

The data acquisition stage relies in different mech-

anisms and protocols which performance affects the

overall operation of the network. The sensor architec-

ture establishes the foundations of a reliable acquisi-

tion regarding, mainly, energy efﬁcient applications.

Energy-saving and reliable strategies are considered

for relevant layered protocols, e.g., MAC and routing.

A preamble sampling approach is adopted for MAC,

while an improved energy-aware collection tree algo-

rithm is implemented for the routing protocol.

2.1 WSN Architecture

WSNs are mainly deployed to monitor and report

physical measured data to a central device or base

station. Therefore, their main concern is to collect

this data from the environment, by means of sensor

nodes, and route it to the base station which perform

appropriate analysis of the collected information. In

a centralized architecture, there is only one base sta-

tion which handles all the incoming messages from

the nodes and take the decisions related to the analysis

to perform. Other alternative, based in this approach,

may be adopted such as multi-tree architectures.

In a multi-tree approach, networks are created ac-

cording with the area they cover. Hierarchical trees

are formed individually one from another. Every tree

is constituted by several sensors nodes and a base sta-

tion. Thus, each tree is a single network that inter-

acts with other networks in the building by means of

their mutual database. This approach presents many

advantages for large buildings for several reasons.

First, by having multiple base stations, sensor mea-

surements do not have to be propagated through large

paths, therefore overall energy is saved because just

a few nodes are involved in data propagation each

time. Another advantage of multiple base stations is

related to networks responsiveness to detected phe-

nomena; since sensor measurements converge faster

to the analysis maker entity, i.e., the base stations,

then, analyisis of data is performed earlier and, con-

sequently, reacting mechanisms such as alarms can be

triggered faster.

2.2 Collection Tree Protocol

The routing protocol designed is a tree-based collec-

tion protocol (CTP) where several number of nodes

are able to announce themselves as roots of the tree.

The routing protocol is address-free, which essen-

tially means that the data is not sent to a particular

root, instead the destination is, implicitly, chosen by

sending the message to the next hop in the upper level

of the tree. The main metric used by the protocol to

select next hops in the routing tree is the Expected

Transmissions (ETX), which produces a routing gra-

dient to allow generating the path towards one root.

While the initial conditions for a root establish its

ETX equals to zero, all other nodes have

ETX(n

) = ETX(n

) + ETX(L (n

, n

)),

where node n

is the parent of node n

, and L is the

function to obtain the link between two nodes. As a

result, the routing protocol performs the selection of

the path with minimum ETX between the source node

and the root node.

To improve the behavior of tradicional CTP, a new

mechanism to compute the metric was developed. Re-

ducing the number of packets transmitted provides an

efﬁcient manner to reduce energy expenditure due to

the high energy cost of the radio operation. Even if,

in CTP, the number of protocol packets decreases in

PECCS 2011 - International Conference on Pervasive and Embedded Computing and Communication Systems

284

time, a large quantity of these ﬂows in the network.

A solution to reduce the number of protocol pack-

ets consits on perform the metric calculation based

on the quantity of energy available to the node. Thus,

the node with larger energy available is responsible

for sending the packets to measure the metric. The

receiving node only sends one packet to update the

sending node. With this simple mechanism, nearly

half of the packets used for the routing protocol are

reduced and, in consequence, the overall energy us-

age of the network is reduced as well.

2.3 Preamble Sampling Protocol

Preamble sampling is a protocol designed to avoid

sharing sleep/wake up schedules among nodes, i.e.,

every node chooses independently a schedule without

synchronizing it with other nodes.

An optimization to the traditional preamble sam-

pling protocol was introduced in (Sanchez et al.,

2009) and was implemented in this deployment.

