Validating the Optimization of a Building Occupancy Monitoring

Software System

Jalil Boudjadar and Simon Thrane Hansen

Department of Electrical and Computer Engineering, Aarhus University, Denmark

Keywords:

Buildings Occupancy Analysis, State Estimation, Monitoring Software, Formal Veriﬁcation.

Abstract:

Real-time occupancy information is key to track and optimize the resource usage in buildings. With the ulti-

mate goal to reduce energy consumption and improve indoor environment quality, different occupancy mon-

itoring solutions have been introduced in the literature. However, these solutions can be part of the original

problem due to its high energy consumption. This paper proposes the design and validation of a knowledge-

driven energy-efﬁcient system to monitor and analyze buildings occupancy. Such a system is in fact an op-

timization of the actual software solution (Ofﬁce Master 3000) used to monitor buildings occupancy in an

academic institution. The key contributions are: 1) Optimization of the actual solution: using the knowledge

about the activity expected to take place, our system proactively identiﬁes the minimal sensors data relevant

to the actual state to conﬁrm/deny such an activity by which non-substantial sensors are operated with loos-

ened frequency; 2) Formal validation: using UPPAAL model checker we prove that the original functional

and non-functional properties are maintained post-optimization. The proposed system is implemented in C++,

tested and validated. The results demonstrate that our optimization reduces sensors energy consumption by up

to 31% while maintaining high accuracy (84%) in identifying the occupancy states and activities.

1 INTRODUCTION

With the ever increasing buildings energy demand and

cost, it is paramount to monitor and optimize the en-

ergy consumption of the different loads and comput-

ing components embedded in buildings (Lasla et al.,

2019; Boudjadar et al., 2014). To conduct such an op-

timization, a highly reliable tracking of the actual oc-

cupancy state of building spaces is needed. The build-

ing state refers to the activities, events and potentially

the number of people occupying a building space at

a given time point. Building occupancy monitoring

systems (BOMS) enable to track the state of build-

ing spaces in real-time (Azimi and O’Brien, 2022;

Elkhoukhi et al., 2018; Shokrollahi et al., 2024; Baer-

entzen et al., 2023; Tien et al., 2022). The knowledge

generated by BOMS provides a solid and informative

ground for buildings management to identify when

and where resource utilization can be reduced, so that

to achieve a better space allocation and efﬁcient re-

sources usage (Salimi and Hammad, 2019; Jabirullah

et al., 2021).

Commonly, state of the art BOMS rely on Big

Data analysis where massive quantities of sensor data

(video, audio, step panels, etc.) need to be collected

and run through classiﬁcation, prediction and opti-

mization engines (Zhang et al., 2019) (Jiang and Yin,

2015). It is expensive to maintain the real-time aspect

of such a brute-force approach to occupancy analysis

(Elkhoukhi et al., 2018), due to massive data process-

ing, and can result in relatively low accuracy, particu-

larly if the input data blinds out the data related to the

cases with a strong potential to optimization (Salimi

and Hammad, 2020; Pan et al., 2014). Moreover, it

can end up with high energy consumption due to the

cost of collecting and processing massive data (Coste-

naro and Duer, 2012).

Different attempts to optimize the energy con-

sumption of BOMS and improve its decision mak-

ing have been proposed in the literature (Lou et al.,

2017; Rai et al., 2015; McKenna et al., 2015; Dai

et al., 2020; Ahmad et al., 2021; Zhang et al., 2022;

Jiang et al., 2022). However, different challenges

have emerged: 1) it is not trivial to design high ﬁ-

delity models to enable systematic state computation

(Austin et al., 2020); 2) an automated semantics-

driven occupancy analysis delivering high conﬁdence

decisions in real-time is expensive computation-wise;

3) optimizing the performance and energy efﬁciency

of a BOMS can lead to violate safety properties and

degrade reliability metrics given that such metrics

can be conﬂicting with the performance (Teng et al.,

2021).

This paper proposes the optimization of a state of

Boudjadar, J. and Hansen, S. T.

Validating the Optimization of a Building Occupancy Monitoring Software System.

DOI: 10.5220/0013559400003964

In Proceedings of the 20th International Conference on Software Technologies (ICSOFT 2025), pages 155-166

ISBN: 978-989-758-757-3; ISSN: 2184-2833

155

the art BOMS, Ofﬁce Master 3000, to reduce the data

samples required and the underlying energy consump-

tion, for both data gathering and processing, while

maintaining high accuracy for the occupancy state es-

timation. The resulting optimized BOMS relies on

an intelligent knowledge-driven occupancy estima-

tion so that it proactively identiﬁes which high-level

symbolic facts are needed to change the current state

into an actual event. These facts are run through a se-

mantically well-deﬁned occupancy model to identify

which minimal combination of non-intrusive sensors

can supply data to conﬁrm or contradict these facts,

rather than blindly collect massive data from all sen-

sors post-hoc to recognize ongoing activities. This en-

ables to reduce the sensors frequency, and potentially

switching sensors off for considerable time durations.

To validate the proposed optimization, we have

modeled the optimized BOM in UPPAAL and used

model checking to verify that the original non-

functional properties (correctness, response time, per-

formance and energy consumption) of the Ofﬁce Mas-

ter BOMS are still maintained post-optimization. Fur-

thermore, the optimized BOMS is implemented in

C++ where we analyzed the underlying performance

and energy, and compare it to the original Ofﬁce Mas-

ter BOMS and to other state of the art BOMS algo-

rithms.

