Simulating the dynamics of occupant behaviour for

power management in residential buildings

Ayesha Kashifa,b, Stéphane Ploixa, Julie Dugdaleb, Xuan Hoa Binh Lea

aLaboratory of Grenoble for Sciences of Design, Optimisation and Production, Universityof Grenoble, 1

46 Avenue Felix Viallet, 38031 Grenoble - FrancebLaboratoire d'Informatique de Grenoble / University of Grenoble, 2

110, Av de la Chimie, 38400, Saint Martin d'Hères, France

Abstract

Inhabitant's decisions and actions have a strong impact on the energy con-sumption and are an important factor in reducing energy consumption andin modelling future energy trends. Energy simulations that take into accountinhabitants' behaviour are benchmarked at o�ce buildings using controlledactivity pro�les and prede�ned scenarios. In this paper we have proposed aco-simulation environment for energy smart homes that takes into accountinhabitants' dynamic and social behaviour. Based on this kind of complexbehaviour, the setpoints for di�erent controllers are adjusted in the physicalsimulator. In this platform, human behaviour is modelled using the Brahmsenvironment and the thermal model and controllers for di�erent appliancesare modelled as a physical simulator. The thermal model computes the tem-perature decrease/increase in a room based on the contextual informationresulting from the behaviour simulator. This information is then given tothe controller to act upon.

Keywords: human behaviour, modelling, simulation, multiagent system

1. Introduction

In Europe, buildings account for 40 - 45% of total energy consumption,contributing to a signi�cant amount of CO2 emissions [1]. Inhabitants' be-haviour in buildings has a strong impact on the energy consumption andis an important factor for energy waste reduction [2]. Existing behavioural

Preprint submitted to Building and Environment 31st January 2012

models that are used for prediction are based on controlled activity pro-�les and prede�ned scenarios. Furthermore they are speci�cally aimed atpredicting power consumption in o�ce buildings. Results from these mod-els cannot be extended to the residential buildings since human behaviour ismore unpredictable and complex in home settings. In this paper, we focus onresidential buildings to model and simulate behaviour dynamically througha human behaviour simulator module. This human behaviour simulator gen-erates dynamic behaviours and is connected to a physical simulator wherethe setpoints for di�erent appliances are stored and adjusted. The setpointis the target value that an automatic control system aims to attain. Changesin the environment are perceived by the agents in the behaviour simulator,who then take actions dynamically to change the state of the objects andappliances in the building. The purpose of the proposed approach is to as-sess the sensitivity of human behaviour for energy control and management.This will help in developing smart environments as well as testing the designof new buildings or houses so that they are more suited to human needs.

1.1. Energy Simulations State of the Art

The majority of works focus on energy simulations in o�ce buildings;in this section a brief overview of those works are presented. Claridge andhis colleagues [3] compiled a library of schedules and diversity factors basedon measured electricity consumption data for energy simulations and peakcooling load calculations in o�ce buildings. They derived multiple sets ofdiversity factors from measured data in 32 o�ce buildings. [4] used theoccupancy and lighting diversity pro�les and found a strong correlation be-tween these two variables through linear regression. [5] however modelledthe lighting and occupancy in buildings using a monte carlo approach basedon survey statistics of how people use o�ce spaces. [6] suggested that moreattention should be given to occupant behaviour in order to increase the ac-curacy of building thermal models in o�ce buildings. [7] proposed stochastic'lightswitch2002' algorithms to predict the manual and automated controlof lights and blinds in single and two person o�ces. These algorithms wereused in a lighting simulator called DAYSIM that demonstrated the impactof manual control on predicted lighting energy requirements. [8] proposeda, sub-hourly occupancy-based control model as a self contained simulatormodule. This integrated the stochastic lightswitch2002 algorithms into awhole building energy simulator ESP-r. [9] developed an event based pat-tern detection algorithm for sensor-based modelling and prediction of user

2

behaviour. They connected behavioural patterns (Markov model) to build-ing energy and comfort management systems through EnergyPlus simulationsoftware for energy calculations.

