As demands for low energy consumption in constantly increasing, architects and engineers is facing great challenges of saving energy while maintaining or even improving current comfort levels. The current trend in reduction of energy use in buildings is oriented towards sustainable measures and techniques aimed to energy need restraint: super-insulation, efficiency in plant systems, improved thermo-physical properties of building materials, attention to form and orientation, etc... However studies confirmed the existence of an outstanding gap between predicted and actual energy consumption in dwelling. These highlighted discrepancies are due to several factors. Socio-demographic characteristics such as household size, age of the Utilities, income, education level and country of origin have been recognized as parameters having an influence on the desired levels of comfort by the occupants. Existing dynamic energy simulation tools, developed in the last 10 years, exceed the static size of the simplified methods through a better and more accurate prediction of energy use; however they are still unable to replicate the actual dynamics that govern energy uses within buildings. Energy performance is dramatically affected by the behaviour of users inside buildings: each architectural project is proposed on the base of a singular spatial organization and occupier's usability, but once the building-plant system has been realized, this could be used differently from the assumptions made in the design phase.

As Humphreys stated (1997), "If a change occurs such as to produce discomfort, people react in ways which tend to restore their comfort". As a matter of fact, in everyday practice and often without considering the consequences of such actions, occupants inter-act with the building-plant system of their own homes and workplaces in order to achieve the desired environmental conditions: opening and closing windows when feeling too hot or too cold, driving shields to prevent glare or excessive illumination, regulating the temperature of the thermostat, etc... Each of these actions generates uncontrollable variables in the design phase, which deny full effect to any previsions in energy consumption of a building. As previously confirmed, occupant behaviour is affected by indoor climate, but since users are naturally prone to restore ideal comfort conditions by interacting within the environment they are living or working in, this relationship could be defined as retroactive. For instance, occupant behaviour also affect indoor climate, consequently combine to bring about large variation in energy consumption in otherwise identical housing. For instance, in 2001 Jeeninga et al. have shown that the household energy consumptions in completely identical apartments can vary as much as 100% only considering user actions forced in order to maintain a state of comfort in their homes. As a consequence of this, the weight of the occupant behaviour assumes a remarkable value within the energy performance of buildings. Additionally, further studies (Paciuck 1989, Lehman and Bordass 1999, Toftum 2009) have stressed the importance of giving the occupants the opportunity to operate directly on the internal environment control: it is shown (Brager, Paliaga and de Dear 2004, Wagner and others 2007) that users allowed to interact with control systems are more satisfied with their own living or working environments since they became likely to adapt themselves to indoor climate conditions and to tolerate greater fluctuations in acceptable temperature ranges.

Therefore, occupant behaviour is emerged as one of the main responsible of the existing discrepancy between predicted and actual energy consumption in buildings. Several researches (Sardianiou 2008, Guerra Santin and Itard 2010) have been carried out primarily in order to investigate the socio-economic implications of energy behaviour in buildings within individual and household members. Sector studies (Isaacs and others 20042, Gardner and Stem 2008, Sanquist and others 2010) have underlined that information and incentives about energy rates are less effective than expected, since occupant behaviour could vary considerably and is mostly habitual. This is probably due to the fact that individuate do not necessarily behave as "economics maximizers", but aim for the maximal level of comfort in indoor environments. Moreover, a study carried out from Andersen (2009) highlights considerable differences in energy consumptions (up to 20 times from the minimum and maximum value) among identical dwellings (age of construction, building typology), as confirmation that occupant behaviour in buildings dramatically affect any deterministic prevision of energy consumption. Individual responses of each human have different origins and depend substantially on factors such as age, sex and type of activity. However, the behaviour of each of these categories of occupants can vary greatly according to physiological and psychological differences. Moreover, occupant's level of comfort depends on several factors: physical and social issues must be considered as having equal weight in achieving occupant comfort within environments. People productivity is generally increasing whether they can reach a state of physical comfort in indoor environments: achieving a satisfactory level of thermal comfort in working places often helps in obtaining a psychologically pleasant environment, and vice-versa. Furthermore, users' emotions and stressing factors are equally relevant in assessing the condition of comfort perceived by the occupants and must be taken into account along with the physiological parameters.

