Study of Human Activity Related to Residential Energy Consumption

Using Multi-level Simulations

Thomas Huraux

1,2,3

, Nicolas Sabouret

and Yvon Haradji

EDF Research & Development, Clamart, France

LIMSI-CNRS, UPR 3251, University of Paris-Sud, Orsay, France

LIP6, CNRS UMR 7606, - Pierre and Marie Curie University, Paris, France

Keywords:

Multi-agent Systems, Multi-level Modeling, Simulation, Electrical Consumption.

Abstract:

In this paper, we illustrate how multi-agent multi-level modeling can help energy experts to better understand

and anticipate residential energy consumption. The problem we study is the anticipation of electricity con-

sumption peaks. We explain in this context the beneﬁt of the coexistence of microscopic (human activity)

and macroscopic (social characteristics, overall consumption) levels of representation. We present brieﬂy the

SIMLAB model (Huraux et al., 2014) that extends the SMACH simulator (Amouroux et al., 2013) with coex-

isting levels on different modeling axes. We then present a model of the households activity and its electrical

consumption consistent with energy experts’ observations in the residential sector. We show the impact of dif-

ferent social factors, such as individual sensitivity to price or to personal comfort, on the apparition of peaks on

the consumption. We illustrate the contribution of multi-level modeling in the understanding of macroscopic

phenomena.

1 INTRODUCTION

As energy consumption increases, one major issue for

electricity suppliers is to adapt in real time to cus-

tomer demand to maintain a stable frequency. In

this context, understanding and analyzing inhabitant

behavior is a major issue for the reduction of en-

ergy consumption. Two solutions are possible to

study and anticipate human behavior. First, experi-

ment in real situations allows psychologists and ex-

perts in ergonomics to study human activity around

the principles of action and situated cognition (Re-

lieu et al., 2004). Second, simulation can, at the cost

of some computer simpliﬁcations, reproduce related

consumption phenomena (Kashif et al., 2012; Mura-

tori et al., 2013; Amouroux et al., 2013). We seek to

assemble these two approaches by directly integrat-

ing in situ knowledge into simulations to provide new

tools for experts to connect behaviors and electrical

consumption. To this purpose, we propose a multi-

level simulation model for the study of human activ-

ity both at the microscopic level (individuals) and at

a macroscopic level (household) and we show how it

can be used to study household energy consumption.

In our model, three dimensions are considered:

the temporality of human activity (from actions to

habits), the diversity of populations (from individu-

als to social groups) and the complexity of the en-

vironment (from electrical appliances to residential

area). We extend the SMACH simulator

, a mod-

eling and simulation platform (Haradji et al., 2012;

Amouroux et al., 2013) for the study of human be-

havior which allows energy experts to study the ev-

eryday life of households in relation to their electrical

consumption, and in particular, how organization of

household activities impact on energy consumption.

We add micro-macro dynamics in these three dimen-

sions and we will show in this paper how studying

back and forth between levels of representation can

enable energy experts to better understand the phe-

nomena related to consumption and human activity.

In section 2, we present works that are interested

in the simulation of human activity and multi-level

simulations. Section 3 brieﬂy present SIMLAB, our

multi-level agent-based model based, among others,

on an inter-level inﬂuence mechanism and the concept

of modeling axis. Section 4 presents the representa-

tion of households with our model. After presenting

our various action strategies, we present different ex-

periments and the corresponding results in section 5.

http://www.youtube.com/watch?v=DViBg3-crxM

133

Huraux T., Sabouret N. and Haradji Y..

Study of Human Activity Related to Residential Energy Consumption Using Multi-level Simulations.

DOI: 10.5220/0005197401330140

In Proceedings of the International Conference on Agents and Artiﬁcial Intelligence (ICAART-2015), pages 133-140

ISBN: 978-989-758-073-4

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

2 RELATED WORK

2.1 Human Activity

In the residential sector, a Markov process calibrated

using time-use data can be used to simulate an av-

erage household (Muratori et al., 2013). The limita-

tion of this approach applied in the residential sector,

is to be based on the use of human behavior mod-

els as average households. But as pointed out by

(Morley and Hazas, 2011), there is no average house-

hold in our society in the context of energy consump-

tion. To overcome this limitation, (Grandjean, 2013)

uses inhabitant and activity proﬁles based on stochas-

tic models, which introduces some diversity in the be-

havior. This approach allows to reconstruct a load

curve in the residential context and to observe over-

all phenomena, but it does not allow a micro vision

for studying practices of everyday life.

