Several authors ^[1][2][3][7,8,9] state that office building users’ behaviors are the most relevant aspect that influences energy consumption. Sometimes office users leave their computers turned on needlessly (e.g., on lunch breaks, on weekends, and during night periods). These behaviors increase electricity bills and are difficult to change with automation systems, demonstrating their energy inefficiency ^[4][10].

AWe propose a methodology is proposed to identify user behaviors for potential behavior-change interventions. Sets of behaviors, interactions, and associated events that occur in a given time and space, with well-defined objectives and rules, can be defined as virtual choreographies ^[5][11]. This concept sees such virtual representations as being independent from the platform on which they are recorded and later analyzed or replayed. It emerged from efforts at analysis of multi-user behaviors in virtual training scenarios where users interact with the environment collaboratively ^[6][12].

According to the main goal, we present the state of the art regarding the role played by individual behaviors in energy consumption. We will also look at how human behaviors can be represented, concluding with the theoretical approach regarding virtual choreographies.

Reducing the energy consumption in buildings is a critical component of carbon reduction commitments and has become a growing relevant area of work and research ^[7][1]. Buildings represent approximately 40% of the total energy consumption ^[8][13]. According to the Buildings Performance Institute Europe report of 2011 ^[9][14], office buildings alone account for 26% of the total energy consumption within the building sector.

Studies on low carbon emissions ^[10][3] recommend transitioning towards greener business practices and improving automation systems, not forgetting individual behaviors. The behaviors and choices of individual users may influence the consumption patterns, and those individual behaviors are responsible for 30 to 40% of the total annual CO₂ emissions in the United States ^[11][4]. However, few studies have addressed the issue of the behavior of individuals in organizations/companies, as discussed below: prior research has mostly looked into the actions of users in the context of households.

There are two components of energy consumption in buildings: regulated and unregulated ^[12][15]. The total operational energy consumption of regulated components, such as heating, cooling, hot water, fans, and pumps, is generally well optimized. However, the consumption patterns associated with unregulated elements in office buildings like IT equipment (office printers, desktop and laptop computers), lab equipment, catering facilities, localized heating or cooling, lighting, etc., cannot be easily controlled by automation systems because they depend mainly on human behaviors ^[13][16].

Occupants’ behaviors significantly influence the relevant discrepancies between buildings with the same climate location and functionalities ^[14][5]. The way that building occupants set their comfort levels and related criteria (for instance, thermal and visual) influences the building energy systems. In addition, the responses to those environmental changes, to achieve comfort levels, directly affect energy use and the overall operation of buildings ^[14][5].

In a typical office building, lights consume 40% of total energy, heating and central cooling systems around 25%, and the rest is from plugged-in electrical equipment (35%) ^[15][6]. Even more relevant, if analyzing the electricity use in buildings with high-efficiency systems, the plugs’ electrical load can represent 50% of total consumption ^[16][17]. Here can divide the energy consumption of these types of buildings into lighting, computers, and air conditioning. Several studies ^[16][17][18][17,18,19] indicate that it is possible to optimize the usage of these three vectors of office equipment with considerable energy savings by changing individual users’ behaviors.

There are several factors upon which those users’ behaviors depend (economic, ethical, and social-related) making it difficult to change their impacts solely with automation systems increasing electricity consumption. For instance, sometimes users frequently leave the computers turned on for long periods intentionally to minimize boot-up time (lunch, weekends, etc.) ^[4][10].

A study conducted in the USA on office buildings ^[19][20] found that most electric equipment is always on, almost 90% of desktop computers are not configured to enter low-power mode, and 50% of computer monitors enter safe mode. Another study ^[20][21] based on the quantities of energy wasted during non-occupied hours in commercial buildings highlights opportunities for implementing individual behavioral changes in service buildings.

Many mechanisms are used in the design phase of buildings that can predict, using simulations, the total energy consumption. However, there is a considerable difference between the expected consumption and the effective one. Individual behaviors and the occupants’ preferences are some of the most relevant factors that influence that identified difference ^[21][22]. So there must exist effective strategies aiming to understand user awareness of its impact and their expectations and concerns. Many research surveys ^[22][23][24][23,24,25] have sought to understand these consumer preferences about energy consumption and their perceptions related to demand response and energy efficiency behaviors. Several authors ^{[1][2][3][25]}[7,8,9,26] state that individual behaviors of office buildings users are the most relevant aspect that influences energy consumption.

There are several research efforts that have sought to influence occupants’ behaviors. Hoes et al. ^[26][27] propose simulation tools; however, this kind of approach does not deal with the diversity and complexity of users’ behaviors. Another approach was the use of power meters to provide the basic information on appliance consumption, but this approach is unable to define usage patterns because they were not made to discriminate energy consumption at the individual user level ^[27][28].

