Accepted Manuscript

Title: Occupant Centered Lighting Control for Comfort and

Energy Efficient Building Operation

Author: Zoltán Nagy Fah Yik Yong Mario Frei Arno Schlueter

PII: S0378-7788(15)00175-9
DOI: http://dx.doi.org/doi:10.1016/j.enbuild.2015.02.053
Reference: ENB 5727

To appear in: ENB

Received date: 11-12-2014

Revised date: 21-2-2015
Accepted date: 23-2-2015

Please cite this article as: Zoltán Nagy, Fah Yik Yong, Mario Frei, Arno Schlueter,
Occupant Centered Lighting Control for Comfort and Energy Efficient Building
Operation, Energy & Buildings (2015), http://dx.doi.org/10.1016/j.enbuild.2015.02.053

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
 Occupant Centered Lighting Control for Comfort and
Energy Efficient Building Operation

t
ip
Zoltán Nagy1 , Fah Yik Yong1 , Mario Frei, and Arno Schlueter
Architecture and Building Systems, Institute for Technology in Architecture

cr
Department of Architecture, ETH Zürich, John-Von-Neumann Weg 9, 8093 Switzerland

us
Abstract

an
We present an adaptive control strategy for lighting control in office spaces,
with the aim to reduce energy consumption and provide occupant comfort.
M
Based on the premise that each occupant is unique, yet consistent in his
actions, the set-points for switching the artificial lights on and off are derived
dynamically from statistical analysis of the occupancy sensor data, and from
d

interaction with the occupant. We present a 6-week case study in 10 offices,

te

where we achieved 37.9% and 73.2% energy savings compared to a standard

setting control baseline and a worst-case scenario, respectively. We show that
p

individual offices determine individual set-points and discuss their adaptive

ce

nature.
Keywords:
Ac

building control, energy efficient buildings, smart buildings, user interaction

Email address: [email protected] (Zoltán Nagy)

1
Equally contributing co-first authors

Preprint submitted to Energy and Buildings February 20, 2015

Page 1 of 40
 1 1. Introduction

2 Increasing energy efficiency is one of the key strategies for the reduction of

t
ip
3 energy consumption and carbon dioxide emission as part of the efforts to mit-
4 igate climate change. Compared to other measures, it is generally more cost

cr
5 effective and can be implemented more quickly. Recently, more and more fo-
6 cus has been placed on energy efficient buildings. It is estimated that almost

us
7 a third of the world’s energy is consumed in buildings, both commercial and
8 residential [1].In Switzerland, lighting in residential building alone accounts

an
9 for up to 12% of total electricity consumption [2] and lighting in the service
10 and agriculture sectors accounts for almost 24% of total electricity used [3].
M
11 This has led to increasing efforts to improve lighting systems in buildings. In
12 this paper, we focus on lighting systems in an office environment.
Good-quality lighting is an essential part of occupant comfort. It can
d
13

14 support visual performance and interpersonal communication and improve

te

15 our feelings of well-being. Poor-quality lighting on the other hand, can be un-
comfortable, confusing and even inhibit visual performance. In general, light
p

16

17 not only provides the physical stimulus necessary for visual performance, but
ce

18 also communicates or reinforces certain cues that influence people’s subjec-

19 tive impressions of the environment surrounding them. These impressions
Ac

20 can make a significant impact on long-term occupant satisfaction [4]. This

21 suggests that good lighting is important in an office environment, and quality
22 lighting should be provided to maximize visual comfort as well as to allow
23 performance of tasks. If no other factors are considered, then this can easily
24 be achieved by providing constant adequate lighting suited to the occupant.
25 However, the balance between cost effectiveness and energy consumption

Page 2 of 40
26 must always be considered.
27 Efforts have been made in designing lighting control systems capable of

t
28 reducing energy consumption and ultimately achieving cost savings. The

ip
29 very first conceptions included the use of occupancy sensors, where the idea

cr
30 was to simply switch on and off the lights automatically whenever the target
31 room is occupied or unoccupied. Many works have shown that the integration

us
32 of such devices alongside a controller is able to achieve significant savings in
33 energy consumption. Simulations have show that it is possible to achieve

an
34 energy savings of up to 38% with occupancy sensors compared to manual
35 switching [5, 6] while an occupation study by Han et al. reports possible
36 savings of up to 40% [7]. An experimental study done by Jennings et al.
M
37 reports 20-26% savings [8].
38 There are however drawbacks to control via occupancy monitoring. A
d