The work of (Lee et al., 2011) relies on a micro

modeling of activities in the professional context and

use simulation to optimize the building performance.

They use simulation to make prediction on real sys-

tems, combining individual activities (e.g. opening

windows) with more macro information (e.g. holiday

periods, meeting habits, ...). In the SMACH model,

we want to go further in the modeling of activities

and interactions between individuals, which naturally

leads us to turn to multi-agent simulations.

2.2 Multi-agent Systems

A large part of human activity results from coopera-

tion, knowledge sharing, negotiation and adjustments

within interactions (Haradji et al., 2012). This is

the kind of dynamic between intrinsic properties, ac-

tions and interactions that can be found with multi-

agent systems (MAS). For instance, this paradigm is

used by (Yang and Wang, 2013) to controls the build-

ing environment through installed actuators with data

from sensors and occupants as inputs. The MAS pre-

dicts occupant preferences through learning their be-

haviors Nevertheless in simulation models, although

some agent based models have studied the electricity

market, such as (Zhou et al., 2007), little work has

been done on the simulation of human activity. Or the

models are focused on one aspect of the activity such

as comfort (Alfakara and Croxford, 2014).

Kashif (Kashif et al., 2012) works in the residen-

tial context to design a detailed model of human be-

havior for energy management. The authors also de-

part from the premise that human behavior is an im-

portant element of energy consumption of buildings.

Based on the analysis of activity logs, they propose an

inhabitant behavior model based on BDI agents (Rao

et al., 1995). The model requires precise frameworks,

which is difﬁcult to implement with domain experts

(not computer scientists) and that does not always

seem necessary (Haradji et al., 2012). That is why

we propose in this paper an approach that combines

a microscopic description of human activity with the

use of domain expert macroscopic knowledge, as the

phenomenon of overall consumption peaks. We pro-

pose in our work to use a multi-level approach.

2.3 Multi-level Simulations

Representation of macro phenomena in MAS can be

done using multi-level models (Morvan, 2012). How-

ever, most models do not consider the coexistence of

different levels within the same model during simu-

lation. Some models combine different levels, as in

hierarchical models like SWARM (Minar et al., 1996)

where macro levels take control of micro levels. Sim-

ilarly, in the crowd simulations (Navarro et al., 2011)

macro entities are seen as aggregations to speed up

simulations. The system automatically selects, for

each agent, the appropriate level of representation to

allow a signiﬁcant computation gain and one level is

enabled at a time. In all these approaches, the micro

and macro levels do not co-exist within a single sim-

ulation.

Among more generic approaches, AA4MM (Ca-

mus et al., 2013) proposes to combine levels with a

multi-modeling approach: coordination is done us-

ing information sharing. The interaction between the

different levels is limited as the levels correspond to

heterogeneous models. PADAWAN (Picault et al.,

2011) allows the co-existence of agents and environ-

ments corresponding to different spacial and temporal

scales in the IODA model (Kubera et al., 2011). All

concepts are represented by agents that can change

level (i.e. environment) based on their activity. The

IRM4MLS approach (Morvan et al., 2011) offers

a meta-model based on the principle of inﬂuence-

reaction (IR) where the inﬂuence represents the “de-

sire” of the agent to modify its environment and the

reaction represents the consequence according to the

state of the world. The links between levels are repre-

sented by directed graphs and allow to have a partic-

ular temporality for each level.

In our work, we share the view that a more sys-

temic approach (i.e. coexisting levels) can enhance

our model by extending it with macro entities which

is not allowed by approaches based on aggregation

with average behaviors. We retain the Picault’s idea

that all concept must be represented by an agent and

Morvan’s idea of representing the inﬂuences between

ICAART2015-InternationalConferenceonAgentsandArtificialIntelligence

134

levels. However, we are not interested by different

time scales or level changes (i.e. conceptualization by

energy experts), but we want to take into account the

different levels in the system modeling itself and to

combine models from different domains. We seek to

show that the direct introduction into the model of lev-

els that coexist during the simulation, not as different

visualizations of a same phenomenon but to connect

expert knowledge within a model to improve it. It will

help us to study large-scale phenomena. So we pro-

posed the SIMLAB model that we will brieﬂy present

in the next section.

