Virtual reality as a new frontier for energy 1 behavioural research in buildings: tests 2 validation in a virtual immersive office 3 environment

7 Occupants’ behaviour and strategies to encourage behavioural changes need to be 8 addressed in workplaces to reduce energy consumption. In this study, the Theory of 9 Planned Behaviour (TPB) was integrated for the first time with an office virtual 10 environment (VE) to investigate the adequacy of the VE in the comfort and behaviour 11 domain while understanding its effect in predicting individuals’ energy-related 12 intention of interaction with the building systems. One hundred four participants, 13 randomly divided into two groups, were recruited to answer questionnaires (TPB, 14 comfort, interactions, sense of presence and cybersickness). Two test sessions were 15 conducted at a constant indoor air temperature: an in-situ experiment was compared 16 with the virtual counterpart. Findings revealed an excellent level of presence and 17 immersivity and the absence of high disorder levels. A good agreement between the 18 two environments was highlighted in terms of thermal comfort, number, and type of 19 interactions (one interaction focused on window opening for 71-81% of subjects). 20 Moreover, no differences were discovered between the results of a multiple regression 21 model in both real and virtual environments. In particular, the analysis identified the 22 knowledge of energy consumption as the main predictor of behaviour because it 23 accounted for about 12% of the variation in the intention of interaction in both tested 24 environments. Thus, the suitability of the virtual environment could offer an effective 25 tool for decision-makers and researchers to develop strategies aimed at designing more 26 comfortable and less energy-consuming buildings.


32
A Renovation Wave for Europe was proposed by the EU Commission in 2020 to allow buildings to be less energy-33 consuming while creating more liveable spaces. In this domain, an important target for researchers, policymakers, and 34 public administrations is a clearer understanding of the factors driving energy consumption in the built environment.

35
The aim is to develop suitable strategies to aid economic and environmental targets while increasing end-users comfort, 36 satisfaction, health, and performance. However, technological progress and investments alone rarely guarantee low or 37 net-zero energy in buildings because «human factors» play a crucial role, and while the awareness of their impact has 38 improved, it is often ignored in building design. Indeed, it is well-established that occupants' behaviour is a major 39 factor affecting the energy performance of buildings. It is important to notice that users' energy-related behaviour 40 differs significantly between domestic and non-domestic use, where the dwellers directly pay for the energy 41 consumption while the company provides free energy for workers. Employees seem less motivated to engage in energy-42 saving behaviour than households that are more willing to save energy in their daily lives. As a result, during the last 43 years, energy consumption in commercial and services has increased, accounting for about 30% of European energy 44 demand [1]. Due to the large amount of time spent in workplaces (60-70% every week), workers constantly try to 45 provide comfortable working conditions [2]. Thus, a hot research topic has emerged to understand the factors affecting 46 people's behaviour and willingness to save energy in workplaces. Accordingly, technological development promoting energy efficiency needs to be integrated with a programme to encourage behavioural changes that could be a potential 48 solution to be adopted immediately.
Most of the research has already indicated that energy behaviour is a relatively complex task to understand because 50 it depends on several drivers: internal (occupants' activities and preferences) and external (building, equipment, 51 environment, time, contextual, random) factors. Thus, various theories and models have been introduced in this field, 52 such as the Theory of Planned Behaviour (TPB) developed by Ajzen et al. [3]. It explains that human behaviour is 53 guided by three factors: behavioural beliefs about the consequences of the behaviour itself, normative beliefs about the 54 expectation of others over the users' behaviour, and control beliefs related to the presence of factors that may facilitate 55 or limit the implementation of the behaviour. In particular: behavioural beliefs produce a favourable or unfavourable 56 attitude toward the behaviour, normative beliefs result in perceived social pressure or subjective norm, and control 57 beliefs determine perceived behavioural control. The combination of the attitude toward the behaviour, subjective norm 58 and perceived behavioural control produces a behavioural intention. In general, the users' intention to perform a 59 behaviour would be greater the more favourable the attitude, the less social pressure, and the greater perceived control.

60
In addition, in the presence of an opportunity and sufficient control, building users are expected to finalise the intention, 61 which is why it is assumed to be an immediate antecedent of the behaviour itself. Figure 1

88
The present study involved an independent-measure design experiment (52 subjects per group) in investigating the adequacy of the virtual environment in the comfort and behaviour domain. Two test sessions were conducted: each 90 participant was randomly assigned to a virtual condition or «immersive virtual environment» (group 1) or an in-situ 91 condition, or «real environment, RE» (group 2) session.