High-trafﬁc nodes are normally closer to the sink due

to their behavior to forward others packets to their

destination in a network without aggregation mech-

anism. Most of the energy used to send is due to the

waiting period of sending nodes before the receiving

nodes wake up. This time can be reduced by follow-

ing an adaptive mechanism which reduces the sleep-

ing time according to the closeness of the nodes to the

sink. The base station is always listening i.e., sleeping

time equals to zero, thus, reducing to the minimum

the sending time of high-trafﬁc nodes which are in

the ﬁrst level of the routing tree. The sleeping time of

those nodes is less than the nominal value in order to

help second-level nodes to send their packets without

too much waiting. Second-level nodes’ sleeping time

is higher than ﬁrst-level nodes’ sleeping time. This

mechanism goes on until the nominal sleeping time is

achieved by nodes that are far from the sink and do

not have to forward others packets.

3 DEPLOYMENT

CONSIDERATIONS

The commissioning system described was imple-

mented at the Politecnico di Torino main site. WSNs

for data collection were developed by using Cross-

Bow’s Telos rev. B (Polastre et al., 2005) nodes

compliant with IEEE 802.15.4 and Zigbee standards

(Tsang et al., 2009). Telos nodes were provided with

sensors for environmental parameters (temperature,

relative humidity, light). In addition, customized no-

des were implemented in parallel for acquiring oc-

cupancy indicators (human presence and CO

lev-

els). Up to 67 nodes were deployed concurrently

within several measurements campaigns that were

performed through one year.

Due to the measurement campaigns deployed in

different buildings and the dimension of the covered

areas, mesh networking strategies were limited to less

than four hops by adopting the multi-tree network ar-

chitecture. The routing protocol was address-free and

represented an optimized version of CTP (Gnawali

et al., 2009) where only the nodes with higher energy

evaluated the link quality, as described in Section 2.2.

The optimal solution was therefore designed on a 15-

minutes sampling time and the improved Low Power

Listening (LPL) MAC protocol (Polastre et al., 2004),

presented in Section 2.3. This low-power, real-time

data acquisition enabled fast response to perturbing

situations with controlled network trafﬁc and power

consumption, providing improvements on the average

lifetime of nodes.

The employed framework consider an end-user in

charge of deﬁning the energy saving strategy based

upon consumption data and comfort of building users.

To facilitate decision making, the processing stage

performs statistical and comfort computations. By ex-

ploiting the presence sensors, the system is able to

statistically single out isolated or weakly-populated

areas. Additionally, the termo-hygrometric comfort

of the considered environment can be computed by

utilizing the PMV index (ISO 7730, 2005).

Fig. 1 presents an area of the deployments: the

Hydraulic and Transport Department. We deployed

sensor nodes throughout two ﬂoors in an area of

roughly 250 m

. We selected one corner of the area as

placement for the base station. Sensors were placed

in different environments of the department; several

were positioned in ofﬁces, while others were placed

inside facilities mainly used by students, such as a

small library and small laboratories. One sensor was

placed in the main hall as a reference for the other

measurements. Another department, the Energetic

Department facility was also chosen for deployment,

mainly, because it has a large laboratory area with dif-

ferent equipment that may affect temperature or hu-

midity, and may represent and interesting focusing

area for energy-saving strategies. The area covered

by sensors was about 250 m

, as well.

MONITORING AND MODELING BUILDING ENERGY EXPENDITURE WITH SENSOR NETWORKS

285

Figure 1: Sensor deployment at Hydraulic and Transport

Department.

4 MEASUREMENT AND

ENERGY-SAVING RESULTS

The reliability of the WSN acquisition system was

more than 97% of delivery ratio, which is higher than

other approaches (Gnawali et al., 2009). The adaptive

preamble approach exploited for the MAC protocol

provides a fairer distribution of the energy expendi-

ture of the network by reducing the cost burden of

high-trafﬁc nodes towards low-trafﬁc nodes. The op-

timization of CTP allows for a reduction of almost

50% of the control trafﬁc in the network, which leads

to a decrease in power consumption and an increment

in the average lifetime of nodes to approximately six

months.