39 conventional occupancy sensor works with a time delay(TD) setting, i.e. the
te

40 room is considered unoccupied and the lights switched off only after a time
41 period of non-activity equal to TD. Higher energy savings can be realised with
p

42 a low TD, but a low TD value will also lead to false triggers where the lights
ce

43 are switched off even though the room is still occupied [5]. This may happen
44 for example when the occupant is working on his computer and is stationary
Ac

45 for a period of time longer than the TD, and usually leads to annoyance of the
46 occupant and reduces the perceived quality of lighting control system. Also,
47 in a system where occupancy sensors are used to switch the lights both on
48 and off, energy waste compared to manual switching occurs, because lights
49 may be turned on automatically even when there was sufficient daylight
50 available [6]. Further, occupants in rooms with occupancy senors are less

Page 3 of 40
51 likely to switch off the lights manually as they rely on the sensors to control
52 the lights for them [9]. This tendency reduced the energy savings from the

t
53 occupancy sensors by as much as 30%, and is an example for the rebound

ip
54 effect observed in energetically retrofitted buildings [10].

cr
55 Recently, more advanced controls have been developed with the aim of
56 achieving further economy on energy consumption as well as improving oc-

us
57 cupant’s comfort. This includes implementations with adaptive and learn-
58 ing capabilities to overcome the drawbacks of occupancy sensors mentioned

an
59 above, as well as inclusion of illuminance sensors to incorporate the avail-
60 abilility of ambient and day lighting. Illuminance sensors were included in
61 addition to occupancy sensors in [11, 12], achieving as much as 65% and 69%
M
62 energy savings respectively compared to the baseline of having the lights on
63 for all day. A simple statistical method to model the occupant’s activities
d

64 is proposed in [13]. By adapting the TD in the occupancy sensors specifi-

te

65 cally to each occupant, 5% more energy savings can be achieved compared

66 to the conventional TD setting for a single occupant. Other such smart or
p

67 intelligent controls have also been proposed. In [14], an experimental study

ce

68 conducted using a similar statistical approach to [13] reported a reduction

69 in energy consumption of 35% to 75%. More complicated works involving
Ac

70 advanced behavioural models [15], fuzzy logic [16] and sensor networks [17]
71 show that there is indeed potential to achieve further energy savings from
72 such smart control systems.
73 Inspired by [13], in this paper we introduce an occupant centered lighting
74 control scheme, which derives its set-point for TD based on statistical analysis
75 of the motion sensor data, and through interaction with the occupant for

Page 4 of 40
76 the light level threshold. This contributes to the comfort of the occupant
77 by adapting the desired illuminance levels and occupancy sensor TD values

t
78 to this preferences, in addition to removing the need to manually switch

ip
79 the lights on. Further, energy consumption is reduced as the light is only

cr
80 switched on when required. This paper discusses the control system, and
81 presents results of a six week experimental study conducted in the offices of

us
82 the authors, in a total of 10 rooms. We show that the individual set-points
83 are achieved relatively quickly, and demonstrate up to 37.9% energy savings

an
84 respectively compared to a standard setting control baseline, and up to 73.2%
85 compared to a worst-case baseline.
86 The main advantages and characteristics of our work are
M
87 • A simple setup which makes use of only two sensors, typically available
in standard building management systems (BMS)
d
88
te

89 • Experimental study in real world environment with multiple partici-

90 pants
p

91 • Presentation of the learning curves which are required for the adapta-
ce

92 tion of the various set-points

• Discussion of the applicability for different room types

Ac

93

94 The remainder of the paper is organized as follows. Section 2 explains

95 the occupant specific set-points, while Section 3 describes the implementation
96 of the hardware and the controller. The experimental study is detailed in
97 Section 4 and its results are presented in Section 5. Section 6 discusses our
98 findings and concludes the paper.