3 THE MULTI-LEVEL MODEL

The SIMLAB model

(Huraux et al., 2014) is based

on two major proposals for modeling and studying

multi-level phenomena. First, agents that share com-

mon features for the study are grouped within mod-

eling axes that capture the representation of cross do-

mains properties at different levels. Second, we dis-

tinguish between interactions of agents (i.e. the ex-

change of information or queries) and intra-axis in-

ﬂuences on properties, which represent the inter-level

dynamics of these cross domains components.

3.1 Modeling Axis

Our model is positioned in the context of multi-expert

modeling and simulation. In each domain of exper-

tise, shared properties can be identiﬁed. But also, in

each domain of expertise, these notions appear with

different levels of abstraction. For example, the en-

ergy consumption is a shared property of the axis

grouping appliances, rooms, housings. This is why

we deﬁne modeling axes. These particular properties

make the modeling process easier by encouraging re-

ﬂection on what is common to all levels.

Let Ω be the set of agents. An axis χ

⊆ Ω is

associated with the concept c of the studied system

and a set of shared properties P

speciﬁc to the axis,

which represents the domain. And in each axis, we

then deﬁne the agents representing the different lev-

els. For example the heater, the room, the housing, all

calculate their energy consumption in the consuming

environment axis. We denote @ an inter-level rela-

tionship between two agents. An axis χ

form a con-

nected subset of Ω for the relationship @. We assume

that the inter-level relationship is non-transitive and

acyclic, but an agent can have several super-agents.

SIMLAB Is Multi-Level Agent Based

3.2 Agent Description

Each agent ω ∈ Ω has certain level of abstraction

in a modeling axis. It is deﬁned by the tuple

hΩ

sup

Ω

sub

, P , A, I , F i where ω

sup

and ω

sub

represent

the access of the agent to his direct super and sub-

agents. P is the set of internal variables manipulated

by the agents, called properties. These concepts cap-

ture the characteristics of the modeled entity. A prop-

erty can be atomic (real, boolean, ...) or more com-

plex (list, set, function, or even a reference to another

agent). A is the set of internal actions to enable the

agent to modify its properties. A precondition is as-

sociated to each action, and at each step of execution,

an agent performs all actions whose preconditions are

satisﬁed. Finally, I and F are the sets of interactions

and inﬂuences of the agent (see 3.3).

3.3 Inter-agent Relationships

Interactions are at the heart of agent based modeling.

However, these are not sufﬁcient when entities of sev-

eral levels are added into the model. Therefore, we

also introduce inter-level inﬂuences to describe the re-

lations between levels within a modeling axis.

3.3.1 Interactions

Our mechanism of interaction between agents is a

simpliﬁed version of the intentional communication

model proposed by FIPA (FIPA consortium, 2003).

Performatives used by agents are modeled speciﬁc

concepts and may involve several axes (for example,

agents-individuals use performative action to inform

each other of their desire and their availability for the

realization of common tasks, as we will see in the

next section). Each agent is provided with a set of

reactions R (ω), i.e. actions triggered only by interac-

tions. Like actions, reactions act on the internal prop-

erties and are associated with pre-conditions. Every

agent ω has a set of interactions I (ω), where each

i ∈ I (ω) is a tuple htarget, reactionsi with target ∈ Ω

the recipient and reactions ⊆ R (target) a set of reac-

tions by the recipient.

Example: Let n and h be two agents ∈ Ω corre-

sponding to an individual and a heater, h.temp ∈ P (h)

the temperature property of h and increase ∈ I (n) the

possible interaction of n on h deﬁne as :

increase = hh, {h.temp := h.temp ∗ 1.1}i

The operator “:=” describe the effect of the action,

given a value ω. p := v means that v is allocated to

the property p of the agent ω. Here, if the individual

performs the increase interaction on the heater, it will

increase the temperature by 10%.