Simulations based on static pro�les or single user behaviour are unreal-istic. Typically in building simulators only the thermal heat generated byappliances and occupants is considered. Moreover, the occupants are con-sidered only as being present or absent, as shown in �gure 1, without takinginto account the way they behave to consume energy. In energy simulations,

Figure 1: Presence of occupants in the house

instead of drawing simple curves for the presence or absence of inhabitants,taking into account the actual behaviour of humans that a�ect energy con-sumption could help in testing building design in a more realistic way. Abetter management that coordinates and orchestrates the use of all kindsof energy according to inhabitants' needs and comfort remains an importantprogress factor. In this paper we focus speci�cally on domestic situations andmodel dynamic behaviour, which we believe is the key for reliable simulationin energy management.

3

2. The Importance of Occupants Actions Regarding Energy Con-

sumption

In order to understand how inhabitants' behaviour a�ects energy con-sumption an experiment has been performed on the IRISE database1. Inthis experiment 2 di�erent categories of houses are selected based on thenumber of occupants: 2 person houses in category 1, 5 person houses incategory 2. Also all houses in both categories have the same appliances.

It can be seen in the results below in �gure 2, that in the 1st categoryin house "H2000902", the inhabitants have the highest consumption for thewashing machine as compared to other appliances. This maybe due to theirbehaviour of frequently washing a small volume of clothes compared to wash-ing a large volume less often. In the 2nd category, it can be seen in �gure 3

Figure 2: Energy Consuming Activities: 2 person houses

1This is part of the European Residential Monitoring to Decrease Energy Use andCarbon Emissions (REMODECE) project. It contains energy consumption data, for eachappliance from 98 French houses, recorded at every 10 minutes, over a one year period.

4

that inhabitants in house "H2000945" have the highest consumption for theTV as compared to other houses possibly because of their behavioural dif-ferences with the inhabitants in other houses. The above experiments showthat the occupants' behaviours vary frequently and have a strong in�uenceon energy consumption.

Figure 3: Energy Consuming Activities: 5 person houses

3. Context and Human Behaviour Representation

The models discussed in section 1.1 were designed speci�cally for o�cebuildings where the occupants follow more or less the same routines andwhere their behaviour is not as complex and unpredictable as it is in homesettings. In addition to behaviour, context is another important factor a�ect-ing the energy related activities of occupants. "Context is any informationthat can be used to characterize the situation of an entity" [10]. [11] anal-ysed user behaviour through contextual factors including user, time, space,

5

environment and object. These authors presented a user behaviour mod-elling approach called 5W1H for: what, when, where, who, why and how,which they then mapped to a home context (object, time, space, user andenvironment). Since human behaviour is one of the most important factorsa�ecting energy utilization in buildings, it is explored in more detail. Humanbehaviour can range from being very simple to very complex. The purposehere is to capture the behaviour that not only represents a simple presenceor absence of an inhabitant in an environment but also represents a realis-tic interaction of the human with the environment. The environment doesnot only includes the di�erent objects and appliances but also includes theother inhabitants. This means that the dynamic, reactive, deliberative andsocial behaviour of inhabitants must also be taken into account in order tofully understand it's possible a�ect on energy consumption. Since behaviourvaries from one inhabitant to another, we need a way to generalize and modelit for all inhabitants, so that it is representative of actual energy consumingbehaviour in the home. Existing theories and behaviour models have beenexplored in order to fully understand and capture various aspects of humanbehaviour. This will help to consider the inhabitants as reactive, intelligentagents instead of simply "�xed metabolic heat generators passively expe-riencing the indoor environment" [6]. The term "behaviour" refers to theactions or reactions of an object, usually in relation to its environment. Inthe literature, perception, cognition, memory, learning, social and emotionalbehaviour and psychomotor are considered to be the basic elements of humanbehaviour [12, 13, 14].

