A MULTI-AGENT HOME AUTOMATION SYSTEM FOR POWER

MANAGEMENT

Shadi Abras

, St

ephane Ploix

§§

, Sylvie Pesty

and Mireille Jacomino

§§

Laboratoire Leibniz-Institut IMAG

CNRS, UMR5552, 46, Avenue F

elix Viallet

38031 Grenoble, France.

§§

Laboratoire d’Automatique de Grenoble

CNRS, UMR5528, BP 46

38402 Saint Martin d’H

eres, France.

Keywords:

Home Automation System, Multi-Agent Systems, Automatic Control, Negotiation and Cooperation, Power

Management.

Abstract:

This paper presents the principles of a Home Automation system dedicated to power management that adapts

power consumption to available power ressources according to user comfort and cost criteria. The system

relies on a multi-agent paradigm. Each agent is embedded into a power resource or an equipment, which

may be an environment (thermal-air, thermal-water, ventilation, luminous) or a service (washing, cooking),

and cooperates and coordinates its action with others in order to ﬁnd acceptable near-optimal solution. The

control algorithm is decomposed into two complementary mechanisms: an emergency mechanism, which

protects from constraint violations, and an anticipation mechanism, which computes the best future set-points

according to predicted consumptions and productions and to user criteria. The paper details a negotiation

protocol used by the both mechanisms and presents some preliminary simulation results.

1 INTRODUCTION

For the next decades, the two major problems con-

cerning energy are the greenhouse effect and the de-

pletion of petrol resources especially the energy pro-

vided by oil and gas. Therefore, by conscience or by

necessity, the resort to renewable resources of energy

such as wind or solar radiations, arrives in the build-

ings knowing that the building represents 47% of the

energy consumption and it is responsible for 25% of

the greenhouse effect (Fontaine, 2003). Moreover,

undoubtedly, the user will be confronted by variable

tariffs of energy according to the hour and the days

and to the energy producers. It is in this varied and

dynamic context of production and consumption of

energy that a building, equipped with a Home Au-

tomation system to control the energy, takes its im-

portance. The role of a Home Automation system

dedicated to power management is to adapt the power

consumption to the available power resources taking

into account user comfort criteria: it permits to limit

the use of supplementary resources which require ad-

ditional investment and to avoid the expensive need of

storage. A Home Automation system has to reach a

compromise between the priorities of the user in term

of comfort and in term of cost while satisfying tech-

nological constraints of equipment and user’s comfort

constraints.

This problem can be formulated as a scheduling prob-

lem. In (Ha et al., 2005), a solution based on a

Resource Constrained Project Scheduling Problem

(RCPSP), to improve the management of thermal-air

equipments, is presented. Its aim is to satisfy resource

constraints by coordinating the control of thermal-

air equipment. Nevertheless, this approach requires

precise predictive models and RCPSP techniques are

hardly adaptable to the context of multi energy re-

sources and multi equipments. In (Ha et al., 2006), an

anticipation mechanism using Bellman-Ford’s algo-

rithm (Cherkassky et al., 1994) is presented for solv-

ing the problem of managing predicted events in a

Home Automation system. The principal advantage

of Bellman-Ford’s approach is that the optimal solu-

tion is guaranteed (if exist) but the major disadvantage

Abras S., Ploix S., Pesty S. and Jacomino M. (2006).

A MULTI-AGENT HOME AUTOMATION SYSTEM FOR POWER MANAGEMENT.

In Proceedings of the Third International Conference on Informatics in Control, Automation and Robotics, pages 3-8

DOI: 10.5220/0001207200030008

 SciTePress

is the high order of complexity.

An alternative approach is to use Multi-Agent tech-

niques. Algorithms based on Multi-Agent Systems

are nowadays used in several areas such as Computer

science or Automatic Control. The ﬁrst MAS ap-

proach for energy distribution have been presented in

(Jennings, 1994) and (H

agg and Ygge, 1995). (Kok

et al., 2005) put forward a market-based control con-

cept for the supply and demand matching (SDM) in

electricity networks. It aims to propose a Multi-Agent

system for the electronic market. Its purpose is to

control tasks in future electricity network which is ex-

pected to develop into a network of networks in which

a vast number of system parts communicate and coor-

dinate with each other.