The rest of the paper is structured as follows: Sec-

tion 2 presents a background about BOMS and UP-

PAAL. Section 3 cites relevant related work. Sec-

tion 4 presents the mathematical modeling and opti-

mization of the proposed BOMS. Section 5 describes

the speciﬁcation and veriﬁcation using UPPAAL. Im-

plementation analysis and validation results are dis-

cussed in Section 6. Finally, Section 7 concludes the

paper.

2 BACKGROUND

This section presents the background related to occu-

pancy analysis, optimization and UPPAAL environ-

ment for formal veriﬁcation and validation.

2.1 Occupancy State Estimation

Occupancy monitoring amounts to observe a speciﬁc

area within a building to detect human presence and

recognize occupancy activities (Ahmad et al., 2021).

This typically entails gathering data from sensors that

capture various features associated with human pres-

ence, such as breathing, movement and noise. The

facts obtained from the sensors are then combined

through an inference process to estimate the occu-

pancy state (Yang et al., 2016).

For building management systems (BOMS) to ef-

fectively utilize this knowledge, real-time data is es-

sential. This allows for timely actions to be taken

based on the current occupancy state, thereby mini-

mizing resource wastage and enhancing indoor com-

fort (Seghezzi et al., 2021; Sun et al., 2020).

With recent advancements enabling the deploy-

ment of BOMS on embedded devices using battery-

powered sensors, optimizing the computational costs

and energy consumption has become crucial. While

reducing data sampling can serve as a means of op-

timization, it may compromise the accuracy of oc-

cupancy estimation (Yang et al., 2016). Hence, it

is imperative to conduct post-optimization validation

to ensure BOMS systems consistency and reliability

(Boudjadar and Tomko, 2022).

2.2 BOMS Optimization

A BOMS optimization is the process of enhancing

different performance metrics such as response time,

resource utilization, and energy consumption (Mi-

noli et al., 2017). The optimization can be achieved

through a variety of techniques, including heuris-

tics, mathematical solvers, and machine learning al-

gorithms. However, it is crucial to ensure that improv-

ing one performance metric does not signiﬁcantly de-

grade others or violate functional and non-functional

properties (Boudjadar and Khooban, 2020; Boudjadar

and Tomko, 2022). In the context of buildings, the

need for optimization arises from their multifunc-

tional nature, where individual spaces may serve dif-

ferent purposes. Therefore, tailoring a building space

to its current activity is essential. For BOMS, op-

timization aims to reduce energy consumption and

computational costs associated with data gathering

and processing. The key objective of our optimiza-

tion is to achieve real-time knowledge of the occu-

pancy state, enabling dynamic adjustment of sensor

frequencies based on the required data to conﬁrm or

deny an activity (Rault et al., 2014).

2.3 UPPAAL Model Checking

Model checking (E. M. Clarke et al., 1999; Baier

and Katoen, 2008) is a widely used automated ver-

iﬁcation technique for examining complex reactive

systems, including hardware components, embed-

ded controllers, and network protocols. The method

works by expressing the speciﬁcations of a system

behavior in a semantically well-deﬁned language and

formalize the properties to look for in logical formu-

ICSOFT 2025 - 20th International Conference on Software Technologies

156

las such as LTL or CTL (Baier and Katoen, 2008),

and then check whether the system behavior satisﬁes

the expected qualities. We use UPPAAL (Boudjadar

et al., 2013) for the post-optimization validation due

to its expressiveness and ability to verify real-time

systems.

UPPAAL is based on Timed Automata the-

ory (Alur and Dill, 1994). As such, each UPPAAL

template (timed automaton) is characterized by the

following:

• A ﬁnite set of locations, with one designated as

the initial location;

• A ﬁnite set of transitions connecting locations;

• A ﬁnite set of variables;

• A ﬁnite set of actions performed on the variables

when executing a transition;

• A ﬁnite set of predicates, called guards allowing

the execution of a transition:

• A ﬁnite set of synchronization channels.

A sequence of transitions represents the evolution of

the speciﬁcation. Each transition may be labeled with

one or more actions, a guard and a synchronization.

The state of a speciﬁcation is given by the current lo-

cation and the actual valuation of the variables. A

speciﬁcation in UPPAAL relies on combining multi-

ple templates (network of timed automata) to create

complex and modular systems. The composition can

be synchronous where two transitions, one from each

timed automaton, labeled with the same synchroniza-

tion channel can be executed simultaneously.

3 RELATED WORK

Buildings occupancy monitoring has been thoroughly

explored in the recent years where different monitor-

ing models, analysis processes and optimizations are

proposed (Lou et al., 2017; Rai et al., 2015; McKenna

et al., 2015; Dai et al., 2020; Ahmad et al., 2021;

Zhang et al., 2022; Jiang et al., 2022). The ultimate

goal is to deliver highly accurate occupancy state esti-

mation while optimizing the resources to deploy in or-

der to achieve that mission such as energy consump-

tion, accuracy and response time (Lasla et al., 2019;

Trivedi and Badarla, 2020).