Although some approaches ^[28][29][30][29,30,31] explore non-intrusive load monitoring in order to obtain data on the energy consumption of buildings at the equipment level, they still fail to correlate the consumption with the occupants’ activities. However, even if it were possible to differentiate this information in some way, as Berges et al. ^[31][32] point out, care should always be taken to correlate consumption with behavior so as not to generate out-of-context results.

Understanding human behavior is fundamental in society. Computer-supported approaches for this understanding require methods to represent human behaviors in information systems. 

The need to represent human behavior stems from the desire to analyze it with software tools. Such tools arose in the early 1990s when the Cold War’s culmination brought new military challenges and tasks to NATO ^[32][33]. All this was due to the advent of innovative technologies that were beginning to have a tremendous impact on implementing simulation systems and decision support tools. It was then that the digital representation of human behaviors became vital to empowering decision-support tools and simulators ^[32][33].

Uwe Dompke is a German Air Force officer who led studies in the NATO Research and Technology Organization (RTO now STO), namely in modelling and simulation to support training, education, and decision-making, especially in the area of human behavior representation. Regarding the term “human behavior”, he defined it as “a purposive reaction of a human being to an idiosyncratic meaningful situation” ^[32][33]. A few years later, Elizabeth Hutchinson ^[33][34] defined human behavior as the interaction between a person and the environment.

In practical terms, human behavior occurs when there is a change from one state into another (bodily and/or mentally) with a particular goal, which can be externally observable. It does not require an associated logic nor an appropriate reaction, and possesses three interconnected components (socio-affective, psycho-motor, and cognitive) ^[32][33]. In order to perceive human behavior, one also needs to consider a multidimensional approach (time, person, environment) ^[34][35].

There are several methods to model human behaviors. Schmidt ^[35][36] presents the PECS (physical conditions, emotional state, cognitive capabilities, social status) reference model that aims to replace the BDI (belief, desire, intention) model initially developed by the philosophical expert Michael Bratman ^[36][37]. However, the USA Department of Defense combined the PECS model and the BDI, thus presenting the Human Behavior Representation (HBR) framework to model human behavior ^[32][37][33,38].

Dompke ^[32][33] states the aspects that should be considered when modelling human behavior:

• Considering that the human behavior has a purpose, besides modelling that behavior, there should always be associated a SMART (specific, measurable, acceptable, realistic, and timed) objective;
• The associated goal should represent the optimal behavior;
• In a simple way, to model a behavior, it should be necessary to determine the initial value(s), the process that leads to the result, and the change to achieve the goal;
• One should represent behaviors that are relevant to one’s analysis needs;

The main goal of the HBR approach is to create a computational model of human behavior that can express the observed variability in behaviors according to differences in the person’s characteristics, in their situation, or in their interplay ^[38][39].

Another perspective in the area of behavior representation relates specifically to behavioral change, which is one of the pillars underpinning this research work. For behavioral change to occur, it is necessary to identify a set of activities specially designed to change specific behavior patterns. These patterns are measured by the number of times they occur in a given population group under study ^[39][40].

There are methods and ways to report, evaluate, and understand behavioral change interventions through specific rating techniques ^[40][41]. Medicine and the natural sciences are, in fact, the leading exponents of these approaches ^[41][42] with the use of taxonomies ^[42][43][43,44]. Ontologies, as described in a scoping review led by Norris et al. ^[40][41], “extend the hierarchical nature of taxonomies” by presenting the following advantages:

• They allow unique identification of entity types (objects, attributes, processes), thus eliminating ambiguity;
• They enable the precise definition and classification of these identifiers;
• They also help organize the relationships between these identifiers.

Therefore, in comparison with taxonomies, ontologies allow a greater and more detailed knowledge at the level of the representation of behaviors ^[40][41]. They also allow different theoretical perspectives with the help of conceptual frameworks and make it possible to compare multiple fields of study with large datasets ^[44][45][45,46]. It is also a relevant fact that ontologies allow manual updates according to the evolution and development of the domain itself, enabling a permanent update ^[46][47].

The use of ontologies has indeed been revolutionary in several domains (e.g., computational modelling of biological systems ^[47][48], or the creation of repositories accessible to the scientific community ^[48][49]. Given the significant impact of ontologies in other areas of knowledge, the scope of behavioral change is also included, namely with the Human Behavior-Change project (https://www.humanbehaviorchange.org/, accessed on 2 December 2020) ^[49][50]. This results from a collaboration between several areas of science (behavioral scientists, computer scientists, and systems architects), proposing tools and guidelines that help researchers and others interested in behavioral change themes ^[40][41].

But how are ontologies used in information science?

There is space for the use of ontologies whenever semantic contexts are used or needed, i.e., giving meaning to information ^[46][47]. In this sense, the consortium responsible for standardizing the technologies associated with the World Wide Web (https://www.w3.org/OWL/, accessed on 2 December 2020) has defined an OWL (Web Ontology Language) ^[50][51] as being responsible for representing ontologies in information systems. In computer science, an ontology is defined as a formal definition (through a well-defined syntax and semantics language) of concepts and their relationships for a given domain ^[46][47].