The PMV computations shows that high thermal

comfort level is guaranteed even in isolated areas,

where usually users do not beneﬁt of that situation, re-

sulting in an inefﬁcient energy usage. Fig. 2(a) shows

the temperature measurement of an isolated area of

the building on January 2010. By hypothesizing a re-

duction of heating power during working hours and

a subsequent modiﬁcation of the internal temperature

by -1.5

C, a decrease of up to 270 DD (DegreeDay) is

obtainable during the whole heating period (180 days

per year).

The same reasoning may be applied to summer pe-

riod, when cooler are activated for air conditioning.

Decreasing the internal cooling of isolated areas al-

lows to obtain a further energy saving still guarantee-

ing an acceptable thermal comfort. The overall theo-

retical energy saving obtainable may be up to 15% of

total annual energy consumption.

5 DATA TRENDS AND

AUTOCORRELATIONS

IN TIME

From the obtained data, autocorrelations functions

(ACF) of temperature, humidity and illuminance are

presented in Fig. 2(a), 2(b) and 2(c), respectively.

Temperature

ACF

-0.4

-0.2

0.2

0.4

0.6

0.8

lag (minutes)

1e+00 1e+01 1e+02 1e+03 1e+04

day

week

(a) Temperature.

Humidity

ACF

-0.4

-0.2

0.2

0.4

0.6

0.8

lag (minutes)

1e+00 1e+01 1e+02 1e+03 1e+04

week

day

(b) Humidity.

Illuminance

ACF

-0.2

0.2

0.4

0.6

0.8

lag (minutes)

1e+00 1e+01 1e+02 1e+03 1e+04

week

day

Figure 2: Autocorrelation functions.

The particular node chosen corresponds to S3 in the

Hydraulic and Transport Department which is consid-

ered as a representative sensor.

Temperature and humidity present speciﬁc time

trends in two particular periods: daily and weekly.

As part of an in-door ofﬁce scenario, the daily perid-

iocity is consistent due to the clear drift of a day.

The weekly frequence behavior becomes clearer by

analysing the differences during the weekend, i.e., an

ofﬁce generally behaves differently in the weekend

than in workdays in terms of temperature and humid-

ity because fewer individuals are present, thus, gen-

erating a weekly peridiocity that in other scenarios is

negligible.

The autocorrelation is highly inﬂuenced by these

peridiocities. A more precise analysis of the energy

consumption and behavior of a particular area of a

building can be achieved by analysing the ﬂuctuations

of the autocorrelation once these cyclic trends and the

seasonal trend are removed. To that end, a model of

the data based on the its descomposition is developed.

The descomposition is constituted by three compo-

nents: the yearly trend, a daily effect and a residual

short term ﬂuctuation.

For every time series M(t), the addition of the

three modes produces the following model

M(t) = Y (t) +

∑

D(t − nT

) + R(t),

where T

represents the time period of one day, Y (t)

is the yearly trend, D(t) represents the daily behavior

and R(t) is the residual ﬂuctuation. Y (t) is deﬁned as

the smoothed average over one day of data, in order

to reduce the inﬂuence due to daily variations. D(t)

is estimated as the daily trend, after taking out Y (t).

To address the different behavior of the weekends and

workdays, the computation of D(t) considers merely

the workdays. D(t) is produced by substracting the

yearly mode and taking the cyclic mean over a period

of one day:

D(t) = mean(M(t) −Y (t)); t ∈ {workday}.

PECCS 2011 - International Conference on Pervasive and Embedded Computing and Communication Systems

286

Temperature

Hours

0 100 200 300 400 500 600

(a) M(t).

M(t) - Y(t)

-4

-2

Hours

0 5 10 15 20 25

(b) Cyclic M(t) −Y (t) and D(t).

R(t)

-5

Hours

0 100 200 300 400

Figure 3: Descomposition values for node S3.

For the node S3, the result of this decomposition is

depicted in Fig. 3(a), 3(b) and 3(c). Fig. 3(a) shows

the complete time series, M(t), for the temperature

sensor of the node. After substracting Y (t), the cyclic

daily values for working days are presented in Fig.