Lasla et al (Lasla et al., 2019) proposed an energy-

efﬁcient monitoring system for buildings using in-

frared sensors. The energy optimization comes from

the optimal placing of sensors identiﬁed through a

design exploration formulated as mixed integer lin-

ear problem. However, according to the authors, the

accuracy of these experiments is dependent on the

amount of data to be collected. The authors of (Nien-

aber et al., 2020) proposed the validation and tuning

of different occupancy detection algorithms for multi-

person ofﬁces where different abstraction models are

considered, and the underlying impact of the decision

accuracy is analyzed. Improvements of the algorithms

have been conducted to achieve a trade-off between

the granularity of the occupancy model, the assump-

tions to make and the decision accuracy. However,

relying only on the CO

emission sensors has been

proven to be less accurate as it depends on the physi-

cal state of the occupants. The authors of (Lou et al.,

2017; Rai et al., 2015) proposed an agent-based oc-

cupancy analysis, that amounts to track the individual

occupants rather than measuring the impact made by

occupants on their surroundings. Although this alter-

native can enable accurate occupancy of the individu-

als, it can suffer from the scalability and the compu-

tation cost.

A state-based occupancy model has been pro-

posed in (McKenna et al., 2015). It consists in creat-

ing stochastic data in terms of how probable an occu-

pancy state will occur. This process enables to infer

the location of occupants and their activities. How-

ever, the bottleneck can be related to the massive data

gathering which drains sensor batteries and computa-

tion resources. Similarly, the authors of (Jiang et al.,

2022) proposed a semantic (trust-based) model for oc-

cupancy analysis where the sensors triggering order

and data are fed to a sequence matching scheme to

estimate the occupancy state. The sequence matching

scheme relies on known human activity sequences,

which may lead to uncertainty when partial states or

unknown scenarios are encountered.

Machine learning-based occupancy monitoring

has also been explored. The authors of (Tien et al.,

2022) proposed an occupancy monitoring model us-

ing convolutional neural networks and cameras to

sample the building spaces. Although this approach

achieves high accuracy, it requires massive data and

may suffer from computation cost and privacy con-

cerns.

Alternatively, a knowledge-based occupancy anal-

ysis has been proposed in (Ahmad et al., 2021). It

consists in designing a domain-speciﬁc model encod-

ing the activities to be carried out in the building and

run the actual data through such a model to identify

the occupancy state. In (Cala et al., 2015), the au-

thors presented the evaluation and validation of an oc-

cupancy detection system based on the concentration

of CO

in the indoor air. The validation amounted

to check the similarity between the proposed BOMS

outputs and the actual occupancy state reported by the

occupants.

Validating the Optimization of a Building Occupancy Monitoring Software System

157

In this paper, we propose a knowledge-driven

proactive BOMS system to analyze the actual occu-

pancy state and identify the minimal data to collect in

order to infer the ongoing activity in real-time, as an

optimization of the actual Ofﬁce Master 3000 BOMS.

We analyze the proposed BOMS post-optimization to

examine the performance improvement and conﬁrm

that the original functional and non-functional prop-

erties are maintained.

4 OPTIMIZED OCCUPANCY

MONITORING

This section presents the proposed occupancy analy-

sis model and optimization.

4.1 Occupancy State Speciﬁcation

The original BOMS we consider monitor building

spaces (classrooms, meeting rooms, etc) using CO

sensors, motion sensors, noise sensors and light sen-

sors. The sensors are installed in different room ends,

where each location contains a sensor from each cate-

gory. The non-intrusive sensors are quantitative: light

sensors in lux, CO

in parts per million (ppm), audio

in decibels (dB), and motion in four levels: none (0),

low (1), medium (2), and active (3).

Following the building functionality and sensors,

we deﬁne the following language grammar L allow-

ing to describe the different occupancy activities.

L ≜ E

|...|E

E ≜ (F,c)|(F,c) ∧ E

F ≜ S|S ∧ F

S ≜ v ∈ [a,b]

c ≜ H|M|L

The language speciﬁes each potential occupancy

event E through a set of features F, each of which

is recognizable through a subset of sensors S if the

value v of each sensor is within certain range [a,b].

The conﬁdence level c states how much a feature is

able to identify an event and can have values high (H),

medium (M) or low (L). One can see that different

events can have different features, which in turn can

have different numbers of sensors associated.

We deﬁne a semantics function ||e|| : E → P (S )×

P (c) to infer for each event the sensors and conﬁ-

dence levels from the event features, according to L .

E is the set of events, whereas S represents the set of

sensors and P () is the power set (set of all subsets).

Thanks to language L describing all potential events

and features in a structured manner, we will be able to

automate a thorough analysis of the occupancy state

in real-time. This in fact will be done by projecting

the sensor readings to identify the features present at

the given time point. The features in turn enable to

recognize the candidate events. An example of a fea-

ture is ”one or few people present in the front of class-

room”. This feature will be identiﬁed if the values of

sensor S

and S

are within these respective ranges:

∈ [410,700ppm] ∧ v

∈ [2,3]. Accordingly, we

specify the lecture event E

as follows:

≜

⟨

⟩

∧

⟨

⟩

∧

⟨

⟩

∧

⟨

⟩

∧

⟨

⟩

∧

⟨

⟩

The conﬁdence level indicates how critical a sen-

sor can contribute in recognizing an event. Nomi-

nally, the sensors of features having low conﬁdence

for a given event e can be discarded while e is being

conﬁrmed through high conﬁdence features. The cal-

culation and optimization of sensors frequency will

be deﬁned in the next section.

We deﬁne a valuation function V[S

](t) to return

the value of a given sensor S

at time point t. A state

of the proposed BOMS s at time t is given by the

readings (V[S

](t)) from each of the sensors sampled

at time t.

s(t) = ⟨V [S

](t),V [S

](t), . . . ,V [S

](t)⟩

For the sake of notation, we use s to refer to the actual

state, i.e. having the latest sensor readings. V returns

the value of any sensor for any time point.