In practice, two fundamental components allow designing the semantic application (knowledge base and inference engine). The knowledge base is entirely linked to the ontology schema (what kinds of statements are possible) and to the facts, represented through a formal language ^[46][47].

Despite the approaches previously listed (HBR and the pure use of ontologies) for representing human behavior in the context of behavioral change, it is considered that, given the need for a more simplified approach and the different characteristics present in the context of this thesis (behaviors related to energy consumption in office buildings), the use of virtual choreographies will be the approach to take into consideration.

Common use of the term choreography occurs when referring to “the skill of combining movements into dances to be performed” or “the movements used by dancers especially in performing ballet, or the art of planning such movements” ^[51][52].

From a computer science point of view, the term is also used as a new view on interacting services associated with the “Web Services” technology ^[52][53]. There is even a Web Service Choreography Description Language (WS-CDL (https://www.w3.org/TR/ws-cdl-10/ accessed on 5 March 2021) that supports a top-down approach in the design and implementation of those services.

However, the approach to the term that will be considered in this work corresponds to the following definition: Virtual choreographies are sets of behaviors, interactions, and associated events that occur in a given time and space, with well-defined objectives and rules ^[5][11]. Those virtual choreographies can be performed by human-controlled actors and/or computer-controlled actors (also known as “bots” or “non-player characters” ^[53][54]) or even by non-embodied entities, such as temperature or conceptual networks. They enable the analysis of the behaviors independently of the physical platform upon which they occur ^[54][55], and thus can be reproduced on different platforms to serve different needs ^[55][56].

There are several contexts that need to use collaborative virtual systems based on multi-user behaviors, and that can include choreographed scenarios (e.g., aircraft maintenance ^[56][57], industry simulation ^[57][58], disaster simulation ^[58][59]). Furthermore, in research, virtual choreographies are also included in some scientific experiments ^[6][12].

However, considering that this work aims to identify which behaviors can reduce energy consumption, it is relevant to identify how to represent these behaviors. In this sense, the direction of the present research will consider the reference identified in the previous section to the use of ontologies in behavioral change and the approach taken by Silva et al. ^[59][64], which presents the choreography representation through an ontology-based model.

It is necessary to create a choreography to take into account the following elements:

• Actors: characters that perform the behaviors in a choreography. This includes both human-controlled and computer-controlled actors, and might include non-embodied concepts;
• Action: this is a specific interaction within the environment; for instance, actors walking, gesturing, talking, manipulating, etc., and also automatic doors opening or machines running, or a conceptual element emerging or fading;
• Objects: elements that are not actors but can be acted upon by actors;
• Roles: higher-order semantic context of an actor or object, providing meaning for their actions, location, and overall features;
• Scenario: the stage where a choreography takes place. It may include objects and general characteristics (such as daytime, gravity, etc.);
• Space-time: dynamic changes and evolution of the choreography, as actors and objects have specific roles and interact with each other in the scenario over time.

4. Conclusion

3. Conclusion

AThroughout our article, a method was presented which consists in identifying virtual choreographies as sets of behaviors within a broader context, by combining three sources of data—observations, meters, and computer logs—so that herewe can obtain more contextualized results regarding users’ final consumption, thus enabling the creation of new methodological approaches that help solve the problem identified by Berges et al. ^[31][32]. These choreographies can be assessed as targets for behavioral change by analysing their energy consumption impact and their potential for successful behavior change. If only isolated actions were used as an analysis parameter, behavior change success would be erratic because there would be no distinction, for example, between leaving the computer for a short while for nature’s call or leaving it for longer periods (e.g., lunchtime). This lack of contextual meaning would result in less targeted efforts. Moreover, the awareness of actual energy consumption impact by choreography instead of individual actions enables prioritization of behavior-change efforts, towards those with the greatest impact potential in energy consumption. The proposed method makes it possible to identify behaviors with a richer semantic context (differentiating going out to lunch from going out to a meeting, for example), which allows for greater clarity in identifying the incentives to be generated for each action detected. Moreover, by associating each behavior with the effect on the target variable that is intended to be affected (in this case, energy consumption), herwe can identify the behaviors whose change is most effective, thus reducing the potential to generate contradictory incentives. In the specific case under study, which aimed to identify the behaviors related to the energy consumption of users in office buildings, five of the actions studied were identified as actions that users usually perform and that can be subject to behavioral change, thus generating a reduction in energy consumption. In this sense, a direct implication of this methodology of identifying behaviors that could be changed was relevant for the design of a gamified mobile application aimed at reducing energy consumption in offices. Thus, we can conclude that this method seems promising in systematically obtaining behaviors to be changed towards a goal. Considering that in a typical office building, as presented in the state of the art, the lights consume 40% of total energy and another 35% is from the electrical equipment [6], if we can promote the targeted identified behavior change choreographies, this can bring significant energy consumption reduction.