Cognition and the organization of knowledge within humans is capturedby many behaviour representation models. [15] proposed ACT (Atomic com-ponents of thoughts) in order to represent human behaviour. ACT mainlyfocuses on human cognition and shows how humans organize their knowl-edge in order to behave intelligently. [16, 17] proposed CAPS (Concurrentactivation based production system) which is a production system whoseprocedural knowledge consists of productions, each specifying the conditionsand consequent actions. [18] proposed COGNET (Cognition as a networkof tasks) that mainly focuses on cognitive behaviour of humans, which ismodelled by assuming that humans are capable of performing multiple taskssimultaneously. [19, 20] proposed CCT (Cognitive complexity theory) as amodel of cognition that is based on the concept of GOMS, which modelshuman performance as Goals, Operators, Methods and Selection rules. [21]proposed DCOG (Distributed cognition), which argues that cognition is not

6

con�ned into an individual rather it is distributed across the environment.In addition to human cognition many of the behaviour models also focus

on the perceptive and motor processes within humans. [22] proposed EPIC(Executive process/interactive control) model that focuses on the perceptual,cognitive and motor processes. It also captures another important factor ofhuman performance which is multitasking. [23, 24] proposed APEX (Archi-tecture for procedure execution), to model human behaviour in a complexand dynamic environment. It makes an abstract sketch of future actionsand �lls out a plan in the form of procedures as soon as the information isavailable and manages the tasks accordingly. [13] proposed Coga� (Cogni-tion and a�ect project) which is a human information processing architecturedivided into three levels, reactive, deliberative, and re�ective. [14, 25], pro-posed Brahms (Business redesign agent-based holistic modelling system) as amodelling and simulation environment for analyzing human work practices.It is able to represent people, things, places, behaviour of people over time,tools and artifacts used etc.. It also focuses on the social behaviour. [12]proposed SOAR (State operator and result), according to which behaviour iscaptured as a search or movement through the problem space at a particulartime and a goal state which represents a solution for the problem.

There are many Human behaviour representation models that take intoaccount the human-system interactions. [26] proposed HOS (Human opera-tor simulator) which provide a model of human capabilities and limitations tosupport the design of human-machine systems. [27] proposed OMAR (Op-erator model architecture), in this model human behaviour is modelled asinteractions among agents. These agents can represent di�erent people ordi�erent functions within a single person. But there is no central executiveor scheduler that controls these parallel activities.

The above studies have suggested that the behaviour representation mod-els capture many characteristics of humans, such as their observable actions,decision making and cognitive abilities and single and group behaviour. Mostof the behaviour models discussed above capture the reactive and delibera-tive behaviour of humans, however, few of them capture the social behaviouras well, including HOS, OMAR, SOAR, and Brahms. Also keeping in viewthe context elements important for energy control and management, Brahmsmodelling and simulation environment is one which is able to simulate theinhabitants as agents. Theses agents interact not only with the objects andappliances at a particular location and at particular instance of time but alsowith other agents in the environment.

7

4. Modelling the Household Behaviour

Human behaviour has been described in terms of physical needs: " Thereare certain physical needs that people must meet in order to survive. Thereare other needs that make people more comfortable. In the speci�c ways theystrive to meet these needs, people are di�erent" [28]. Inhabitant behaviourat home is considered to correspond to the inhabitants' activities with a pur-pose to satisfy their needs according to their contexts. The questionnairepresented in �gure 4 was used to collect information concerning the contextelements in home situation. In home situations we have the information

Figure 4: Questionnaire for collecting the context and information of inhabitant behaviourat home

on energy consumption per appliance by the household but the informationabout the activities of people behind the consumption of home appliances ismissing. This study was conducted in order to assess human behaviour inhome situations by capturing their activities. The above questionnaire, gath-ered information on the various types of behaviour, the needs of inhabitantsand the way inhabitants use household objects to satisfy the needs. Somephysical needs of the inhabitants have been identi�ed (e.g. drinking, eating,going to the toilet, sleeping, taking a bath, dressing, etc). Each inhabitanttends to repeat the behaviour that has been successful in satisfying theseneeds. This repetition becomes a behaviour pattern and forms the daily rou-tines of inhabitants. Routines are used "as a resource for e�ciently organisingindividual and collective activities at a low cognitive cost" for the design ofdomestic situations [29]. These behaviours can be modelled and simulatedby a stochastic process with an approximated timetable. However, for evalu-ating possible power management solutions, not only the time, duration andlocation of the daily activities are necessary but also the detailed informationabout which and how domestic electric appliances are used in these activitiesare needed. For example, an inhabitant wants to have dinner; he goes to the