4.2 Knowledge-Driven Optimized

Occupancy Analysis

The proposed BOMS optimization amounts to infer

the sensors to use in order to recognize the expected

event ε.The expected event can initially be obtained

from the schedule of the monitored space. Alterna-

tively, it can be set through a brute force data sampling

by activating all sensors with the highest frequency,

then the event recognized will be set to be the ex-

pected one. The optimization algorithm changes the

expected event dynamically at runtime following the

features recognized through occupancy analysis, as

one can see in Algorithm 1. The sensors are planned

on-the-ﬂy where a sliding of the frequencies is cal-

culated and applied following the expected event and

the actual occupancy state.

Sensing Plan. A sensing plan R speciﬁes the sen-

sors and the frequency for each to sample oc-

cupancy data. Namely, R is given by R =

⟨(S

),(S

),...⟩, where S

are sensors

ICSOFT 2025 - 20th International Conference on Software Technologies

158

and r

are frequencies. For the sake of notation, we

omit to specify time points as part of the sensing plans

given that the knowledge expected for the occupancy

state deﬁnes what sensors to operate and how often

rather than the actual values of sensors. We simply

write R to refer to the actual sensing plan.

Given that a sensor can be utilized to recognize

different features of a given event, such a sensor runs

the same (highest) frequency for the different features

of the event but with respective reading ranges for

each of the features.

Our occupancy analysis is carried out by combin-

ing the sensor readings and comparing the inferred

occupancy state to the features of the expected event,

or any other event of the language L, in case the ex-

pected event cannot be approved. This results in 3

different cases: 1) good matching; 2) no matching; 3)

partial matching.

Good Matching. This corresponds to the case when

the actual occupancy state (sensor readings) matches

a single event e. This results in updating ε to e, where

a sensing plan is computed as follows:

R = ⟨(S

),..,(S

) | ∀i ∃c

) ∈ ||e|| ∧



2 if c

= H

4 if c

= M

8 if c

= L

⟩

For the frequency calculation of each sensor, we

associate frequencies of 2, 4 and 8 minutes respec-

tively for sensors having high (H), medium (M) and

low (L) conﬁdence levels respectively. Those fre-

quencies are analyzed empirically and proven to be

sufﬁcient to capture the occupancy features using the

underlying conﬁdence levels. Hence, an event be-

ing conﬁrmed through few consecutive sampling it-

erations leads to an optimization of the sensing plan

where sensors with the lowest conﬁdence are dis-

carded from the sensing plan as follows:

R = R/{(S

) | (S

) ∈ ||e||∧ r

= 8}

Discarding a sensor from the sensing plan

amounts to operating the sensor with the lowest fre-

quency or switching off the sensor if it has already

been operating at the lowest frequency for many con-

secutive iterations. Further elaboration on the opti-

mization of sensing plans is provided in Section 4.3.

No Matching. No-match situation emerges when

the outcomes from the actual sensing plan does not

match any event of the language L . To resolve such

a problem, the sensing plan is expanded to include all

sensors with high frequency (2 minutes) to be able to

get a starting point so that either a good matching or

a partial matching of an event occurs.

R = ⟨(S

,2),(S

,2),...,(S

,2)⟩

In fact, the occurrence of a no-match case can be

interpreted as a rejection or disapproval of the ex-

pected event ε.

Partial Matching. This case occurs when the actual

occupancy state partially matches one or more events.

To calibrate the sensing plan efﬁciently, we need to

distinguish how many events partially match the ac-

tual state.

In case of a partial matching of the actual state to

a single event e, we compute a new sensing plan as an

update of the actual sensing plan R where all the sen-

sors belonging to the features with high conﬁdence in

event e are included. Such sensors will run with high-

est frequency since they are given high conﬁdence H.

The new sensing plan is calculated as follows:

R = R ∪ {(S

)|(S

) ∈ ||e||∧ r

= 2}

In case this is insufﬁcient to conclusively con-

ﬁrm the event that partially matches the actual state,

the sensing plan will undergo an upgrade by incor-

porating sensors associated with features that exhibit

medium conﬁdence in event e. These sensors will

operate at a medium frequency of 4 minutes, corre-

sponding to their assigned medium conﬁdence level

R = R ∪ {(S

)|(S

) ∈ ||e||∧ r

= 4}

In case the partial matching persists, the sensing

plan will be expanded further to include all the sensor,

being Off, belonging to the features with low conﬁ-

dence in event e. This will lead to either conﬁrm the

actual event e (good matching case), or disapprove it

(no-match case). Accordingly, the update of the sens-

ing plan R is made as follows:

R = R ∪ {(S

)|(S

) ∈ ||e||∧ r

= 8}

When facing partial matches to multiple events, it

becomes crucial to pinpoint the ongoing event with-

out engaging all sensors. To accomplish this, we be-

gin by determining the semantic differences between

the candidate events. For instance, if two events e

and e

exhibit partial alignment with the current state,

we compute the set of distinguishing features between

them, denoted as D = ||e

|| − ||e

||. The sensing plan

is then updated as follows: R = R ∪ D. This enables

to decide which of the events is actually ongoing.