In a view of these factors, in recent years there has been a shift in the direction of research related to energy and environment performance of building towards a focus on human-centre concerns. In this aim, the international project IEA Annex 53 ECBCS "Total energy use in buildings - analysis and evaluation methods" has as its primary objective a better description of total energy consumption in buildings, also taking into account the influence of users behaviour, through the clarification of the main factors, also called drivers, which literally drive, push, occupants, to the realization of a certain action in order to improve, more or less consciously, their standard of comfort within an environment, based on precedent experiences. For instance, the main six drivers recognized by the literature as influencing energy occupant behaviour in buildings are physical parameters (indoor and outdoor temperature), psychological (preferences, attitudes), biological (age, gender) environmental (type of environment where the occupants are located) and from group interaction (income, lifestyle).

As shown by this complex interaction profiles, behavioural actions can't be easily predictable, since internal and external parameters continuously interface between each other, driving at any time subjects to different stimuli. As a consequence of this, a large variation of occupants' reaction also called "action scenario" establishes a difficult task in predicting effects on indoor environment and as a consequence on energy consumptions. Several investigations have been conducted in literature in order to comprehend the most influencing drivers related to occupant behaviour in residential building. Accordingly to previous researches, is possible to highlight that users in residential building principally restore their comfort condition by opening and closing windows and adjusting thermostats. A more in deep description of drivers will be performed in this research study.

Furthermore, the international project IEA Annex 53 focuses on the development and the proposal of a new methodological approach able to understand the complex networking connection of internal and external drivers and to replicate the process leading occupants to interact with building systems and envelope in order to reach or restore a condition of comfort in a desirable indoor environment. Main objective of the project is therefore to describe occupant behaviour by probabilistic models, which necessarily take into account the possibility that an action is carried out through statistical input parameters. As a matter of fact, the new method defines behavioural models by using an algorithm based on a statistical data set built on monitoring occupants' real interaction within the building-plant system.

Main purpose of the presented research is to investigate the users' level of interaction within building control systems in order to perform a better prediction of energy consumption. The traditional approach to dynamic simulation tools considered energy consumptions as fully deterministic, on the one hand taking in to account standardized input parameters, on the other hand using fixed schedule unable to replicate the actual users' behaviour inside building. Firstly, climatic input conditions simulated during the design stage were considered as stable and do not take in to account the possible fluctuation of indoor and outdoor parameters, due to yearly climate variation and human interaction with operable control. Secondly, pre-determined schedules replicating human behaviour (lighting level, occupancy, air change rate, thermostat set-point) will never provide results that truly overlap with the actual energy consumption, since many parameters influencing human comfort (lighting, thermal, air quality) could vary significantly and unpredictably during building operational life. The method of the research is based on the assumption that only switching from a determinist approach to a probabilistic one, it will be feasible to obtain energy consumption prediction closer to reality. By following this theoretical approach the thesis is structured in five chapters focusing on five main research topics:

1- Philosophy behind statistical modelling: theoretical and empirical study of occupants' behaviour influence on energy consumption of residential buildings

The philosophy behind this new methodological approach as described in literature is basically founded on probability distribution: events in nature never replicate themselves as equal over time, on the contrary are distinguished for intrinsic fluctuations in performance and intensity. For instance, input and output are now considered as a probabilistic distribution, trying to overcome the gap between real and predicted energy consumption due to the previous widespread deterministic consideration of single and fixed value. The practical approach of the methodology is based on tour step operations:

1. Collecting real data from field measurements (objective and subjective)

2. Statistical analysis of the amount of data and definition of the most influencing parameter (coefficients and variables) on occupant behaviour (logistic regression)

3. Implementation of the logistic regression in energy simulation programs

4. Consideration of a probabilistic distribution of the output

Window opening and closing and heating set point adjustment have been considered both as probabilistic input and implemented in conjunction in a logistic regression model by using dynamic energy simulation software.