8

kitchen and prepares the dinner by warming food in microwave for 30 secondsat 500 Watts, cooking food on hot plate for 10 minutes at its maximal powerand then eating the meal in 5 minutes; during all this time period, he turns on100 Watts light in the kitchen. The information about power consumption ineach period of this behaviour is necessary for evaluating power managementsolutions. The inhabitants' behaviour for satisfying environmental comfortneeds (e.g. thermal comfort, visual comfort, etc.) is also important and hasto be considered. Also these comfort levels are individual and vary fromone inhabitant to another. These behaviours are not triggered at regulartimes. They depend solely on the value of some environmental factors whichis one of the context elements. When the physical state of the environmentexceeds the inhabitant comfort tolerances, it causes a psychological state inthe inhabitant. This prompts the inhabitant to perform certain activities toadjust his or her environment. For example, an inhabitant enters a room;the room's temperature is higher than 30◦C; the inhabitant believes that heis feeling hot and wants to open window or turn on the air-conditioning tolower the room temperature. These behaviours can change power consump-tion at home, hence, it is necessary to model and simulate these behavioursfor evaluating power management solutions. If there are many inhabitants athome, an inhabitant can ask others to perform activities to satisfy his need.For instance, the inhabitant can ask someone else in the same room to turnon the air conditioner. Such type of behaviours have also to be modelled.For modelling these various types of behaviours and needs of inhabitants, acausal model of inhabitants' behaviour is proposed and presented in detailin the next section. A causal model is an abstract model that uses cause ande�ect logic to describe the behaviour of a system [30].

4.1. Example of Household Behaviour Model

An inhabitant living at home has some needs. To satisfy his need, he cando one or many activities. To do an activity, he can use one or several house-hold objects. The causal model representing these relations is presented in�gure 5. This model shows that an activity can cause other activities. Anexample of this relation is when an inhabitant prepares a meal, he needs toprepare ingredients and cook the food. These actions cause state changes(e.g. turn on, turn o�, open, close, etc.)in household objects or appliances.In �gure 6, two additional causal inputs of inhabitant needs are introducedinto the model: usual time and environmental factors. When the usual timefor some action comes up, it may cause an inhabitant need (e.g. to get up,

9

Figure 5: Causal model of inhabitant behaviour to satisfy a need

Figure 6: Causal model evolved of inhabitant behaviour

to go to work, to have dinner, to go to sleep etc.). When an environmentalfactor changes and exceeds the inhabitants' comfort tolerances, it causes aninhabitant comfort need to change. Both the usual time and the environmen-tal factors constitute the inhabitant's context at home. The change of othercontext elements (inhabitant, space and object) can also cause an inhabitantneed to change. For example, when a visitor is present, inhabitant may needto prepare a meal for the visitor. The context elements are considered as anexternal cause, coming from the environment whereas inhabitants' psycho-logical state is considered to be an internal cause that triggers an inhabitants'need. The complete causal model of inhabitants' behaviour is presented inthe �gure 7 2.

5. Brahms Modelling Language

A descriptive language to record and simulate the causal relations ofinhabitant behaviour is needed. The Brahms language [14] is compatiblewith our requirements. It is a full-�edged multi-agent, rule-based, activityprogramming language. An agent based approach is needed as agents can

2In �gure 7, CC stands for the causal condition: if a cause is satis�ed, an e�ect iscreated. In the case of many inhabitants, a need of an inhabitant can cause not onlypersonal activities but also activities of other inhabitants. For instance, in a family theparent asks their children to go to the table to take dinner all together.