3(b). For convenience, the mean value D(t) is also

shown in bold. Finally, the resulting R(t) is shown in

Fig. 3(c).

When a day does not follow closely the usual

cyclic pattern, R(t) presents peaks. The analysis of

the different measures can be performed considering

only R(t) which isolates the differences of the sensor

from the usual trend. Energy managers may utilized

this information to single out problematic areas or un-

usual behaviors in the different areas of the building.

While Y (t) and D(t) tend to be similar among dif-

ferent nodes, R(t) represents small ﬂuctuations which

can be used to compare against a reference measure-

ment.

6 CONCLUSIONS

Commissioning buildings represents usually a huge

investment for tenants; however signiﬁcant savings

may be obtained by optimizing energy usage and re-

ducing wastes. Flexibility and interoperability can be

increased considerably by exploiting wireless sensors

networks which reduce also deployment time and cost

of the monitoring system. Optimized protocols play

an importatn role on the savings and sutainability of

the sensor network. A model for analyzing sensor net-

works data allows signiﬁcant improvements and its

exploitment helps to minimize energy consumption

for data collection. The experimental results obtained

at the Politecnico di Torino case study demonstrated

the feasibility of the approach and may serve as a

preliminary reference model in developing the mon-

itoring/controlling framework of an On-going Com-

misioning system.

REFERENCES

Deshpande, A., Guestrin, C., Madden, S. R., Hellerstein,

J. M., and Hong, W. (2004). Model-driven data acqui-

sition in sensor networks. In VLDB ’04: Proceedings

of the Thirtieth international conference on Very large

data bases, pages 588–599. VLDB Endowment.

Gehrke, J. and Madden, S. (2004). Query processing in

sensor networks. Pervasive Computing, 3(1):46–55.

Gnawali, O., Fonseca, R., Jamieson, K., Moss, D., and

Levis, P. (2009). Collection tree protocol. In Proceed-

ings of the 7th ACM Conference on Embedded Net-

worked Sensor Systems, pages 1–14.

ISO 7730 (2005). ISO 7730:2005: Ergonomics of the ther-

mal environment - Analytical determination and inter-

pretation of thermanl comfort using calculation of the

PMV and PPD indeces and local thermal comfort cri-

teria. International Organization for Standardization,

Geneva, Switzerland.

Madden, S., Franklin, M. J., Hellerstein, J. M., and Hong,

W. (2003). The design of an acquisitional query pro-

cessor for sensor networks. In SIGMOD, pages 491–

502, New York, NY, USA. ACM Press.

Perez-Lombard, L., Ortiz, J., and Pout, C. (2008). A review

on buildings energy consumption information. Energy

and Buildings, 40(3):394–398.

Polastre, J., Hill, J., and Culler, D. (2004). Versatile low

power media access for wireless sensor networks. In

SenSys ’04: Proceedings of the 2nd international

conference on Embedded networked sensor systems,

pages 95–107, New York, NY, USA. ACM.

Polastre, J., Szewczyk, R., and Culler, D. (2005). Telos: en-

abling ultra-low power wireless research. In IPSN ’05:

Proceedings of the 4th international symposium on

Information processing in sensor networks, page 48,

Piscataway, NJ, USA. IEEE Press.

Sanchez, E., Chaudet, C., and Rebaudengo, M. (2009).

Improving preamble sampling performance in wire-

less sensor networks with state information. In Wire-

less On-Demand Network Systems and Services, 2009.

WONS 2009. Sixth International Conference on, pages

101–108.

Tsang, K. F., Tung, H. Y., and Lam, K. L. (2009). ZigBee:

From Basics to Designs and Applications. Prentice

Hall.

Yao, Y. and Gehrke, J. (January 2003). Query processing in

sensor networks. In CIDR. ACM Press.

MONITORING AND MODELING BUILDING ENERGY EXPENDITURE WITH SENSOR NETWORKS

287