Validating the Optimization of a Building Occupancy Monitoring Software System

159

4.3 Sensing Plans Optimization

Our BOMS optimization is achieved through mini-

mizing the real-time sensing plan to involve less sen-

sors in tracking the actual occupancy state. It consists

of tracking the actual occupancy state and matches it

to the events deﬁned in L, so that our BOMS always

ends with a good match case. Following the actual

state, the optimization algorithm alternates between

the different cases (good, partial, and no-match) with

the lowest number of sensors active.

A sketch of the optimization algorithm is shown in

Algorithm. 1. The optimization results from the case

an event occurrence is consistently conﬁrmed across

consecutive iterations, resulting in a reduction in the

number of sensors required to track the event.

Upon initial identiﬁcation (good match) of an

event, no optimization takes place. However, upon the

second consecutive conﬁrmation, all sensors assigned

to features with low conﬁdence are deactivated.

Similarly, with further consecutive conﬁrmations

of the expected event, sensors assigned to features

with medium conﬁdence are deactivated. Beyond this

point, no further optimization occurs as the sensing

plan comprises only a few high-conﬁdence sensors

that are sufﬁcient to track and identify the ongoing

occupancy event.

Algorithm 1: Controller Strategy.

1: E

Expected

∈ Events ▷ Select an event to be

monitored.

2: AdjustSensors(E

Expected

) ▷ Sensors for the event.

3: Monitor(E

Expected

) ▷ Monitor the event.

4: switch MonitorStatus(E

Expected

) : ▷ Status of

event.

5: case eventOccurred: ▷ The event has occurred.

6: OptimizeSensors(E

Expected

)

7: Go to step 3 break.

8: case otherEventOccurred: ▷ Other event has

occurred.

9: E

Expected

← E

Other

▷ Select other event.

10: AddSensors(E

Expected

) ▷ Add sensors for the

event.

11: Go to step 3. break.

12: case eventMaybeOccurred: ▷ Event may have

occurred.

13: Go to step 2. break. ▷ Monitor the event

again.

14: case default: ▷ Unexpected event.

15: TurnOnSensors() ▷ Gather more information

using more sensors.

16: Go to step 3. break. ▷ Monitor the system.

5 FORMAL VERIFICATION AND

VALIDATION

To conduct formal veriﬁcation and validation of the

proposed optimization, we modeled our BOMS in

UPPAAL and speciﬁed the properties to look for

in CTL logic. The model speciﬁcation is modular

and consists of three components (templates), a Con-

troller, a Sensor and an Environment. The templates

can be instantiated with different parameters to create

different system architectures and conﬁgurations, and

integrate new sensors, events and environments. The

model is available here.

5.1 Model of the BOMS Control

The Controller process monitors the environment us-

ing Sensor instances, while the Environment simu-

lates the occupancy events. The Controller template

is illustrated in Figure 1, and summarized functional-

ity of the Controller is depicted in Algorithm 1. The

Controller recognizes occupancy events while mini-

mizing the energy consumption of the Sensors. This

is achieved by activating and deactivating the Sen-

sor instances according to the occupancy current state

using the turn on and turn off channels, respectively

(which correspond to the functions AddSensors() and

OptimizeSensors() in the algorithm) and by adjusting

the sampling rate of the Sensors using the global vari-

able sample frequency (which corresponds to Adjust-

Sensors() in the algorithm) since the energy consump-

tion of a sensor depends on its sampling rate and the

time duration it is active. The Controller only mon-

itors one event at a time (E

Expected

) and switches be-

tween the different events to monitor based on the oc-

cupancy state. In Algorithm 1, line 5 corresponds to a

good matching to the expected event, line 8 describes

no matching to the expected event but another event

is recognized. Line 12 is a partial matching of the ex-

pected event, while line 12 describes no matching to

any of the events in L.

5.2 Sensors Modeling

The Sensor template, depicted in Figure 2, com-

prises two locations: On and Off. It samples the

Environment and reports the sampled value to the

Controller when active. The sensor value is read

from the global array environment state, that con-

tains the current state of the Environment and is up-

dated by the Environment template, through func-

tion read() which corresponds to Monitor() in the

algorithm. The periodic sensing, according to the

frequency sensor f requency[id], is modeled using a

ICSOFT 2025 - 20th International Conference on Software Technologies

160

Figure 1: The Controller template.

state invariant in the On location and a clock con-

straint on transition performing the sampling based

on the sampling rate of the Sensor (frequency) and

its local clock variable sample clock. The Controller

Validating the Optimization of a Building Occupancy Monitoring Software System

161

adjusts the sampling frequency by updating the ar-

ray sensor f requency[id], with which each sensor

synchronizes using its Id. Furthermore, each sensor

keeps track of its energy consumption by accounting

for the number of samples taken and the number of

time units spent in the location On through a global

variable energy consumption.

Figure 2: The Sensor template.

5.3 Environment Modeling

The Environment template, depicted in Figure 3,

comprises two locations and periodically generates

occupancy events. The events are generated non-

deterministically upon which a global variable repre-

senting the current occupancy state is updated. Such a

variable is read by the function read() of the different

sensors. The periodic behavior of the Environment

is modeled using a clock constraint on the transition

that generates the new events based on the global vari-

able event period that determines the time between

two events. The event period has to be much larger

than sampling period to ensure that the Controller has

time to detect the event. The Environment is made ﬁ-

nite by constraining the number of events generated

before the Environment goes into an Off state.

5.4 Validation Properties

We speciﬁed and veriﬁed the following properties for

the optimized BOMS using UPPAAL. The queries are

given in the formulas 1, 2, 3, 4, 5 and also available

here:

• Correctness: The Controller should be able to rec-

ognize all the events generated by Environment.