10

Figure 7: Complete causal model of inhabitants' behaviour at home

have needs, they can perform certain activities based on their needs and canalso communicate for the ful�lment of various needs. It has similarities tobelief-desire-and-intention (BDI) architectures and other agent-oriented lan-guages, but is based on a theory of work practice and situated cognition.The notion of work practice includes how people behave in situations, atspeci�c moments in the real world. Situated cognition claims that "everyhuman thought and action is adapted to the environment, that is, situated,because what people perceive, how they conceive of their activity, and whatthey physically do develop together" [31]. It has an activity subsumptionarchitecture which can model an activity that causes other activities. For allthese reasons, the Brahms language is chosen to describe our causal model.Its corresponding simulator is also used to interpret the model in each con-crete example. Figure 8 presents the transformation of the causal model intothe Brahms language.

11

Figure 8: Transformation of the causal model into the Brahms language

5.1. Results in Brahms Simulation Environment

In �gure 8, TF and WF correspond respectively to the thoughtframe andworkframe in Brahms language. Thoughtframes are used to model the rea-soning behaviour of agents and are represented as production-rules creatingnew beliefs of agents or objects whereas workframes (rule-based) performagents and objects activities (simple or composite). The outside context andthe inside psychological state is perceived by the agents as a thoughtframe,which becomes its need. This need is accomplished by the workframe throughprimitive or composite activities. Figure 9 presents the simulation results ofa simple example of inhabitants' behaviour. This example introduces twoinhabitant behaviours: in the upper part of �gure 9, when the temperaturerises, one inhabitant communicates with the other to agree on opening thewindow. The vertical lines show the messages exchanged between the twoagents. After receiving the message from the �rst agent, if the second oneagrees then the �rst one will proceed to open the window. In the lower part of�gure 9, when it is time for dinner, one agent is already present in the kitchenand prepares food, then moves from the kitchen to the living room where itasks the other agents to have dinner together, they then have dinner while

12

Figure 9: Simulation result of an example of inhabitants' behaviour

watching TV in the living room. The horizontal line below the primitiveactivity3 means that some appliance is used in performing this activity.

6. A Multi-simulator environment

This section describes a co-simulation environment to improve energymanagement in buildings. In this simulation environment, there are �ve mod-

3In this �gure, wf stands for workframe, ca for composite activity and pa for primitiveactivity.

13

ules: two modules with models describing the thermal and electrical aspectswithin the building; a module dedicated to control algorithms and energy-saving policies; the module presented earlier on in this paper that deals withthe simulation of inhabitants behaviour and a �fth module for predictingthe outdoor environment. The interoperability between these modules ispresented in �gure 10.

Figure 10: Interoperability between the di�erent modules in a co-simulation environment

6.1. Coupling the physical simulator with the User Behaviour Simulator

The activities of inhabitants, their presence at di�erent locations in thehouse, their control over di�erent appliances and objects, and their communi-cations can be modelled in the Brahms simulation environment. However inorder to model the environmental variables a physical simulator is requiredthat provides the information about physical aspects such as temperatureinside di�erent parts of the house, humidity and outside weather conditions,etc. The values of these environmental parametrs are fed into the Brahmssimulation environment, causing the inhabitants to change their beliefs andperform some activity. In order to perform certain activities, the inhabitantschange their locations, change the states of di�erent objects (e.g. doors, win-dows), and di�erent appliances (e.g. heater, air conditioner) in the house.As soon as these state changes happen, this information is sent back to thephysical simulator where new setpoints for the environmental parameters,such as the temperature, are adjusted. This process continues in a cycle and

14

the impact of all types of inhabitants' behaviour over energy consumption iscaptured.