Page 5 of 40
99 2. Occupant specific set-points

100 In this Section, we discuss the occupant specific set-points for the time-

t
ip
101 delay and the illuminance threshold.

cr
102 2.1. Time Delay

103 The activity level of an occupant is assumed to vary with the time of

us
104 the day and shows a particular trend unique to that occupant. It is further
105 assumed that this daily trend is relatively constant over a period of time,

an
106 e.g., a few weeks. Since the behaviour of each occupant is unique, the TD
107 setting of the occupancy sensor should be adapted to the occupant in order
to maximize potential energy savings.
M
108

109 Our approach is inspired by [13]. The occupancy of the room is moni-
110 tored with passive infrared (PIR) motion sensors. Every period of activity
d

111 and inactivity is registered throughout the day. Figure 1 shows a sample of
te

112 the signal registered by an occupancy sensor, with the arrows representing
113 the periods of inactivity in a room. By taking into account all periods of inac-
p

114 tivity after a sufficiently long time, a histogram of the inactivity period, and
ce

115 subsequently an empirical cumulative distribution can be obtained. Given

116 this function, the TD value is then taken at a certain desired confidence
Ac

117 level. This level describes the probability, pT D , with which the room can be
118 considered as unoccupied. With this approach, the target probability can be
119 defined globaly, but it results in different settings for each room based on the
120 individually recorded data. As an example, Figure 2 shows a TD value of
121 546 seconds being determined for pT D = 95%. In other words the probability
122 of a false trigger is 5%. Finally, by considering data only from a certain time

Page 6 of 40
123 period, e.g., the last 1-2 months, the setting can adapt itself to changing of
124 occupancy, reorganization, or potential seasonal variations.

t
ip
125 [Figure 1 about here.]

cr
126 [Figure 2 about here.]

us
127 2.2. Illuminance

128 A similar approach is taken for the illuminance threshold, or the minimum

an
129 illuminance threshold, below which the control system switches the lights on,
130 in case of occupancy. Based on the premise that individual occupants un-
131 dertake actions consciously and consistently [18], we register the illuminance
M
132 of the room at each time the occupant manually switches the lights on. We
133 assume further that the occupant will only switch on the lights manually
d

134 when the illuminance falls below his desired level. Then, all samples of this
te

135 threshold are taken after a sufficiently long period of time to obtain a his-
136 togram. Since much fewer data samples are collected than in the case of TD,
p

137 a cumulative distribution function cannot be determined. Rather, we simply

ce

138 determine the minimum illuminance threshold as the lower of the median and
139 the mean of the gathered data. To prevent large values from unnecessarily
skewing the data to a larger set-point, we introduce a cut-off value above
Ac

140

141 which the illuminance values are discarded. This cut-off value is defined as
142 the illuminance that the lights in the room provide at night, and determined
143 for each room individually.

144 [Figure 3 about here.]

Page 7 of 40
145 Figure 3 shows sample illuminance levels when the occupant switched the
146 lights on. In this case, values above the cut-off of 439 lux were neglected in

t
147 the calculation of the mean and median, and the lower value of 143 lux (the

ip
148 median in this case) was taken to be the minimum illuminance threshold.

cr
149 A third set point, maximum illuminance threshold, is derived from the
150 minimum illuminance threshold, to evaluate if the illuminance in the room is

us
151 sufficiently high to switch the lights off. This maximum illuminance threshold
152 is determined as the simple addition of this cut-off value to the minimum

an
153 illuminance threshold value in each room.
154 In conclusion, by having a pair of set points for each room, the preferences
155 of the occupants can be well adapted to. The objective is then to decide
M
156 the optimal confidence level in order to maximize energy savings without
157 compromising the comfort of the occupant.
d
te

158 3. Hardware and controller implementation

159 The target office environment is equipped with a powerline communica-

p

160 tion (PLC) system based on digitalStrom (dS) technology [19]. Figure 4
ce

161 shows a typical room in the study. In each room, a push-button light switch
162 and basic occupancy sensors are available and connected through dS. For
Ac

163 our purposes, the basic dS system is augmented with a light sensor at about
164 10cm above desk level in order to log desk illuminance levels, and a main
165 control server with a database to perform the data analysis and execute con-
166 trol actions. By mounting the light sensor 10cm above the desk, we can avoid
167 that the sensor is accidentally covered by objects, such as paperwork, while
168 still gathering relevant illuminance data. Figure 5 shows the control system

Page 8 of 40
169 schematically.The following information is logged:

170 i Light level in each room (logged every minute).

t
ip
171 ii Every trigger of the occupancy sensor in each room.
172 iii Every action on the manual light switch in each room.

cr
173 iv On/Off status of the lights in each room(logged every minute).

us
174 [Figure 4 about here.]

175 [Figure 5 about here.]