The developments of solutions based on Multi-Agent

Systems, well suited to solve spatially distributed and

opened problems, permit to imagine an intelligent

Multi-Agent Home Automation system. This paper

presents a Multi-Agent Home Automation System

(MAHAS). It focuses on the deﬁnition of a negoti-

ation protocol between agents embedded into equip-

ments as well as in energy resources. The paper is

organized as follows: section 2 describes, in a general

view point, the Multi-Agent Home Automation Sys-

tem. Section 3 presents the two main mechanisms of

this system: the emergency and anticipation mecha-

nisms. Section 4 presents, in detail, the principle of

the negotiation protocol for emergency and anticipa-

tion mechanisms. Then, the paper presents some pre-

liminary results and highlights the future work which

will be done.

2 MULTI-AGENT HOME

AUTOMATION SYSTEM

The three main features of the Multi-Agent Home Au-

tomation System (MAHAS) (Figure 1), which con-

sists of agents embedded into energy resources and

into the different equipments, are the following:

• Distributed: the energy resources and equipments

are distributed spatially and their control systems

are independent.

• Flexible: the energy resources are few but also

some equipments can accumulate energy (thermal-

air, thermal-water) or satisfy with delay to demands

of services (washing service, cooking service).

• Opened: the number of connected resources and

equipments may vary with time (equipments or re-

sources can be connected or disconnected) without

having to completely redeﬁne the control mecha-

nism.

In Multi-Agent Systems, the notion of control in-

volves operations such as coordination and negotia-

Solarradiations

Cooking

Washing Refrigerator

Heating

Hot water

Wind

Figure 1: Energy network and communication between em-

bedded agents housing.

tion among agents, elimination of agents that are no

longer present and adding new agents when needed.

2.1 Agent Architecture

The main functionalities of an agent in MAHAS are

shown in ﬁgure 2.

Prediction of

power needs

Co-Construction

of a plan

Characterization of

user behavior

Application or

revision of a plan

Prediction of

power resources

Prediction of

user behavior

Equipment

Actions

Applied set points

Plan

Predict set points

power

resources

Power

needs

Current state

User Constraints

& criteria

Weather

forecast

Information

from energy

producers

Agents

Agent

Local control

Anticipation

mechanism

Emergency

mechanism

Figure 2: Structure of an agent in MAHAS.

Depending on weather forecast, energy resource in-

formation and user habits:

• Resource agent calculates the available power re-

sources: to determine what is and what will be the

available power. For the moment, the energy re-

sources are represented by a virtual energy resource

which manages operations between the different re-

sources.

• Equipment agent calculates the prediction of power

consumption: to determine what are the future

power needs taking into account the usual behav-

iour of users.

ICINCO 2006 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION

From these predictions and taking into account

the user constraints and criteria, a plan is jointly

constructed by the different agents which negotiate

their future power consumption (section 4). The

construction of a plan by cooperating and negotiation

between agents is called the anticipation mechanism

(subsection 3.2). This plan includes predicted values

of the variables that characterize the environments

(for example: the room temperatures) or the end dates

of services (an oven for instance). Then, this plan is

applied but it can be modiﬁed in case of unforeseen

perturbations (for example: consumption peak). If

the perturbation is so important, the agents rene-

gotiate in order to recalculate plans. The real time

adjustment of a plan in order to match constraints is

achieved by cooperation between agents: it is called

the emergency mechanism (subsection 3.2).

A third mechanism may exist: the local control mech-

anism i.e. the controllers endowed into equipments

by the manufacturers. It’s time response is very

fast. This mechanism receives set points from the

agents. Besides, some information on its current state

(power needs) are sent back to the agents so that they

can be taken into account in the future plans. This

mechanism is not mentioned in this paper because

other mechanisms are slower and local controls are

assumed to be transparent.

One of the objectives of the MAHAS is to fulﬁl user

comfort. A notion bound directly to the comfort is

the satisfaction function (Simonin, 2001). Satisfac-

tion functions have been deﬁned for energy resources

as well as for equipments. The equipment satisfac-

tion function will be expressed by a function deﬁned

on the domain of the characteristic variable corre-

sponding to the interval [0, 100%] where zero means

”inadmissible” and 100% is ”perfect”. For example:

thermal air environment satisfaction function, which

is deﬁned on room temperature values corresponding

to an interval selected by user, can be represented by

ﬁgure 3. The resource satisfaction function is also ex-

Figure 3: Thermal air environment satisfaction function.

pressed by a function where the characteristic variable

corresponds to produced power. When the produced

power exceeds the resource capacity, the satisfaction

function falls to 0%. The nominal power of the re-

sources corresponds to 100%.