StudyofHumanActivityRelatedtoResidentialEnergyConsumptionUsingMulti-levelSimulations

135

3.3.2 Inﬂuences

Levels are linked by an inﬂuence function that change

the value of a super or sub-agent property, based on

some properties of the inﬂuencing agent. These in-

ﬂuences are deﬁned in the agents properties and not

on instances of these properties. For example, if a

property of an individual inﬂuences a group, all the

individuals in this group beneﬁt from this inﬂuence.

An inﬂuence f ∈ F is characterized by a tuple

hω

, ω

, P

src

, p, in f li. It allows an agent ω

to inﬂu-

ence the value of a property of one of its super or

sub-agents called ω

, based on some of its properties.

src

⊆ P (ω

) is the set of properties of ω

used to

compute the inﬂuence on ω

, p ∈ P (ω

) the changed

property and in f l the inﬂuence function which change

the property. We denote f (P

src

) the value that ω

must

integrate in p (with the relation :=).

Example: Let n be a sub-agent of f . They cor-

respond to an individual and its family. t ∈ [0, 1] is an

individual’s property which represents its tendency to

increase the heater. prio ∈ {co, sp} describes if the

family gives priority to its comfort or its spending.

We deﬁne f ∈ F the following inﬂuence:

h f , n, prio, t,



t ∗ 0.5 prio = sp

t prio = co

Here, the individual’s tendency to increase the heater

is halved when its family give priority to spending.

The introduction of inﬂuences allow us to deﬁne

speciﬁc properties called recursive. They are proper-

ties deﬁned for all the agents of an axis and the in-

ﬂuence function is directed from sub-agents to super-

agents (n.b. recursive properties are also shared prop-

erties). A change of value on a recursive property

spreads as a recursive function whose computation is

from the agent to the sub-agents. This mechanism al-

lows the modeler to establish automatic consistency

in behavior between different levels (e.g. establishing

a link between the preferences of an individual and

those of a group he belongs).

In the next section, we present a practical example

of modeling that combines a micro representation of

individuals and human activity with a macro represen-

tation of household and housing. We show how the

dynamics between levels can help business experts to

better understand the complex system studied.

4 HUMAN ACTIVITY

SIMULATION

We now illustrate the potential of multi-level model-

ing to facilitate energy studies, we simulate everyday

life of households in relation with the electrical con-

sumption and their thermal comfort. In the follow-

ing, we present our multi-level modeling realized with

SIMLAB. It is based on three axes with several levels.

We will show in section 5 some simulation results and

why they are interesting for the experts.

4.1 Three Axes Representation

The problem is based on three concepts : population,

human activity and consuming environment. To ap-

ply the previously presented model, we deﬁne respec-

tively the following three modeling axes (as shown on

ﬁgure 1):

The populations axis assemble individuals or

groups of individuals. Each agent in this axis can per-

form activities and have preferences for these. To se-

lect the current activity, agents use a priority function.

The human activity axis, in this example, is re-

duce to a set of tasks characterized by preconditions

(e.g. do the laundry is precondition of ironing) and

links with possible objects in the consuming environ-

ment (e.g. ironing necessitate an iron). When an ac-

tivity is triggered, associated elements of the environ-

ment are activated. The task model in SMACH has

been presented in details in (Amouroux et al., 2013).

The consuming environment axis represents en-

tities related to energy consumption (from the appli-

ance to the housing). An agent in this axis is charac-

terized by an electrical power, a consumption function

(as a recursive properties, see section 3.3) and an ac-

tivation state (off, standby or on).

4.2 Micro Level

The micro level can be summarized as follows, each

individual will perform tasks which may activate ap-

pliances within rooms that will produce electricity

consumption. For each of these agents, we will

present the various characteristics.