176

177
an
With this information, the control software runs every minute to determine
if it is necessary to turn the lights on or off.
M
178 Note that no extra hardware or interaction feature is provided to the oc-
179 cupant, i.e., no special action is required from the occupant for the system
d

180 to work. The operation of the lighting is exactly as it was before the instal-
te

181 lation of the control system, with the exception of the added (unintrusive)
182 light sensor.
p

183 Table 1 shows the control table that summarizes the main actions of the
ce

184 control system. The four main control inputs are occupancy, current light
185 status and whether the room is determined as dark or bright. There are
basically only three cases when the control is activated:
Ac

186

187 [Table 1 about here.]

188 i When the room is unoccupied but the lights are on, the control will
189 proceed to turn off the lights.
190 ii When the room is occupied and determined to be dark but the lights are
191 not on, the control will proceed to turn on the lights.

Page 9 of 40
192 iii When the room is occupied, the lights are on, but the light level is de-
193 termined to be excessive (bright), the control will proceed to turn off the

t
194 lights.

ip
195 3.1. Example of control operation

cr
196 A successful operation of the control in one of the rooms (G25.2) on 28

us
197 May 2014 is described as follows and depicted visually in Fig. 6.
198 At about 08:30 (A), the occupant enters the room and control switched
on the lights as the light level was below the minimum illuminance threshold.

an
199

200 The light was turned off and on several times due to the occupant leaving and
201 entering the room. When the occupant reentered the room again at about
M
202 10:30 (B), the control did not switch on the lights as the light level was
203 above the maximum illuminance threshold. However, the occupant deemed
d

204 the room to be dark and turned on the lights manually at about 10:45 (C).
The control switched off the lights again when the occupant left the room at
te

205

206 around noon (D), and back on on his return afer lunch (E). Shortly after, the
p

207 light levels were deemed sufficient to switch off the lighting. The occupant
switched on the lights manually again at about 17:00 (F) with decreasing
ce

208

209 light level in the room, and finally the control switched off the lights at
210 about 19:00 (G) when the occupant left the office at the end of the workday.
Ac

211 [Figure 6 about here.]

212 4. Experimental Study

213 An experiment was conducted for a period of six weeks from 22 April to 4
214 June 2014 in ten different rooms of the author’s research group (see Fig. 7):

10

Page 10 of 40
215 [Figure 7 about here.]

• G22, Kitchenette • G26.1, Workshop

t
216 221

ip
217 • G23, Double-occupant Office 222 • G26.2, Double-occupant Office

cr
218 • G24, Multi-occupant Office 223 • G27, Conference Room

us
219 • G25.1, Single-occupant Office224 • G31.1, Double-occupant Office

220 • G25.2, Double-occupant Office

225 • G31.2, Printer Room

226

227 an
Three different control modes are defined and investigated as shown in
Tab. 2. The baseline mode (M0) has no active control other than after 19:00
M
228 when lights are switched off with TD=15min. This was implemented to
229 prevent excessive energy wastage in the case where occupants do not switch
d

230 off the lights when leaving the office at the end of the day. Modes M1 and M2
te

231 are used to compare the influence of TD. Initially, the difference between pT D
232 for M1 and M2 was more dramatic (pT D (M1) = 95%, and pT D (M2) = 70%),
p

233 which caused too many false triggers. Hence, it was decided to increase it to
ce

234 pT D (M2) = 90%. An arbitrary limit of 20min was set on the maximum TD
235 to ensure that the values are not artificially large.
Ac

236 To achieve a relatively fair comparison between the modes, and to average
237 out the effects of weather and office location, the control mode for each office
238 is determined randomly at midnight each day. In addition, the experiment
239 was conducted double-blind as to not to bias the results. All occupants were
240 informed to continue using their lighting as usual.

241 [Table 2 about here.]

11

Page 11 of 40
242 The TD values for all the rooms in M1 were initialised to 15 min and
243 M2 to 10 min. The minimum illuminance thresholds were initialised to 0 lux

t
244 and the maximum illuminance thresholds were initialised by recording the

ip
245 light levels in each room with lights on at night (i.e. no natural light). The

cr
246 controller updates TD once a day based on the occupants behaviour in each
247 room. The illuminance thresholds values are calculated every hour for the

us
248 first two weeks and every 4 hours thereafter to ensure quick adaptation.

5. Results and Discussion

an
249

250 5.1. Occupant specific set-points

The minimum illuminance threshold and TD values for each room at the
M
251

252 end of the experiment are shown in Fig. 8.