• Energy Consumption: The energy consumption of

the system should be less than certain threshold.

• Response Time: The response time of the system,

deﬁned as the time it takes for the Controller to

detect an event, should be less than certain thresh-

old.

• Resilience: The system should exhibit resilience,

ensuring that the Controller is always able to de-

vise a strategy to monitor the occupancy, even if

the system is in an unexpected state.

• Accuracy: Each event recognized by the Con-

troller must match the corresponding event gen-

erated by the Environment.

• Performance: The total number of accumulated

sensor samples is always less than or equal to cer-

tain threshold.

Such properties are speciﬁed as follows:

Correctness ≜ Env.Execute → (Ctrl.event = Env.event)

(1)

Energy ≜ A□ energyConsumption ≤ E

max

(2)

Response time ≜ A□ ctrl.timeSinceDetection ≤ R

max

(3)

Resilience ≜ A□ not deadlock (4)

Performance ≜ A□ active sensor ≤ X (5)

For the readers who are not familiar with the UP-

PAAL syntax, it should be noted that p → q corre-

sponds to A[](p => A <> q) in CTL.

The response time of the system is monitored us-

ing the clock variable timeSinceDetection that is reset

every time the Controller detects an event. The re-

sponse time property ensures that the Controller will

always detect an event within certain time period (3

sampling periods in our case: 2, 4 and 8 minutes). The

resilience property ensure liveness of the optimized

BOMS system, by which the Controller can always

derive a strategy and evolve following the occupancy

state. The energy consumption property is simply the

sum of the energy consumption of each Sensor, and

is veriﬁed by ensuring that the energy consumption

of the system is less than a certain threshold obtained

from the proactivation algorithm. Finally, the correct-

ness property is veriﬁed to ensure that any event that

occurs in the BOMS will eventually be detected by

the Controller.

Our formal veriﬁcation demonstrates that the

aforementioned properties of the original state of the

art BOMS are satisﬁed by our proactivation BOMS,

that is an optimization of the Ofﬁce Master 3000

BOMS.

ICSOFT 2025 - 20th International Conference on Software Technologies

162

Figure 3: The Environment template.

6 IMPLEMENTATION AND

EXPERIMENTAL VALIDATION

The proposed BOMS system and optimized occu-

pancy tracking algorithm have been implemented in

C++. The implementation is made modular as an in-

tegration of the three components (sensors, environ-

ments and controllers) so that new sensors, events and

environments can be integrated and monitored in a

straightforward manner. For reproducibility, the soft-

ware implementation and data are available here.

We carried out an experiment of a 12-hour period

to evaluate the performance and energy saving of our

optimized BOMS system. To this end, we considered

only one environment (building space) together with

the corresponding controller and 2 hubs of sensors

each of which is composed of 4 sensors (CO

, mo-

tion, noise, light).

The key metrics we considered in the evaluation

are the decision accuracy, the percentage of spared

sensor samples and the energy saved due to hav-

ing sensors turned off and/or operating with low fre-

quency. The number of sensors being active at run-

time is depicted in Figure 4. One can see that the sens-

ing plan of our optimized BOMS rarely operates all of

the eight sensors thanks to the knowledge-driven anal-

ysis. The number of active sensors ﬂuctuates from

3 to 6 in the ﬁrst 500 minutes due to different fea-

tures and occupancy states detected. After 500 min-

utes, the monitored building space is mostly empty

where the only event recognized is Empty, this is why

the number of active sensors is reduced and ﬂuctuates

between 2 and 4.

As a result, the accumulated number of sensor

samples collected by our optimized BOMS for 12

hours is 4212. The total number of samples needed

for the original Ofﬁce Master 3000 BOMS, where all

the sensors run regularly with the same 2-minutes fre-

quency would be 6480. Thus, our BOMS enables to

save 35% of the data sampling compared to the Of-

ﬁce Master 3000 BOMS and also the state to the art

algorithms (Ortiz Perez et al., 2018; Abraham and Li,

2014) where sensors are permanently active, while

achieving the same accuracy and state knowledge.

As for energy consumption, for a 12-hours opera-

tion our optimized BOMS enables a saving of 9.75kw,

(6480-4212)*0.0043 (Vafamand et al., 2020), com-

pared to the original Ofﬁce Master 300s that would

need 27.8kw, which represents a saving of 35%.

The accuracy analysis is depicted in Figure 5. It

refers to the matching of the occupancy events iden-

tiﬁed by our BOMS algorithm and the actual oc-

cupancy events. The achieved accuracy ﬂuctuates

mostly between 40% and 100% with an average mean

of 84%. This means that in 84% of the experi-

ment time, the expected event (computed according

to the optimized sensing plan) matches the actual

event, whereas no-matching cases (lowest value for

blue line) represent 3.7% of the experiment time, and

partial-matching pattern is 12.3%.

Furthermore, one can see that there is a correla-

tion of the accuracy analysis with the optimized sens-

ing plan. The lower the number of sensors and the

higher the state dynamics are, the lower the accuracy

is. However, upon each low accuracy a new sensing

plan is computed by integrating more sensor samples

incrementally to solve the partial-matching and no-

matching cases. This led to shorten the time periods

Validating the Optimization of a Building Occupancy Monitoring Software System

163





■ Active sensors























































Time (minutes)

Figure 4: Number of sensors active at runtime.