4.2.1 Individual

Individuals represent household’s members and are

characterized by an age, a gender, a thermal comfort,

a cold-sensitivity level and a clothing level. Individ-

uals interact with the tasks to achieve them based on

their internal states. An individual associates prior-

ities for each tasks and performs the task with the

highest priority. Household’s members must commu-

nicate to exchange information or request the partici-

pation of others in tasks. In addition to the individuals

internal state, the priority may be inﬂuenced by exter-

nal factors such as energy price level or other individ-

uals invitations. Our model also includes inhabitants

ICAART2015-InternationalConferenceonAgentsandArtificialIntelligence

136

Figure 1: A simpliﬁed view of our three axes representation.

adaptation capabilities to the energy price level. The

cold-sensitivity changes the thermal comfort level felt

by individuals which is computed using the Fanger’s

equation (van Hoof, 2008) depending on their prop-

erties, mainly the temperature and the clothing level.

To communicate, the individuals have a set of speech

acts (e.g. to ask another inhabitant what it does, to

encourage someone to perform an activity, ...).

4.2.2 Task

A task represents a generic activity in the house. It

can be done individually or collectively and can have

a rhythm corresponding to a certain regularity in its

realization which modiﬁes the probability that an in-

dividual achieves it (e.g. dinner occurs every day be-

tween 7 and 9 pm and lasts approximately 1 hour).

A task can have pre-conditional tasks (e.g. the din-

ner must be prepared). An individual can perform a

given task if and only it has the information that all

pre-conditional tasks. Finally, a task can interact with

an appliance to change its activation state (e.g. doing

the laundry activates the washing machine).

4.2.3 Appliance

An appliance updates its electrical consumption

and can perceive the room temperature. Electrical

consumption proﬁles of appliances come from the

database REMODECE

measured in real situations.

The appliances can have a standby power (such as

television, computer, ...) and a thermostat (heater).

Heaters can interact with their room to modify the

REMODECE : European database on residential con-

sumption - http://remodece.isr.uc.pt

temperature. All appliances inﬂuence the room’s tem-

perature according to their radiation property, charac-

teristic of running electrical appliances.

4.2.4 Room

We deﬁne a room as a super-agent of appliances char-

acterized by a number of present individuals, a cur-

rent and a target temperature for heaters. A room

update its temperature using a thermodynamic model

and perceive the number of present individuals.

4.3 Macro Level

We extend our model by adding two macroscopic en-

tities: the household and the housing. These new

agents are represented with a gray background in the

ﬁgure 1.

4.3.1 Household

A household is deﬁned by its type (couple, single-

parent, ...), its comfort sensitivity (eco-oriented,

medium, comfort-oriented) and its income. It has a

thermal comfort and acts on the energy manager in

the housing to regulate the target temperatures in the

rooms depending on its properties. As it is deﬁned

as a recursive property (see 3.3.2), the houshold up-

dates its comfort depending on the inﬂuences of the

sub-agents (individuals). Household properties inﬂu-

ence individuals properties and also modify their ac-

tivity. We associate an inﬂuence on tasks priorities

of individuals. Depending on the household sensitiv-

ity, this inﬂuence reduces the priority when an energy-

consuming appliance (e.g. an oven or a washing ma-

chine) is needed to perform the corresponding task.

StudyofHumanActivityRelatedtoResidentialEnergyConsumptionUsingMulti-levelSimulations

137

Figure 2: An example of cold discomfort feels by an inhab-

itant.

4.3.2 Housing

A housing is composed of rooms. It is characterized

by a presence indicator which perceives if at least one

individual is present in the housing. The housing can,

depending on its energy manager, interact with the

rooms to change the target temperatures. The energy

manager also controls the presences of individuals to

not uselessly heat the housing. The electrical con-

sumption, deﬁned as a recursive property, depends on

the inﬂuences of the sub-agents (rooms).

5 EXPERIMENTS

This section describes two experiments we have con-

ducted. The ﬁrst one illustrates the micro level as an

important aspect for energy experts studies. The sec-

ond one studies the inﬂuence of household types on

individuals and its effect on the consumption.

5.1 Experimental Setup

For our experiments, we generate 50 households us-

ing a populations generator with socio-statistical data

based on a modiﬁed version of the Gargiulo’s algo-

rithm (Gargiulo et al., 2010). Also, generated house-

holds properties such as the income, the surface of the

housing or the number of children are statistically rel-

evant. As it is sufﬁcient for our experimental needs,

we consider that all individuals have the same set of

tasks they can perform (working, eating, sleeping, ...).