253 [Figure 8 about here.]

d

It is evident that each room has arrived at unique set-points. It can be

te

254

255 observed that in both modes, M1 and M2, the determined TD values are
p

256 well below the industry standard of 12min, except for the special rooms G22
(kitchen) and G31.2 (printer room), which will be discussed below. Further,
ce

257

258 minimum illuminance threshold values are also below the typical light level
259 recommendations for an office, i.e., 400 lux [11]. This shows that all the occu-
Ac

260 pants in each of the rooms actually require less light than the recommended
261 value and hence would consume less energy.
262 From this, we can confirm previous findings in this research field, i.e.,
263 that there is a significant potential in reducing energy consumption by im-
264 plementing an adaptive control system. We now present the actual energy
265 savings.

12

Page 12 of 40
266 5.2. Energy Savings

267 The energy consumption is estimated from the power ratings and the

t
total time during which the lights were switched on per day. Each office

ip
268

269 is equipped with 10 light bulbs of 20W each. Thus, one hour of operation

cr
270 results in 20Wh electricity consumption. The savings of M1 and M2 are then
271 calculated by dividing the energy consumption of each mode by the energy

us
272 consumption of M0. Figures 9 and 10 show the energy consumption (in lines)
273 and the percentage energy savings relative to mode M0 (in bars) in rooms

an
274 G25.2 and G31.2, respectively. The values shown are calculated at the end
275 of each week and are cumulative.
M
276 [Figure 9 about here.]

277 [Figure 10 about here.]

d
te

278 The results observed in G25.2, a double-occupant office, were as expected.

279 It can be seen that the system took approximately 3 weeks to adapt to
p

280 the occupant before displaying energy savings in M1. The final percentage
ce

281 savings obtained at the end of the experiment in M1 and M2 are 34% and
282 72%, respectively.
The results observed in G31.2, the printer room, were contrary to expec-
Ac

283

284 tation. M1 and M2 consumed more energy than the baseline mode M0. This
285 can be attributed to the fact that G31.2 is used as a common facility room.
286 The normal behaviour exhibited in this room is that occupants enter the
287 room to collect their prints without switching on the lights. However, with
288 the control system, the lights are turned on for the occupants automatically,

13

Page 13 of 40
289 resulting in higher energy consumption. Thus, our control approach is un-
290 suitable for this type of room. For this reason, in the following the results for

t
291 the three utility rooms, namely the kitchen G22, the conference room G27

ip
292 and the printer room G31.2 were neglected for the calculation of the overall

cr
293 energy consumption.
294 Figure 11 shows the cumulative overall energy consumption and savings

us
295 of the control system, aggregated and averaged over all the rooms excluding
296 G22, G27 and G31.2. In general, it shows that the control system took

an
297 about 1 week to adapt to all occupants across all rooms. Energy savings
298 were evident after only two weeks of operation. The final percentage savings
299 obtained at the end of the experiment in M1 and M2 are 23.2% and 37.9%
M
300 respectively.
d
301 [Figure 11 about here.]
te

302 It can be observed that M2 achieved almost twice as much energy savings
303 than M1. This was achieved without much occupant discomfort, judged
p

304 qualitatively by the lack of complaints from occupants. However, another

ce

305 study will have to be conducted to ascertain the actual effects on the comfort
306 of occupants, and determine the relationship between pT D and the savings
Ac

307 made. In our setup, a difference of 5% in pT D roughly equates a difference

308 in savings of 20%.
309 Figure 12 shows the average overall energy consumption in absolute num-
310 bers of the operation in each control mode. The baseline consumption was
311 690 Wh/day per room and the lowest consumption obtained in M2 was 428
312 Wh/day per room. The achieved savings compared to our own baseline are

14

Page 14 of 40
313 23.2% and 37.9% in M1 and M2, respectively. For a comparison with other
314 published results, we compare our results to a hypothetical all-day, 8hrs lights

t
315 on baseline with a consumption of 1600 Wh/day. Then, we achieve energy

ip
316 savings of 66.8% and 73.2% in mode M1 and M2, respectively.

cr
317 [Figure 12 about here.]

us
318 5.3. Adaptation of Time Delay

319 Figures 13 and 14 show the evolution of the time delay values in each room

an
320 for modes M1 and M2, respectively. It is interesting to note that the TD
321 values stabilised after just a week of operation, with only minor changes of
+/-1 min in the subsequent weeks. This shows that the occupant’s mobility
M
322