3 AGENT MECHANISMS

3.1 Emergency Mechanism

The emergency mechanism is a real time adjustment

mechanism which is triggered out when the level of

satisfaction of an agent falls below the weak values

(10% for example). This mechanism, which relies

on the negotiation protocol (section 4), permits to

react quickly to avoid violations of energy constraints

and to guarantee a good level of user satisfaction. It

is considered as transparent for anticipation mech-

anism because emergency adjustments have very

small impact on the period considered by anticipation.

Therefore, the emergency mechanism adjusts, in

real time, set points coming from the predicted plan,

equipment’s current state (equipment satisfaction

value) and constraints and user criteria. The predicted

set points can be directly transmitted to the local

control mechanism or modiﬁed in case of emergency.

When the emergency mechanism is triggered, each

agent has multiple roles:

• It evaluates, at predeﬁned intervals, its current sat-

isfaction. Therefore, it uses an inﬁnite internal

loop. This interval of time is called checking pe-

riod.

• It can request help from other agents, by sending

messages, when its satisfaction falls below a level

of emergency.

• It analyzes the other agent demand and makes some

propositions.

• When it receives some answers to its demands, it

chooses and accepts the interesting propositions (to

have a maximum value of satisfaction).

• It can allow, according to received messages, to

activate or inactivate its associated equipment.

If an equipment agent satisfaction decreases, it sends

messages requesting help from resource agents to ini-

tiate a negotiation. Other agent answers are collected

during a ﬁxed delay and are sorted out according to

their satisfaction values. Then a solution which max-

imizes the satisfactions of equipments and resources

is chosen.

3.2 Anticipation Mechanism

The emergency mechanism is sufﬁcient to avoid

constraint violations but a MAHAS can be improved

in order to avoid emergency situations. This im-

provement is obtained thanks to the anticipation

mechanism. The objective of this mechanism is

A MULTI-AGENT HOME AUTOMATION SYSTEM FOR POWER MANAGEMENT

to compute the predicted set points depending on

predictions of consumptions and on predictions of

energy resources. The anticipation mechanism relies

on the fact that there is on the one hand, some electric

equipments which are capable of accumulating

energy and on the other hand, some services that

have a variable date as for their execution: some

services can both be delayed or advanced. From these

preliminary observations, it is possible to imagine

that if the equipment consumption can be anticipated,

there is a way to organize it better.

The anticipation mechanism relies on learning

algorithms which are not explained in this paper.

As for the emergency mechanism, the anticipation

mechanism relies also on a negotiation protocol

(section 4). It works on a time window (anticipation

period) larger than the checking period and works

with average values of energy, because it is difﬁcult to

make precise predictions, in order to keep emergency

mechanism transparent for it.

During anticipation mechanism, each agent has mul-

tiple roles:

• When requested, it predicts future needs or re-

sources over a given number of anticipation period.

This period is a multiple of the checking period.

• It analyzes the other agent demands and makes

some propositions.

• When it receives some answers to its demands, it

chooses and accepts the best propositions (to have

a maximum value of satisfaction for all).

• It calculates, according to received messages, its

predicted set points.

The message exchanges between agents during emer-

gency and anticipation negotiations are deﬁned by a

protocol which is presented in the next section.

4 NEGOTIATION PROTOCOL

The negotiation protocol has been deﬁned on the ba-

sis of the contract negotiation model (Mathieu and

Verrons, 2004), CNP protocol ((Smith, 1980), (Yang

et al., 1998)) and algorithms of distributed constraint

satisfaction problems (Makoto and Hirayama, 2000).

This protocol can be used for agent mechanisms ac-

cording to the checking period for emergency mech-

anism and anticipation period for anticipation mecha-

nism.

The negotiation protocol is characterized by suc-

cessive messages exchanged between resource and

equipment agents. Agents exchange messages for two

objectives:

• To avoid to overpass the maximum available en-

ergy.

• To keep the satisfactions over a certain value: ac-

ceptable characteristic variable for environments

which accumulate energy and acceptable shifts for

services.

The agreements issued from negotiations are based on

satisfactions of equipments (representing user com-

fort criteria) and on satisfactions of resources (repre-

senting the ideal power production).