The housings have a set a appliances (e.g. TV, heater,

computer, ...) spread in the rooms. The tasks prefer-

ences are randomly set following a linear distribution.

5.2 Observation of Micro Behavior

During the simulation experts will want to observe

situations that can impact the comfort or the energy

consumption and to connect it to the activity that was

performed at that time, so as to detect typical every-

day life situations. For example, we can focus on

the periods when inhabitants are cold and to compare

this with the clothing level. The ﬁgure 2 presents

on the right the thermal comfort (red line) and the

clothing level (blue line) of an inhabitant and, on the

left, an extract of its activity diagram corresponding

to this period. This graphical representation of activ-

ity, where each color corresponds to a task, allow the

experts to visualize inhabitant’s behavior during the

day. So, this diagram allows to precisely retrieve the

activity of an inhabitant on a given simulation step.

The example in ﬁgure 2 shows a decrease in the

clothing level (the agent removes his clothes) causing

an important degradation oh his thermal comfort level

around 10h45. By observing the extract of the corre-

sponding activity diagram, we observe that the com-

fort gap is caused by the realization of the task ”taking

a shower” (in sky blue, boxed in the diagram).

This simple example illustrate how a micro rep-

resentation of activity between several axes identify

phenomena very localized in time that cannot be tak-

ing into account by more macro models. Indeed, a

study of just an average comfort is not pertinent for

a household because, as showed by (Z

elem, 2013),

two individuals in a same household can have differ-

ent perceptions (or even opposite perceptions) of there

comfort, and also different reactions: one can increase

the heater temperature, the other put on a pullover or

change his activity. The interest of a ﬁne grain de-

scription is to reproduce the diversity and the com-

plexity of human behaviors.

This simple example also illustrates how our

model can be used to simplify the modeling process.

Once the modeling axes (population, activity and en-

vironment) have been identiﬁed and the correspond-

ing entities are reiﬁed in the model, the expert only

needs to determine the key observable parameters and

to relate them. Using agents for every entities gives a

more direct access to experts knowledge speciﬁc to

each axis: experts are familiar with the individual-

centered modeling. For example, ergonomists can de-

ﬁne the notions that intervene in the deﬁnition of the

activity, while energeticians will deﬁne the consump-

tion based on the thermal characteristics, etc. We

show in the next section how the intra-axes common

structure of agents, based on inﬂuences between prop-

erties, allows us to easily deﬁne macro-level agents

that interact with each other. This addition of such

agents is limited to the deﬁnition of the new proper-

ties, speciﬁc to these new agents, and the inter-level

inﬂuences on these properties.

ICAART2015-InternationalConferenceonAgentsandArtificialIntelligence

138

5.3 Adding Macro Agents

We now study the consuming activity proﬁle of a set

of households. We aim to investigate the addition

within our model of macro agents at the household-

housing level. As describe in 4.3.1, the household

agent is deﬁned as a super-agent of individuals (the

household members), it will inﬂuence the priority of

the individuals’ tasks and modify the energy manager

of housing based on its comfort (calculated from in-

dividuals’ inﬂuence) and its sensitivity. Thus an eco-

oriented household will have a lower target temper-

ature in the housing that individuals should be com-

pensate, to the extent possible by their clothing levels

(to each type of sensitivity is also associated a level

of minimum and maximum clothing, for example in

some households more focused on comfort, members

wear rarely more than a T-shirt). Moreover, the mem-

bers of such a household (eco-oriented) are incitated

by an inﬂuence to avoid performing consuming tasks.

We perform one-day simulations of our 50 house-

holds (see 5.1) whose number of consuming tasks per

hour is shown in the ﬁgure 3. Contrary to the ﬁrst

experiment (see 5.2), we now take the similar house-

holds but we add the two macro agents (the household

and the housing) with their associated interactions and

inﬂuences. The households we choose to study in this

particular experiment are eco-oriented as describe in

the previous paragraph. The consuming tasks proﬁle

is shown graphically in ﬁgure 3 with the green curve.

The red curve represents the reference proﬁle without

the macro agents and their inﬂuences.