323 in each room is rather consistent and does not change much with time. It also
324 suggests that the control system is capable of readapting itself to potential
d

325 changes in occupancy.

te

326 [Figure 13 about here.]

p

327 [Figure 14 about here.]

ce

328 As expected the TD values for M2 are generally lower than for M1. The
329 sudden increase in TD values at the start of the third week (black dashed
Ac

330 line) is due to the change in PT D , from 70% to 90%. This was due to TD
331 values being too small and causing unwanted false triggers of the lights.
332 Another interesting observation from the two figures is that the two high-
333 est TD values correspond to the kitchen, G22 and the printer room, G31.2.
334 This can be attributed to the fact that these rooms have a rather irregular
335 occupancy, and hence, the method we used to determine TD is not suitable.

15

Page 15 of 40
336 5.4. Adaptation of Minimum Illuminance Threshold

337 The adaptation of the illuminance thresholds differs is shown in Figure 15.

t
It can be seen that while in some rooms the values stabilized (G25.1, G26.1),

ip
338

339 in others there is still a trend developing. This may be due to the fact that

cr
340 the data that has been collected is insufficient to converge to a stable result.
341 On the other hand, this also shows the dynamic nature of adaptive set-

us
342 points. Since the threshold is defined based on interaction data from the
343 past four weeks, convergence to a fixed-set point may not be achieved or even

an
344 desired. Rather, the evolution of the set-point reflects the latest preference
345 settings of the occupant. In addition, if changes in occupancy occur, such a
system can easily reprogram itself to adapt to a new situation.
M
346

347 [Figure 15 about here.]

d

348 6. Discussion and Conclusion

te

349 An automatic control system can reduce the need for the occupant to
p

350 manually operate the lights if configured correctly, but it can also cause
ce

351 annoyance to the occupant by not operating as intended. To balance these

352 conflicting goals for each occupant individually, in this paper, we presented an
Ac

353 occupant centered lighting control strategy, and its successful implementation
354 in a case study. For 10 offices, occupant-specific set-points for time-delay,
355 minimum and maximum illuminance thresholds were determined based on
356 interaction with the user, and statistical data analysis. The results of the 6
357 weeks case study show that there is a large savings potential as the derived-set
358 points are both individual to each office, and well below standard settings.

16

Page 16 of 40
359 A double-blind occupany study was conducted and the system performed
360 as intended, adapting well to the occupants as well as achieving significant

t
361 energy savings. The best result achieved is 37.9% energy savings in M2

ip
362 compared to our own baseline mode, and 73.2% compared to a worst-case, all-

cr
363 day lights on mode. We cannot exclude any bias effects from the occupants,
364 e.g., waiting for the lights to be switched on or off automatically. However, we

us
365 assume that positive and negative effects average out over the experimental
366 time and across the offices.

an
367 In addition, it has been shown that the dynamic nature of our proposed
368 statistical determination of set-points results in time-variant set-points. This
369 allows for adaptation in case of, e.g., occupancy changes. The effects of time
M
370 variant set-points on building operation and occupant comfort need to be
371 further studied. From a long-term study, the desired set-points could be cor-
d

372 related with external lighting conditions to investigate seasonal variations,

te

373 which could lead to an adaptive lighting comfort model, similar to the adap-
374 tive thermal model [20].
p

375 One potential drawback of our method is that when the function of the
ce

376 room is not one of normal occupancy, e.g., the kitchen, conference and printer
377 rooms, the occupant centered control may lead to higher energy consumption.
Ac

378 For these cases, simple visual prompts to remind occupants to switch off
379 the lights while leaving the rooms is a viable alternative to ensure energy
380 savings [21]. In addition, we have only qualitatively evaluated the comfort
381 of the occupants by the lack of complaints. A proper post-occupancy study
382 in a long-term study, e.g., a living lab environment, has to be performed to
383 quantify the comfort gains.

17

Page 17 of 40
384 Finally, the study in this paper was limited to lighting control due to
385 hardware constraints. Of course the concept extends to all types of HVAC

t
386 systems as long as interaction with the occupant is allowed. In the case of

ip
387 heating, the comparatively low response types needs to be considered. All in

cr
388 all, we can conclude that occupant centered control holds a great potential
389 to ensure energy efficient buildings and satisfied occupants due to its rather

us
390 simple implementation and interactive nature.