4.1 Phases of Negotiation Protocol

The negotiation protocol (Figure 4) may be decom-

posed into three phases:

• Energy demand phase: During this phase, the re-

source agents request equipment agents for propo-

sitions that lead to satisfactions greater or equal to

an attempting satisfaction value and wait for equip-

ment agent answers.

• Proposition phase: A conversation between re-

source agents and equipment agents takes place

during which new propositions are exchanged.

Then, resource agents analyse these propositions

and can either accept them or request for equipment

agents to send all the solutions for a new attempting

satisfaction.

• Final decision phase: The resource agents take the

decision, so equipment agent demands can either

be accepted or refused.

Figure 4: Negotiation protocol.

ICINCO 2006 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION

The global success of negotiation is reached when all

the equipments have reached quite similar satisfac-

tions. When an event is under negotiation and no so-

lution is possible, a negotiation with the user starts to

modify user constraints.

4.2 Primitives of Negotiation

Protocol

The primitives of negotiation protocol are decom-

posed into two groups.

Energy resource agent primitives:

• Request: The resource agents initiate a negotiation

by asking equipment agents to send them their

power needs in order to reach a satisfaction greater

or equal than an attempting satisfaction provided

by resource agent. It collects the answers, it

veriﬁes if there is a global solution. Next request

indicates to the equipment agents that there is no

solution for the attempting satisfaction because

the energy asked by equipments exceeds the

maximum available energy provided by resources,

so resource agents request equipment agents to

send them other propositions about their needs for

a smaller attempting satisfaction. A request may

be deﬁned as:

request(mechanism-name, period,

satisfaction)

where mechanism-name has two values ”emer-

gency” or ”anticipation”. period value may be

equal to the checking period or to the anticipation

period. satisfaction is the attempting satisfaction

value provided by resource agent.

• Accept: This message indicates to equipment

agents that one of the proposed solutions has been

accepted by a resource agent. This message may

be deﬁned as:

accept(proposition)

where proposition is one of the solutions proposed

by the equipment agents.

• Renegotiate: This message indicates that there is

no solution that satisﬁes the constraints deﬁned by

the user. A negotiation with the user starts. This

message may be deﬁned as:

renegotiate(constraints)

where constraints is the set of constraints that can-

not be satisﬁed.

Equipment agent primitive:

• Help: This message initiates a negotiation. It is

sent when an emergency situation is detected or

foreseen for the next checking period. It may be

deﬁned as:

help().

• Propose: This message replies to a request from

resource agents. It contains a set of propositions

of possible sequences of energy consumption

covering one period for an emergency mechanism

or several periods for an anticipation mechanism.

This list may be empty if there are not any possible

propositions.

This message may be deﬁned as:

propose(set-of-powers,

satisfactions)

where set-of-powers are the propositions of equip-

ment agent during the checking or anticipation

period. satisfactions are the predicted satisfaction

values corresponding to each proposition.

4.3 Preliminary Results

In this subsection, an illustrative example is presented

for Home Automation system which consists only of

thermal air environments which are the largest part

of consumption of electricity in buildings in winter.

This system consists of three electrical heaters of

1kW each and a 2100W energy resource knowing

that the initial temperatures in rooms are ﬁxed to

C and the desired value of temperature is 20

(satisfaction function takes its values between 0% for

C and 100% for 20

C). The temperature values

Figure 5: Simulated temperature in room 1.

for other rooms are quite similar to room 1 (Figure

5). The control system, in this example, is capable of

maintaining temperature values for each environment

above 18

C: because of the lack of power, the tem-

peratures remain close to the minimum acceptable

value. An example of exchanged messages between

the energy resource agent and the equipment agents

is presented below:

Heater2: help(heater2)

Resource: request

("emergency",15s,90%)

Heater2: propose (900W,90%)

Heater1: propose (900W,90%)

Heater3: propose (900W,90%)

Resource: request

("emergency",15s,80%)

A MULTI-AGENT HOME AUTOMATION SYSTEM FOR POWER MANAGEMENT

Heater3: propose (800W,80%)

Heater1: propose (700W,70%)

Heater2: propose (750W,75%)

Resource: request

("emergency",15s,70%)

Heater1: propose (650W,65%)

Heater3: propose (700W,70%)

Heater2: propose (600W,60%)

Resource: accept(650W, 600W, 700W)

Heater1: help(heater1)

Resource: request

("emergency",15s,90%)

Heater2: propose (900W,99%)

Heater1: propose (900W,90%)

Resource: accept(900W, 900W, 0W)

Heater2 agent has requested help from the resource

agent to start the negotiation. Then a conversation

between the agents takes place during which the re-

source agent requests the equipment agents to send

their propositions for an attempting satisfaction value,

and during which the equipment agents send their

propositions, which may be empty, to the resource

agent.