We ﬁrst notice that adding macro agents reduce

signiﬁcantly the consuming activity due to the inﬂu-

ence of the eco-oriented households. As illustrated by

the red curve, we note that meals are structuring tasks

in our modeling as they produce two gaps at 12am and

8pm. They correspond to a certain synchronization of

the meal tasks which are speciﬁed as non-consuming.

Also, meal preparations produce consumption peaks

as shown with the green curve. As the need of elec-

tricity remain important for these tasks, the number of

consuming tasks tend to decrease only slightly with

the inﬂuence of the household.

This choice of measure (number of consuming

tasks) allows us to focus on human activity without

being concerned about avoiding eventual bias in the

consumption model of electrical appliances. It must

be emphasized that some other models are able to

reproduce a residential load curve, however the in-

terest of multi-level modeling is the ability to relate

macro knowledge, e.g. here households typologies

and energetics housing management with more mi-

cro elements (as the example presented in the previ-

Figure 3: Number of consuming tasks per hour.

ous section). Experts can easily test different work-

ing hypotheses such as the impact of different pric-

ing policies on consumption practices, to test the ef-

fectiveness of energy strategies (i.e. heating manage-

ment) or indeed the relationship between human ac-

tivity and consumption peaks. The previous experi-

ment fall within an approach to study how different

types of household with the same housing can pro-

duce different consumption proﬁles.

As mentioned previously, macro knowledge is

available in many areas. Their explicit representation

helps the experts to study the connection with micro

concepts to make them co-evolve. More precisely, if

a household gives more importance to its comfort or

expenses, there is a theoretical link (i.e. an inﬂuence)

on how people will manage their heating energy con-

sumption. However, this relation does not prevent the

diversity of behavior in so far as household members

(individuals) preserve their autonomy.

6 CONCLUSIONS

In this paper, we brieﬂy introduced SIMLAB, a multi-

agent model based on the use of agents of different

levels and inﬂuences to capture the inter-level dynam-

ics. The main interest of our approach is the deﬁnition

of axes for the analysis of complex systems, high-

lighting shared properties. We have shown how multi-

level modeling can enable energy experts to better

understand and anticipate the residential energy con-

sumption. We believe that our model is able to repro-

duce micro and macro phenomena, both interesting

for energy experts, allowing energeticians to consider

possible incentives to reduce consumption peaks.

We consider several perspectives in the context of

our work. We are currently working on extending our

model with more macro concepts such as lifestyle and

social groups to perform large-scale simulations with

several households and housings. We believe that

StudyofHumanActivityRelatedtoResidentialEnergyConsumptionUsingMulti-levelSimulations

139

the addition of these new levels from different ﬁelds

such as sociology, allow us to validate the model as

generic, in terms of modeling axes and inﬂuences. An

large-scale study will be soon performed by EDF us-

ing the SMACH platform. Interviews of households

will be used to create more realistic simulation. Re-

sults will be compared to real data from a several year

in situ experiment with sensors in the clients’ hous-

ings. We are also exploring the possibility of our

agents to dynamically change the MAS organization

proposing to add observations and transformations to

detect and reify potentially useful macro-entities to

help the modeler. To go further in this direction, it

would be interesting to study in more detail the char-

acterization of emergent phenomena.

REFERENCES

Alfakara, A. and Croxford, B. (2014). Using agent-based

modelling to simulate occupants’ behaviours in re-

sponse to summer overheating. In Proceedings of the

Symposium on Simulation for Architecture & Urban

Design, page 13. Society for Computer Simulation In-

ternational.

Amouroux, E., Huraux, T., Sempe, F., Sabouret, N., and

Haradji, Y. (2013). Simulating human activities to in-

vestigate household energy consumption. In Proc. of

the 5th International Conference on Agents and Arti-

ﬁcial Intelligence (ICAART).

Camus, B., Bourjot, C., and Chevrier, V. (2013). Multi-

level modeling as a society of interacting models. In

Proceedings of the Agent-Directed Simulation Sympo-

sium, page 3. Society for Computer Simulation Inter-

national.

FIPA consortium (2003). FIPA Communicative Act Library

Speciﬁcation and FIPA ACL Message Structure Spec-

iﬁcation. Technical report, Foundation for intelligent

physical agents.