2- The development of statistical hybrid models on occupants' windows operation and heating set point adjustments based on different behavioural pattern level of interaction

The study developed an experimental approach into subsequent scenarios, adding at any further step more probabilistic variables to the models. Starting from a standardized and fully deterministic approach, the project defines hybrid models, considering all variables due to occupant's interaction with window operation and heating set point adjustments as probabilistic. Firstly, the study has treated dwelling energy performance as normally performed in the design stage of energy consumption simulation. Secondly, a model considering probabilistic the interaction between users and window opening and closing has been built, even though heating set point has been still considered in a deterministic way as a fixed input value, dependent on the comfort category. Finally, both window opening and closing and heating set point adjustments have been described through probabilistic models as logarithmic functions. Specifically, five models have been defined, and consequently five scenario of the research have been settled following an incremental philosophy in the switch from a deterministic to a probabilistic approach.

3- Influence of window opening and closing behaviour and heating set-point adjustments on heating energy consumptions in dwellings.

To estimate the influence of occupant behaviour on dwelling energy consumption, the probability equations determined by R from previous studies were implemented in the dynamic building simulation software IDA Ice [version 4.1]. Five occupants' behavioural pattern scenarios have been implemented by using a two room model consisting in living room and bedroom, developed in a previous study on window opening behaviour by Fabi, Andersen, Corgnati (2012). Aiming at getting an indication of the ability of the models to reproduce occupant behaviour interaction within building envelope and system control, a first reference model using standard window and thermostat control inputs has been implemented.

This simulation model uses deterministic input for both variables according to European Standard 15251:2006. Reference model was run for three climatic locations (Athens, Frankfurt, Stockholm) for each comfort category (I, II, III). Results of this fully deterministic model have been considered as singular (no probabilistic distribution of the output was needed) and subsequently compared with probabilistic models results. Results of this research underlines that energy consumption of dwellings in which occupants personal control (window opening and heating set point adjustments) is performed by probabilistic functions, raised up to 61%, in comparison to dwellings where the occupants interaction within building envelope and control system is regulated in a deterministic way by fixed schedules. Switching from a deterministic to a probabilistic approach, the progressive augmentation in variability is due to further addition of stochastic functions related to personal control in dynamic software.

4- Influence of window opening and closing behaviour and heating set-point adjustments on thermal comfort in dwellings.

Aim of this research was to confirm that discrepancies in energy consumption prediction are triggered by occupants' interaction with building envelope and system control, in order to restore or obtain a condition of comfort. In this effort, results of the research have been focusing in highlighting differences between predicted and actual indoor environmental quality and delivered energy consumption for heating within residential buildings. Findings demonstrated that pre-defined heating set point preference and air change rate in dwelling used in widespread standards (such as categories of comfort range acceptability of European Standard 15251:2006) are far away from actual occupants preferences in buildings. For this reason, not to consider human interaction within building envelope and control systems, will necessary lead designer and modellers to underestimation of heating delivered energy and therefore total energy performance of buildings.

Accordingly to final results, maximum variation in heating energy performance prediction has been recorded for Athens, in particularly referring to category III of Standard 15251:2006. This discrepancy could be attributed to users' interactions specifically within windows opening. As a matter of fact, in warmer climate naturally ventilated building tend to get over heated during summer periods and consequently users tend to open Windows more often. This interaction necessarily leads to an increase of ventilation losses and hence heating delivered energy. Moreover, in case dwelling occupants are allowed to interact within building envelope and system controls (window opening and heating set point adjustments), they always prefer comfort condition closer to category I. This study is therefore confirming that a gap between deterministically predicted and actual delivered heating energy consumption in dwelling is partly due to occupant actions within control systems, performed in order to restore a comfort condition in indoor environments. Moreover, the importance of personal control in reducing the need for high-energy solutions and heating delivered energy become increasingly clear in the topic of building sustainability and it is confirmed by this study

5- Investigation of different behavioural pattern leverage of window opening and heating set point adjustments on energy consumption and thermal comfort in residential buildings

The purpose of this research was to investigate the influence of different behavioural pattern of window opening and heating set point adjustments on energy consumption and thermal comfort in residential buildings. Behavioural patterns for active, medium and passive users have been combined and subsequently merged in order to simulate more accurately the variation in actual energy consumption, due to human interactions within buildings. In this effort, the study has highlighted which combination of users' level of interaction consist the most energy-saver or energy-waster behaviour in residential buildings.