Figure 5: Accuracy of the optimized sensing plan.

for each accuracy drop. In average, each accuracy

drop does not last more than 4-6 minutes.

A partial matching case is depicted in Figure 6

where the occupancy state recognized matches par-

tially the features from 3 different events. As an ex-

ample to solve this inconclusive situation would be

to operate much more sensors to conﬁrm/deny one of

the features so that to exclude either the ﬁrst 2 op-

tions (Lecture, Meeting) or the last option (Exercise

session).

The outcomes resulting from the software experi-

ment match the veriﬁcation results conducted on the

model using UPPAAL such as the maximum duration

to recognize an event and the accumulated number of

sensors for the same data set is always below 4300.

7 CONCLUSION

This paper proposed an intelligent knowledge-driven

real-time occupancy monitoring solution (BOMS) as

an optimization of the actual Ofﬁce Master 3000

BOMS. The proposed analysis enables tuning the fre-

quency of the sensors on-the-ﬂy to reduce data sam-

pling and energy consumption. Rather than collecting

large amounts of sensor data to perform occupancy

analysis post hoc; we proactively identify the minimal

data relevant to the actual state following the seman-

tics of the expected activities.

The proposed BOMS is mechanized in UPPAAL

to perform formal veriﬁcation and validation of the

non-functional properties of the original BOMS post-

optimization. An early proof-of-concept prototype

has been implemented in C++. Furthermore, our so-

ICSOFT 2025 - 20th International Conference on Software Technologies

164

Figure 6: Partial matching to multiple events.

lution has been tested and compared to baseline occu-

pancy analysis. The experiment results showed that,

while achieving a considerable reduction in computa-

tion cost (up to 35%) and energy consumption (up to

31%), it maintains high accuracy for the occupancy

tracking (up to 84%).

As future work, we plan to integrate much more

features and events to achieve a real-world occupancy

state catalog. Moreover, we will also conduct a thor-

ough analysis and validation of the proactivation on

different case studies and compare the optimization

results to those of machine-learning techniques.

REFERENCES

Abraham, S. and Li, X. (2014). A cost-effective wireless

sensor network system for indoor air quality monitor-

ing applications. Procedia Computer Science, 14(34).

Ahmad, J., Larijani, H., Emmanuel, R., Mannion, M.,

and Javed, A. (2021). Occupancy detection in non-

residential buildings – a survey and novel privacy pre-

served occupancy monitoring solution. Applied Com-

puting and Informatics, 17(2).

Alur, R. and Dill, D. L. (1994). A theory of timed automata.

Theoretical Computer Science, 126(2).

Austin, M., Delgoshaei, P., Coelho, M., and Heidarine-

jad, M. (2020). Architecting smart city digital twins:

Combined semantic model and machine learning ap-

proach. Journal of Management in Engineering,

36(4).

Azimi, S. and O’Brien, W. (2022). Fit-for-purpose: Measur-

ing occupancy to support commercial building opera-

tions: A review. Building and Environment Journal,

22(212).

Baerentzen, M. U., Boudjadar, J., ul Islam, S., and Schultz,

C. P. L. (2023). A knowledge-based proactive intel-

ligent system for buildings occupancy monitoring. In

ICSOFT, pages 680–687.

Baier, C. and Katoen, J.-P. (2008). Principles of Model

Checking. The MIT Press.

Boudjadar, A., Vaandrager, F., Bodeveix, J.-P., and Filali,

M. (2013). Extending uppaal for the modeling and

veriﬁcation of dynamic real-time systems. In Arbab,

F. and Sirjani, M., editors, Fundamentals of Software

Engineering. Springer Berlin Heidelberg.

Boudjadar, J., David, A., Kim, J. H., Larsen, K. G., Nyman,

U., and Skou, A. (2014). Schedulability and energy

efﬁciency for multi-core hierarchical scheduling sys-

tems. In International Embedded Real-time Systems

Sysmposium ERTS2.

Boudjadar, J. and Khooban, M. (2020). A safety-driven cost

optimization for the real-time operation of a hybrid

energy system. In Proceedings of the 27th Interna-

tional Conference on Systems Engineering (ICSEng).

Boudjadar, J. and Tomko, M. (2022). A digital twin setup

for safety-aware optimization of a cyber-physical sys-

tem. In Proceedings of the 19th International Con-

ference on Informatics in Control, Automation and

Robotics.

Cala, D., Matthes, P., Huchtemann, K., Streblow, R., and

uller, D. (2015). Co

based occupancy detection

algorithm: Experimental analysis and validation for

ofﬁce and residential buildings. Building and Envi-

ronment Journal, 86.

Costenaro, D. and Duer, A. (2012). The megawatts be-

hind your megabytes: Going from data-center to desk-

top. In ACEEE Summer Study on Energy Efﬁciency in

Buildings.

Dai, X., Liu, J., and Zhang, X. (2020). A review of stud-

ies applying machine learning models to predict occu-

pancy and window-opening behaviours in smart build-

ings. Energy and Buildings, 20(223).

E. M. Clarke, J., Grumberg, O., and Peled, D. A. (1999).

Model Checking. MIT Press.

Elkhoukhi, H., NaitMalek, Y., Berouine, A., Bakhouya, M.,

Elouadghiri, D., and Essaaidi, M. (2018). Towards

a real-time occupancy detection approach for smart

buildings. Procedia Computer Science, 18(134).

Jabirullah, M., Khan, A., Ali, M., Wajih, S., and Hussain,

M. (2021). Iot-based occupancy monitoring tech-

niques for energy efﬁcient smart buildings. Turkish

Online Journal of Qualitative Inquiry, 21(12).