Acknowledgement

an
391

392 The support through the Building Technology Accelerator (BTA) Pro-
M
393 gram of the Climate-KIC initiative of the European Institute of Innovation
394 & Technology (EIT) is gratefully acknowledged.
d

395 References
te

396 [1] International Energy Agency, Technology Roadmap: Energy Efficient

Building Envelopes (2013).
p

397
ce

398 [2] Schweizerische Eidgenossenschaft Bundesamt für Energie, Analyse des

399 schweizerischen energieverbrauchs 2000-2012 nach verwendungszwecken
(2013).
Ac

400

401 [3] Schweizerische Eidgenossenschaft Bundesamt für Energie, Energiever-

402 brauch in der industrie und im dienstleistungssektor (2013).

403 [4] M. S. Rea, The IESNA Lighting Handbook: Reference & Application,
404 Illuminating Engineering Society of North America, 2000.

18

Page 18 of 40
405 [5] B. Von Neida, D. Manicria, A. Tweed, An analysis of the energy and cost
406 savings potential of occupancy sensors for commercial lighting systems,

t
407 Journal of the Illuminating Engineering Society 30 (2) (2001) 111–125.

ip
408 [6] C. F. Reinhart, Lightswitch-2002: a model for manual and automated

cr
409 control of electric lighting and blinds, Solar Energy 77 (1) (2004) 15 –
410 28.

us
411 [7] J. S. Han, Y. K. Jeong, I. W. Lee, Analysis of electric energy con-

an
412 sumption patterns: A case study of a real life office building, Applied
413 Mechanics and Materials 330 (2013) 158–162.
M
414 [8] J. D. Jennings, F. M. Rubinstein, D. DiBartolomeo, S. L. Blanc, Com-
415 parison of control options in private offices in an advanced lighting con-
416 trols testbed, Journal of the Illuminating Engineering Society 29 (2)
d

417 (2000) 39–60.

te

418 [9] S. Pigg, M. Eilers, J. Reed, Behavioural aspects of lighting and oc-
p

419 cupancy sensors in private offices: A case study of a university office

ce

420 building, in: Proceedings of the 1996 ACEEE summer study on energy
421 efficiency in buildings, Vol. 8, 1996, p. 8.
Ac

422 [10] M. Sunikka-Blank, R. Galvin, Introducing the prebound effect: the gap
423 between performance and actual energy consumption, Building Research
424 and Information 40 (2012) 260–273.

425 [11] R. Hughes, S. Dhannu, Substantial energy savings through adaptive

426 lighting, in: Electric Power Conference, 2008. EPEC 2008. IEEE
427 Canada, 2008, pp. 1–4.

19

Page 19 of 40
428 [12] M. Chiogna, A. Mahdavi, R. Albatici, A. Frattari, Energy efficiency of
429 alternative lighting control systems, Lighting Research and Technology

t
430 44 (4) (2012) 397–415.

ip
431 [13] V. Garg, N. Bansal, Smart occupancy sensors to reduce energy con-

cr
432 sumption, Energy and Buildings 32 (1) (2000) 81 – 87.

us
433 [14] T. Leephakpreeda, Adaptive occupancy-based lighting control via grey
434 prediction, Building and Environment 40 (7) (2005) 881 – 886.

an
435 [15] D. Bourgeois, C. Reinhart, I. Macdonald, Adding advanced behavioural
436 models in whole building energy simulation: A study on the total energy
impact of manual and automated lighting control, Energy and Buildings
M
437

438 38 (7) (2006) 814 – 823, special Issue on Daylighting Buildings.

[16] A. Guillemin, N. Morel, An innovative lighting controller integrated in

d
439

440 a self-adaptive building control system, Energy and Buildings 33 (5)

te

441 (2001) 477–487.

p

442 [17] A. Schaeper, C. Palazuelos, D. Denteneer, O. Garcia-Morchon, Intelli-

ce

443 gent lighting control using sensor networks, in: Networking, Sensing and
444 Control (ICNSC), 2013 10th IEEE International Conference on, 2013,
pp. 170–175.
Ac

445

446 [18] H. B. Gunay, W. O’Brien, I. Beausoleil-Morrison, A critical review of

447 observation studies, modeling, and simulation of adaptive occupant be-
448 haviors in offices, Building and Environment 70 (0) (2013) 31 – 47.