6.2. Connection between the Simulators

The connection between the occupant's behaviour simulator (Brahmssimulation environment) and the physical simulator, is established throughjava interface. The physical simulator is created in matlab/simulink and con-sists of the thermal model of the house and the controllers for appliances.This interface actually drives Brahms virtual machine and manipulates dif-ferent attributes of the occupant's behaviour model to be simulated. Thisis done by setting agents and objects attributes and handling the startingtime of the simulation. It keeps track of the current location of agents andof the current values of di�erent attributes of objects. In addition, the in-terface is also responsible for the synchronization between Brahms and mat-lab/simulink, it veri�es the termination of a simulation step and advancesthe Brahms virtual machine to the next step of simulation and process andprepares the data to be exchanged between Brahms and Matlab/Simulink(data to be exchanged between occupant behaviour simulator and physicalsimulator). This interface is utilized in matlab/simulink by compiling it intoa jar �le and further giving the path of this �le and further calling its built-infunctions in the level-2 matlab s-function. The thermal model is de�ned inthe matlab function �le which uses the output of the Brahms simulation suchas presence of occupants, appliance and window state and based on the out-side temperature and the heating appliance's power gives the temperatureinside the room. The heating appliance's power is calculated by the con-troller which maintains the setpoint temperature for the room. The �gure11 shows how the co-simulation works.

6.3. Application Example

The combined architecture of the physical simulator connected to theBrahms simulation environment is shown in the �gure 12. An example ofthe application, being run in matlab/simulink, and the results are shown in�gure 14 and 15. The physical simulator, consisting of the thermal modelof the house and the controllers for air conditioner and heater is connectedto the Brahms simulator. In the �gure 12 only the thermal model of theliving room is considered. The information about the temperature inside theliving room is sent to the Brahms simulator. If an agent is present in theliving room, it will continuously perceive the environmental temperature and

15

Figure 11: How the co-simulation works

as soon as the temperature value exceeds or falls below a speci�c setpoint itwill perform the appropriate action. Figure 12 shows the physical simulatorwhich stores the values for the environmental variables. In this scenario theenvironmental variable considered is the temperature in di�erent rooms insidethe house. This physical simulator when connected to the Brahms simulationenvironment adjusts the temperature in di�erent rooms. The inhabitants aremodelled as agents moving throughout the house. We will now describe asimple scenario to see how the co-simulation works and how the decisionstaken by the inhabitants a�ect the energy consumption. Consider a simplescenario consisting of a 4 person house, husband, wife and their 2 children.Figure 13 shows that at the start the 2 children are present in the bedroom2,the husband in bedroom1 and the wife in the living room, after some timethe husband moves to the bathroom and then to the kitchen and �nally tothe living room. Inhabitants' beliefs are changed based on the perceivedenvironmental values and they perform di�erent activities accordingly. Theseactivities when performed a�ect the appliances or objects present in thehouse. For example, if the temperature in the physical simulator for theliving room is adjusted to a value lower than a speci�c setpoint, say 20◦C,the inhabitant will be comfortable with this, but as soon as the temperature

16

Figure 12: Combined architecture of brahms and physical simulator

Figure 13: Movements of Inhabitants in di�erent locations in the house

value exceeds 20◦C, he will start feeling hot and will react in some way or theother to lower the temperature. The husband and wife have di�erent ways of

17

Figure 14: Communication between the inhabitants

lowering the temperature. The wife being mindful of cleanliness prefers notto open the window, especially when the weather is windy. Instead she alwaysprefers turning on the air conditioner. On the other side, the husband beingworried about the electricity bills, prefers opening the window, whatever theoutside weather condition. If however, they both are present in the roomand temperature goes up, both of them make certain compromises becauseof the social in�uence of the other. In this case, the husband, who alwaysprefers opening the window, will take into account the weather and if it is�ne, he will disregard his wife's opinion and open the window. However ifthe weather is windy or stormy, he will not want to annoy his wife and willask her permission �rst to open the window. If she disagrees he will turnon the air conditioner in order to lower the temperature. From �gure 12the temperature outside is 30◦C and the weather is windy, represented bythe value of 2. The inhabitants in the living room will react as mentionedin the scenario. Since the weather is windy, the wife did not agree to openthe window and the husband turns on the air conditioner. The negotiationbetween the inhabitants in shown in the matlab command window in �gure

18

Figure 15: Window's status is not changed, Inhabitant has turned on the air conditioner

14. As a result of this communication between the 2 agents, the �nal decisionmade by the agent is to turn on the air conditioner. In �gure 15, it can be

Figure 16: Living room's temperature and setpoint for air conditioner

seen that the window is closed and the air conditioner is turned on, which isthe result of the weather outside and the negotiation between the inhabitantsin lowering the living room temperature.