Jiang, J., Wang, C., Roth, T., and Nguyen, C. (2022).

Residential house occupancy detection: Trust-based

scheme using economic and privacy-aware sensors.

IEEE Internet of Things Journal, 9(3).

Jiang, W. and Yin, Z. (2015). Human activity recognition

using wearable sensors by deep convolutional neural

networks. In Proceedings of the 23rd ACM Interna-

tional Conference on Multimedia.

Lasla, N., Doudou, M., Djenouri, D., Ouadjaout, A., and Zi-

zoua, C. (2019). Wireless energy efﬁcient occupancy-

monitoring system for smart buildings. Pervasive and

Mobile Computing Journal, 19(59).

Lou, X., Lam, K. P., Chen, Y., and Hong, T. (2017). Perfor-

mance evaluation of an agent-based occupancy simu-

Validating the Optimization of a Building Occupancy Monitoring Software System

165

lation model. Technical report, School of Architec-

ture, Carnegie Mellon University.

McKenna, E., Krawczynski, M., and Thomson, M. (2015).

Four-state domestic building occupancy model for en-

ergy demand simulations. Energy and Buildings,

15(96).

Minoli, D., Sohraby, K., and Occhiogrosso, B. (2017).

Iot considerations, requirements, and architectures

for smart buildings - energy optimization and next-

generation building management systems. IEEE In-

ternet of Things Journal, 4(1).

Nienaber, F., Wolf, S., Wesseling, M., Cal

ı, D., Muller,

D., and Madsen, H. (2020). Validation, optimisation

and comparison of carbon dioxide-based occupancy

estimation algorithms. Indoor and Built Environment

Journal, 29(6).

Ortiz Perez, A., Bierer, B., Scholz, L., Wollenstein, J., and

Palzer, S. (2018). A wireless gas sensor network to

monitor indoor environmental quality in schools. Sen-

sors Journal, 18(12).

Pan, S., Bonde, A., Jing, J., Zhang, L., Zhang, P., and Noh,

H. Y. (2014). Boes: Building occupancy estimation

system using sparse ambient vibration monitoring. In

Proceedings of Sensors and Smart Structures Tech-

nologies for Civil, Mechanical, and Aerospace Sys-

tems SPIE.

Rai, S., Wang, M., and Hu, X. (2015). A graph-based

agent-oriented model for building occupancy simu-

lation. In Proceedings of the Symposium on Agent-

Directed Simulation.

Rault, T., Bouabdallah, A., and Challal, Y. (2014). Energy

efﬁciency in wireless sensor networks: A top-down

survey. Computer Networks, 14(67).

Salimi, S. and Hammad, A. (2019). Critical review and re-

search roadmap of ofﬁce building energy management

based on occupancy monitoring. Energy and Build-

ings, 19(182).

Salimi, S. and Hammad, A. (2020). Sensitivity analysis

of probabilistic occupancy prediction model using big

data. Building and Environment, 20(172).

Seghezzi, E., Locatelli, M., Pellegrini, L., Pattini, G.,

Di Giuda, G. M., Tagliabue, L. C., and Boella, G.

(2021). Towards an occupancy-oriented digital twin

for facility management: Test campaign and sensors

assessment. Applied Sciences, 11(7).

Shokrollahi, A., Persson, J. A., Malekian, R., Sarkheyli-

agele, A., and Karlsson, F. (2024). Passive infrared

sensor-based occupancy monitoring in smart build-

ings: A review of methodologies and machine learn-

ing approaches. Sensors Journal, 24(5).

Sun, K., Zhao, Q., and Zou, J. (2020). A review of building

occupancy measurement systems. Energy and Build-

ings, 20(216).

Teng, S. Y., Tou

s, M., Leong, W. D., How, B. S., Lam,

H. L., and Masa, V. (2021). Recent advances on in-

dustrial data-driven energy savings: Digital twins and

infrastructures. Renewable and Sustainable Energy

Reviews, 21(135).

Tien, P. W., Wei, S., Calautit, J. K., Darkwa, J., and Wood,

C. (2022). Real-time monitoring of occupancy activ-

ities and window opening within buildings using an

integrated deep learning-based approach for reducing

energy demand. Applied Energy, 22(308).

Trivedi, D. and Badarla, V. (2020). Occupancy detection

systems for indoor environments: A survey of ap-

proaches and methods. Indoor and Built Environment

Journal, 29(8).

Vafamand, N., Boudjadar, J., and Khooban, M. H. (2020).

Model predictive energy management in hybrid ferry

grids. Energy Reports, 6:550–557.

Yang, J., Santamouris, M., and Lee, S. L. (2016). Review

of occupancy sensing systems and occupancy model-

ing methodologies for the application in institutional

buildings. Energy and Buildings, 16(121).

Zhang, H.-B., Zhang, Y.-X., Zhong, B., Lei, Q., Yang, L.,

Du, J.-X., and Chen, D.-S. (2019). A comprehen-

sive survey of vision-based human action recognition

methods. Sensors Journal, 19(5).

Zhang, W., Wu, Y., and Calautit, J. K. (2022). A review

on occupancy prediction through machine learning for

enhancing energy efﬁciency, air quality and thermal

comfort in the built environment. Renewable and Sus-

tainable Energy Reviews, 22(167).

ICSOFT 2025 - 20th International Conference on Software Technologies

166