449 [19] G. Dickmann, Digitalstrom: A centralized plc topology for home au-
450 tomation and energy management, in: Power Line Communications

20

Page 20 of 40
451 and Its Applications (ISPLC), 2011 IEEE International Symposium on,
452 2011, pp. 352–357.

t
ip
453 [20] F. Nicol, M. Humphreys, S. Roaf, Adaptive thermal comfort: Principles
454 and practice, Taylor & Francis, 2012.

cr
455 [21] R. M. Tetlow, C. P. Beaman, A. A. Elmualim, K. Couling, Simple

us
456 prompts reduce inadvertent energy consumption from lighting in office
457 buildings, Building and Environment 81 (0) (2014) 234 – 242.

an
M
d
p te
ce
Ac

21

Page 21 of 40
 t
ip
cr
us
an
M
d
te

Figure 1: Sample of Motion Sensor Signal

p
ce
Ac

22

Page 22 of 40
 t
ip
cr
us
an
M
d
p te

Figure 2: Determining TD Value from Cumulative Distribution Function

ce
Ac

23

Page 23 of 40
 t
ip
cr
us
an
M
d
pte

Figure 3: Determining Minimum Lux Threshold

ce
Ac

24

Page 24 of 40
 t
ip
Light

cr
PIR Motion
Sensor

us
an
M
Push
Button
d
te

Light
Sensor
p
ce
Ac

Figure 4: Typical room setup for experimental study

25

Page 25 of 40
 t
ip
cr
us
an
M
d
te
p
ce

Figure 5: Overview of Control System

Ac

26

Page 26 of 40
 t
ip
max lux A B C D E F G
threshold 1000

cr
a)
500

min lux

us
threshold 0
06 09 12 15 18 21

Occupied
b)

an
Empty

06 09 12 15 18 21
M
c) Push- ON
button
events OFF

06 09 12 15 18 21
d
te

d) State ON
of lights
OFF
p

06 09 12 15 18 21
ce

ON
e) Control
Events OFF
06 09 12 15 18 21
Ac

Hour of Day

Figure 6: Operation of Control System in Room G25.2 on 28 May 2014. a) Light levels
with min and max illuminance thresholds, b) Occupancy, c) Interaction with user through
push-button, d) State of lights, e) Control events. See text for explanation

27

Page 27 of 40
 t
ip
cr
us
an
M
d
te
p
ce

Figure 7: Floor Plan of Experiment

Ac

28

Page 28 of 40
 t
ip
cr
us
an
M
d
p te
ce

Figure 8: Illuminance threshold and TD Values on 4 June 2014

Ac

29

Page 29 of 40
 t
ip
cr
us
an
M
d
p te
ce

Figure 9: Energy consumption (line plot) and savings (bar plot) in G25.2
Ac

30

Page 30 of 40
 t
ip
cr
us
an
M
d
p te
ce

Figure 10: Energy consumption (line plot) and savings (bar plot) in G31.2
Ac

31

Page 31 of 40
 t
ip
cr
us
an
M
d
p te
ce

Figure 11: Overall energy consumption (line plot) and savings (bar plot)
Ac

32

Page 32 of 40
 t
ip
cr
us
an
M
d
te

Figure 12: Overall absolute energy consumption and savings for each mode
p
ce
Ac

33

Page 33 of 40
 t
ip
cr
us
an
M
d
te
p
ce

Figure 13: Evolution of TD in mode M1

Ac

34

Page 34 of 40
 t
ip
cr
us
an
M
d
te
p
ce

Figure 14: Evolution of TD in mode M2

Ac

35

Page 35 of 40
 t
ip
cr
us
an
M
d
pte
ce

Figure 15: Evolution of illuminance thresholds

Ac

36

Page 36 of 40
 t
ip
cr
us
an
M
d
te
p
ce

Table 1: Control Table

Ac

37

Page 37 of 40
 t
ip
cr
us
an
M
Mode: M0 M1 M2
Type: Baseline Comfort Mode Savings Mode
Desc: After 19:00, TD=15min pT D = 95% pT D = 90%
d

Table 2: Control Modes

p te
ce
Ac

38

Page 38 of 40
*Highlights (for review)

 Simple setup using only two sensors


 Experimental study in real world environment with multiple
participants

 Presentation of set-point adaptation curves

t
ip
 Discussion of the applicability for different room types

cr
us
an
M
d
p te
ce
Ac

Page 39 of 40
Graphical Abstract (for review)

an
M
ed
pt
ce
Ac

Page 40 of 40