Figure 18 shows the graphical model for the above scenario according tothe formalism in �gure 8.

From �gure 16 we can see how the physical simulator adjusts the tem-perature inside the room to a value where the inhabitants feel comfortable,20◦C in this scenario. If however, the outside temperature falls below the

19

Figure 17: Living room's temperature and setpoint for heater

Figure 18: Graphical model of the household

setpoint, the physical simulator will start using the heater controller to heatthe living room to the setpoint temperature as shown in �gure 17.

20

7. Conclusion and future works

In this article we have proposed a dynamic behaviour model to simulateinhabitants' non trivial decisions and actions in residential buildings thatdirectly impact the energy waste. Furthermore a co-simulation platform hasbeen proposed that takes into account dynamic inhabitants' behaviour andcaptures energy loss resulting from inhabitants' behaviour using a physicalsimulator. We have successfully demonstrated that dynamic behaviour iscrucial for an accurate energy simulation in order to predict energy trendsand reduce energy waste consumption. Simulations results demonstrate thebene�t of our approach in terms of optimized power savings. It can be usedin smart homes to optimize power savings while maintaining inhabitants'comfort levels. The work presented in this paper di�ers from work usingprede�ned static scenarios because changes in the thermal model of the houseor in the weather condition in the simulation also change the inhabitants'beliefs; this �nally a�ects their decisions to act upon certain objects andappliances in the environment. Also by introducing the social element it isa more realistic representation of how actually people behave in daily lifewithin home settings.

8. Acknowledgement

This work has been performed in the scope of the SUPERBAT andSIMINTHEC ANR projects with participation from Électricité de France(EDF). The authors also want to acknowledge public research institutions:G-SCOP and LIG.

References

[1] P. Huovila, Building and Climate Change: Status, Challenges and Op-portunities, Paris, France: United Nations Environment Programme,Sustainable Consumption and Production Branch, 2007.

[2] W. F. Van Raaij, T. M. M. Verhallen, A behavioral model of residentialenergy use, Journal of Economic Psychology 3 (1) (1983) 39-63.

[3] D. E. Claridge, B. Abushakra, J. S. Haberl, A. Sreshthaputra, Electric-ity diversity pro�les for energy simulation of o�ce buildings , ASHRAETransactions 110(1).

21

[4] B. Abushakra, D. Claridge, Accounting for the occupancy variable ininverse building energy baselining models, in: Proceeding of the Interna-tional Conference for Enhanced Building Operations (ICEBO), Austin,2001.

[5] L. O. Degelman, A model for simulation of daylighting and occupancysensors as an energy control strategy for o�ce buildings (1999) 571-578.

[6] G. Newsham, Manual control of window blinds and electric lighting:Implications for comfort and energy consumption, Indoor Environment3 (1994) 135-144.

[7] C. Reinhart, Lightswitch 2002: a model for manual control of electriclighting and blinds, Solar Energy 77 (1994) 15-28.

[8] D. Bourgeois, C. Reinhart, I. Macdonald, Adding advanced behaviouralmodels in whole building energy simulation: A study on the total energyimpact of manual and automated lighting control, Energy and Buildings38 (2006) 814-823.

[9] B. Dong, B. Andrews, Sensor-based occupancy behavioral pattern recog-nition for energy and comfort management in intelligent buildings, in:7th International IBPSA conference, Glasgow, Scotland, 2009.

[10] A. K. Dey, Understanding and using context, Personal Ubiquitous Com-put. 5 (2001) 4�7.

[11] T. S. Ha, J. H. Jung, S. Y. Oh, Method to analyze user behaviour inhome environment, Personal and Ubiquitous Computing 10 (2006) 110-121.