Gargiulo, F., Ternes, S., Huet, S., and Deffuant, G. (2010).

An iterative approach for generating statistically real-

istic populations of households. PloS one, 5(1):e8828.

Grandjean, A. (2013). Introduction de non lin

earit

es et non

stationnarit

es dans les mod

eles de repr

esentation de

la demande

electrique r

esidentielle. PhD thesis, Th

ese

de doctorat, Mines Paristech.

Haradji, Y., Poizat, G., and Semp

e, F. (2012). Human Activ-

ity and Social Simulation, pages 416–425. CRC Press.

Huraux, T., Sabouret, N., and Haradji, Y. (2014). A Multi-

Level Model for Multi-Agent Based Simulation. In

Proc. of the 6th International Conference on Agents

and Artiﬁcial Intelligence (ICAART), Angers, France.

Kashif, A., Ploix, S., Dugdale, J., and Le, X. H. B. (2012).

Simulating the dynamics of occupant behaviour for

power management in residential buildings. Energy

and Buildings (online pre-print).

Kubera, Y., Mathieu, P., and Picault, S. (2011). Ioda: an

interaction-oriented approach for multi-agent based

simulations. Autonomous Agents and Multi-Agent

Systems, 23(3):303–343.

Lee, Y. S., Yi, Y. K., and Malkawi, A. (2011). Simulating

Human Behaviour and its Impact on Energy Uses. In

Proc. of the 12th Conference of International Building

Performance Simulation Association (IBPSA), pages

1049–1056.

Minar, N., Burkhart, R., Langton, C., and Askenazi, M.

(1996). The swarm simulation system : a toolkit for

building multi-agent simulations. GEMAS Studies in

Social Analysis, Working Paper 96-06-042.

Morley, J. and Hazas, M. (2011). The signiﬁcance of dif-

ference: Understanding variation in household energy

consumption. ECEEE Proceedings of the 2011 Sum-

mer Study.

Morvan, G. (2012). Multi-level agent-based modeling-

bibliography. Technical report, LGI2A, Univ. Artois,

France.

Morvan, G., Veremme, A., and Dupont, D. (2011).

Irm4mls: the inﬂuence reaction model for multi-

level simulation. In Multi-Agent-Based Simulation XI,

pages 16–27. Springer Berlin Heidelberg.

Muratori, M., Roberts, M. C., Sioshansi, R., Marano, V.,

and Rizzoni, G. (2013). A highly resolved modeling

technique to simulate residential power demand. Ap-

plied Energy, 107:465–473.

Navarro, L., Flacher, F., and Corruble, V. (2011). Dynamic

level of detail for large scale agent-based urban simu-

lations. Proc. of 10th Int. Conf. on Autonomous Agents

and Multiagent Systems (AAMAS 2011), pages 701–

708.

Picault, S., Mathieu, P., et al. (2011). An interaction-

oriented model for multi-scale simulation. In IJCAI

Proceedings-International Joint Conference on Artiﬁ-

cial Intelligence, volume 22, page 332.

Rao, A. S., Georgeff, M. P., et al. (1995). Bdi agents: From

theory to practice. In ICMAS, volume 95, pages 312–

319.

Relieu, M., Salembier, P., and Theureau, J. (2004). Intro-

duction au num

ero sp

ecial activit

e et action/cognition

situ

ee. Activit

es, 1(2):3–10.

van Hoof, J. (2008). Forty years of Fanger’s model of ther-

mal comfort: comfort for all? Indoor Air, 18(3):182–

201.

Yang, R. and Wang, L. (2013). Development of multi-agent

system for building energy and comfort management

based on occupant behaviors. Energy and Buildings,

56:1–7.

elem, M.-C. (2013). Le confort thermique, norme tech-

nique ou norme sociale ? D

ebat National sur la Tran-

sition Energ

etique, Note 12.

Zhou, Z., Chan, W. K. V., and Chow, J. H. (2007). Agent-

based simulation of electricity markets: a survey of

tools. Artiﬁcial Intelligence Review, 28(4):305–342.

ICAART2015-InternationalConferenceonAgentsandArtificialIntelligence

140