93
2.1 Test room 94 An office was set up like a test room located inside the Department of Engineering, Civil, Construction and Architecture 95 (Università Politecnica Delle Marche, Ancona, Italy). The test room had an internal dimension of 5.93x4.38m and a floor 96 ceiling height of 3.00 m. The room contained furniture to replicate an office working environment and was equipped with 97 a computer station to carry out the tests and the equipment for the IVE visualisation ( Figure 2). The thermal environment 98 depends only on the central HVAC system of the room, and the indoor air temperature was recorded by several probes 99 (temperature range: from +5° to +60° and accuracy ±0.3°) located at the feet (0.10m), waist (0.60m) and head (1.10m) 100 of the seated participants and above the table where the test was performed. To detect participants' energy-related 101 intention of interaction, a window, a fan, a heater, and an air conditioner were added to the room, but they were set off 102 and did not influence the thermal environment. Indeed, the participants did not directly interact with the climatic systems; 103 they only reported the adaptive response they would have wanted to carry out to improve their thermal comfort induced 104 by the HVAC of the room. So, no thermal outcome was experienced by the subjects. This strategy is supported by the 105 TPB, which states that the intention of interaction is antecedent to the behaviour itself, and as the occasion occurs, the 106 users would perform the intended behaviour. 107 108

Virtual environment 109
To create an IVE that can adequately replicate the double-occupancy office space, an extremely detailed 3D model 110 was created using CAD software and afterwards exported to Unity software [14] to apply materials, lights and cameras.

111
The luminance parameter (L*) and chromatic components (a*, b*) of the CIELab model were detected using a 112 spectrophotometer (CM-2500d Konica Minolta) to address the correct representation of surfaces' colour and materials.

113
Indeed, 5 measurements were carried out with a diameter of 8 mm for each surface of the office room: walls, desk, 114 chair, and floor tiles. Then, the resulting L*a*b* parameters were converted into RGB coordinates for the Unity model.

123
To create a model coherent with its real office counterpart for validation, the climatic systems (a window, a heater, 124 a fan, and an air conditioner) were also added in the virtual environment. After selecting their intention of interaction, 125 the subjects did not experience dynamic visual changes and thermal outcomes as in the real environment.    6 performed a productivity task (3 minutes) to stay focused and simulate a traditional working scenario during the test 174 session. However, no task performance assessment was later carried out in this study. Then, they answered a post-175 experimental survey.

176
In particular, in the IVE experiment, participants wore and adjusted the head-mounted display before the operative 177 phase, rested with their eyes closed for 30 seconds and adapted to the virtual scene for 3 minutes. In this way, any 178 psychological fluctuations related to the virtual environment exposure were reduced, and immersion was facilitated.

179
Responses to the productivity test and questions displayed on the virtual computer monitor were given by voice and 180 recorded by the researchers.

181
Each test session lasted about 20-25 minutes to reduce overall fatigue and exposure to the virtual environment.  Figure 4 shows the participants' percentage of votes across the real 232 and the virtual experiments. As expected, the temperature significantly influences TSV in both environments: at least 233 94% of the subjects felt from «slightly warm» to «hot». Therefore, the thermal condition was evaluated as not fully 234 comfortable (from «slightly uncomfortable» to «uncomfortable») by 66%-83% of the subjects, respectively, because 235 the selected temperature set-point was +4°C away from the usual winter thermal comfort temperature (20°C

291
In conclusion, the measurement model (CFA) confirms that the overall AC, AT, KE, and PBC contribute to analysing 292 the intention of interaction with the building systems of the total sample size (n=104). and perceived behavioural control. The significance level was set equal to 0.05 (5%). Table 5     experienced realism, involvement, and spatial presence) revealed that the virtual environment created an excellent level 326 of presence and immersivity, and most subjects did not report high disorder levels.

327
Secondly, a good agreement between the real and the virtual environment was discovered in terms of thermal comfort 328 and the number and type of interactions. In both environments, the temperature has a significant influence on thermal 329 sensation (at least 94% of the subjects felt from «slightly warm» to «hot»), and the selected temperature condition was 330 evaluated as not fully comfortable because the set-point was +4°C away from the usual winter thermal comfort 331 temperature (20°C). Thus, the majority (between 79% and 90%) of the subjects would have wanted to feel at least 332 «slightly cooler» and «cooler». Therefore, opening the window was highlighted as the best strategy to improve the 333 thermal sensation by decreasing the indoor temperature and enhancing air change in both RE and IVE.

334
After establishing a good model-of-fit (CFA analysis), multiple regression models of the environments were 335 compared to evaluate the suitability of the TPB in IVE in predicting participants' intention of interaction. The 336 comparison of the results did not reveal differences between RE and IVE, thus, supporting the adequacy of the 337 integration of TPB within the VR technology. In particular, the analysis identified the knowledge of energy 338 consumption as the main predictor, even if only a few subjects knew how much energy the electric appliances 339 consumed. This implies that a higher knowledge about this topic could significantly positively affect energy-related 340 behaviour, allowing individuals to interact correctly with the building equipment to make them comfortable while 341 saving energy in the workplace.

342
In conclusion, the suitability of the virtual environment could offer an effective tool for decision-makers and 343 researchers to develop strategies aimed at designing more comfortable, liveable and less energy-consuming buildings.

344
However, future studies should be conducted after adjusting the TPB survey to include other predictors in the model, 345 such as personal and social norms, habits in energy-saving behaviours, and time availability. Thirdly, the data were 346 collected on a hundred subjects, which may restrict the generalizability of the results, but the findings may be effective