[12] J. F. Lehman, J. Laird, P. Rosenbloom, A gentle introduction to Soar,an architecture for human cognition, Vol. 4, MIT Press, 1996.

[13] A. Sloman, Varieties of A�ect and the CogA� Architecture Schema, in:C. Johnson (Ed.), Proceedings Symposium on Emotion, Cognition, andA�ective Computing AISB'01 Conventions, 2001, pp. 39-48.

[14] M. Sierhuis, W. J. Clancey, R. van Hoof, Brahms - a multiagent model-ing environment for simulating work practice in organizations, Interna-tional Journal of Simulation and Process Modelling 3(3) (2007) 134-152.

22

[15] J. R. Anderson, C. J. Lebiere, The atomic components of thought, Mah-wah, NJ: Lawrence Erlbaum Associates, 1998.

[16] R. Thibadeau, M. A. Just, P. A. Carpenter, A model of the time courseand content of reading, Cognitive Science 6 (2007) 157-203.

[17] M. A. Just, P. A. Carpenter, S. Varma, Computational modelling ofhigh-level cognition and brain function, Human Brain Mapping 8 (1999)128-136.

[18] W. W. Zachary, J. M. Ryder, J. H. Hicinbothom, Making decisionsunder stress: Implications for individual and team training, Washington,DC:APA., 1998, Ch. Cognitive task analysis and modeling of decisionmaking in complex environments, pp. 315-344.

[19] S. K. Card, T. P. Moran, A. Newell, The psychology of human-computerinteraction, Hillsdale, NJ: Lawrence Erlbaum Associates, 1983.

[20] D. Kieras, P. G. Polson, An approach to the formal analysis of usercomplexity, Man-Machine Studies 22 (1985) 365-394.

[21] R. G. Eggleston, M. J. Young, K. L. McCreight, Distributed cognition:A new type of human performance model, in: AAAI Fall Symposiumon Simulating Human Agents, North Falmouth, MA, 2000.

[22] D. E. Kieras, D. E. Meyer, An overview of the epic architecture for cogni-tion and performance with application to human-computer interaction,Tech. rep., MI: The University of Michigan (1995).

[23] M. A. Freed, Simulating human performance in complex, dynamic envi-ronments, Ph.D. thesis, Northwestern University, IL USA (1998).

[24] R. J. Firby, Adaptive execution in complex dynamic worlds, Ph.D. the-sis, Yale University, New Haven, CT (1989).

[25] W. J. Clancey, P. Sachs, M. Sierhuis, R. van Hoof, Brahms: Simulatingpractice for work systems design, Energy and Buildings 49 (1998) 831-865.

[26] R. Wherry, Monitoring behavior and supervisory control, New York,NY: Plenum Press, 1976, Ch. The Human Operator Simulator-HOS,pp. 283-293.

23

[27] S. E. Deutsch, J. MacMillan, N. L. Camer, Operability model archi-tecture: demonstration �nal report (al/hr-tr-1996-0161), Tech. rep.,Wright-Patterson AFB, OH: Human Resources Directorate, LogisticsResearch Division. (1997).

[28] M. E. Milliken, Understanding human behavior: a guide for health careproviders, The Free Press, 1965.

[29] P. Salembier, J. Dugdale, M. Frejus, Y. Haradji, A descriptive model ofcontextual activities for the design of domestic situations, in: EuropeanConference on Cognitive Ergonomics: Designing beyond the Product� Understanding Activity and User Experience in Ubiquitous Environ-ments, ECCE '09, VTT Technical Research Centre of Finland, VTT,Finland, Finland, 2009, pp. 13:1�13:7.

[30] K. D. Anthony, Introduction to causal modeling, bayesian theory andmajor bayesian modeling tools for the intelligence analyst, Tech. rep.,USAF National Air and Space Intelligence Center (NASIC) (2006).

[31] W. J. Clancey, Situated cognition: on human knowledge and computerrepresentations, Cambridge University Press, 1995.

24