In the absence of MAHAS but with an unbalancing

system, always the same heater is penalized when all

heaters simultaneously consume energy according to

the user’s predeﬁned priorities. Contrary to MAHAS,

the maximum user satisfaction cannot be guaranteed.

5 CONCLUSION AND

PERSPECTIVES

This paper has presented a Multi-Agent Home Au-

tomation system allowing the agents to cooperate and

coordinate their actions in order to ﬁnd the accepted

near-optimal solution for power management. Nego-

tiation protocol has been detailed. The experimen-

tal results have showed the performance of the ne-

gotiation algorithm. This paper have provided evi-

dence that cooperation and negotiation capabilities of

Multi-Agent systems can be advantageously used in

automatic control systems for spatially distributed and

opened systems.

The implementation of a simulator for the emergency

and anticipation mechanisms is not ﬁnished yet. This

simulator will be tested on a reduced-scale model of

an apartment composed of two thermal environments

and several services (washing machine,...). Each en-

vironment contains a reduced-scale electric heater, a

temperature sensor and a micro-controller card with

an embedded Java Virtual Machine.

REFERENCES

Cherkassky, B. V., Goldberg, A. V., and Radzik, T. (1994).

Shortest paths algorithms: theory and experimental

evaluation. In SODA ’94: Proceedings of the ﬁfth an-

nual ACM-SIAM symposium on Discrete algorithms,

pages 516–525, Philadelphia, PA, USA. Society for

Industrial and Applied Mathematics.

Fontaine, N. (2003). Livre blanc sur les

energies. d

ebat national sur les

energies,

http://www.industrie.gouv.fr/energie/politiqu/ploe.htm.

Ha, D. L., Ploix, S., Zamai, E., and Jacomino, M. (2005).

Control of energy consumption in home automation

by ressource constraint scheduling. In The 15th Inter-

national Conference on Control System and Comput-

erScience, Bucharest, Romania.

Ha, D. L., Ploix, S., Zamai, E., and Jacomino, M. (2006).

A home automation system to improve household en-

ergy control. In The 12th IFAC Symposium on Infor-

mation Control Problems in Manufacturing.

agg, S. and Ygge, F. (1995). Agent-oriented programming

in power distribution automation. PhD thesis, Univer-

sity of Karlskrona/Ronneby, Ronneby, Sweden.

Jennings, N. R. (1994). The ARCHON system and its appli-

cations. In Second International Working Conference

on Cooperating Knowledge Based Systems (CKBS-

94), pages 13–29, Keele, UK.

Kok, J. K., Warmer, C. J., and Kamphuis, I. G. (2005).

Powermatcher: multiagent control in the electricity in-

frastructure. In AAMAS ’05: Proceedings of the fourth

international joint conference on Autonomous agents

and multiagent systems, pages 75–82, New York, NY,

USA. ACM Press.

Makoto, Y. and Hirayama, K. (2000). Algorithms for

distributed constraint satisfaction: A review. Au-

tonomous Agents and Multi-Agent Systems, 3(2):185–

207.

Mathieu, P. and Verrons, M. H. (2004). Three differ-

ent kinds of negotiation applications achieved with

GeNCA. In Proceedings of the International Confer-

ence on Advances in Intelligent Systems - Theory and

Applications (AISTA) In cooperation with the IEEE

Computer Society, Centre de Recherche Public Henri

Tudor, Luxembourg-Kirchberg, Luxembourg.

Simonin, O. (2001). Le mod

ele satisfaction-altruisme :

coop

eration et r

esolution de conﬂits entre agent situ

eactifs, application

a la robotique. PhD thesis, Uni-

versit

e Montpellier II.

Smith, R. G. (1980). The contract net protocol: High-

level communication and control in a distributed prob-

lem solver. IEEE Transaction on Computers, C-

29(12):1104–1113.

Yang, J., Havaldar, R., Honavar, V., Miller, L., and Wong,

J. (1998). Coordination of distributed knowledge net-

works using contract net protocol. IEEE Information

Technology Conference, Syracuse, NY.

ICINCO 2006 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION