Multi-agent Model for Domotics and Smart Houses

Guillaume Guerard, Loup-No

e Levy and Hugo Pousseur

Pole Universitaire L

eonard de Vinci, Research Center, Paris La D

efense, France

Keywords:

Multi-Agent System, Microgrid, Smart House, Home Automation, Demand-Response.

Abstract:

Most of the demand-side management programs focus on the interactions between an aggregator and its users.

Moreover, renewable energy production being irregular, increasing their number implies to predict consump-

tion and energy storage or discharge in real time. This is why the consumption patterns of every device

connected to the grid must be organized in order to optimize the global consumption of the grid. Studying the

smart grid through modeling and simulation provides us with valuable results which cannot be obtained in the

real world due to time and cost related constraints. In this paper, we focus on a multi-agent model to simulate

a microgrid and domotics through automaton and energy consumption scheduling.

1 INTRODUCTION

Our society is electrically dependent. The Power Grid

supplies energy to households, businesses, and indus-

tries. Nevertheless, disturbances and blackouts are

becoming common. With the pressure from ever-

increasing energy demand and climate change, ﬁnd-

ing new energy resources and enhancing energy efﬁ-

ciency have become the priority of many nations in

the 21st century.

The classical electric power infrastructure that has

served us sufﬁciently to a certain extent, also known

as the grid, is rapidly running up against its limita-

tions. Our lights may be on, but systemically, the

risks associated with relying on an often overtaxed

grid grow in size, scale and complexity every day.

The Power Grid is evolving into a Smart Grid, where

power systems, information and communication tech-

nologies meet in order to generate, transport, dis-

tribute and consume energy in a more efﬁcient man-

ner.

A Smart Grid is deﬁned as following (Amin,

2011): it is capable of adapting, predicting and com-

municating with the different agents it is interact-

ing with (production, consumers, weather...) to op-

timize production, transport and consumption of en-

ergy. It can be seen as a complex system optimiz-

ing efﬁciency, reliability and robustness of the elec-

trical grid. It is made of intelligent nodes interacting

autonomously to deliver power to consumers by in-

tegrating advanced control and communication tech-

niques.

A smart grid must allow customers to make in-

formed decisions about their energy consumption, ad-

justing both the timing and quantity of their electric-

ity use. Such technologies detect the need for a load

shedding, communicate the demand for participating

users, schedule load shedding, and verify compliance

with the grid.

From 1980s, many technologies have emerged.

Automatic meter reading was used for monitoring

loads from important customers, and evolved into

the Advanced Metering Infrastructure, whose meters

could store how electricity was used at different times

of the day. Monitoring and synchronization of wide

area networks were revolutionized in the early 1990s

when the Bonneville Power Administration expanded

its smart grid research with prototype sensors that are

capable of very rapid analysis of anomalies in elec-

tricity quality over very large geographic areas.

By the late 1990s, home automation was com-

monly used. Automation describes any system in

which informatics and telematics were combined to

support activities at home. Security, privacy, reliabil-

ity and robustness are important aspects concerning

power grid operations. According to the Federal En-

ergy Regulatory Commission, demand-response (DR)

is deﬁned as:

The changes in electricity usage by end-use

customers from their normal consumption pat-

terns in response to changes in the price of

electricity over time, or to incentive payments

designed to induce lower electricity use at

times of high wholesale market prices or when

Guerard, G., Levy, L. and Pousseur, H.

Multi-agent Model for Domotics and Smart Houses.

DOI: 10.5220/0006757202230230

In Proceedings of the 7th International Conference on Smart Cities and Green ICT Systems (SMARTGREENS 2018), pages 223-230

ISBN: 978-989-758-292-9

223

system reliability is jeopardized.

Because a Smart Grid evolves through time and archi-

tecture, a multi-agent model is needed to understand

its complex behavior. A centralized point of view is

unsuitable to solve every problem in a Smart Grid;

where a distributed or systemic approach gives tools

to model, to understand and to simulate various agents

with their own behaviors in interaction.

The paper presents a simulation of a microgrid

with random devices in a context of Smart Grid. By

the way, the goals are to reduce peak demand, to

smooth the consumption curve and to adapt devices’

behavior in function of the local requirements and the

overall system.

This paper is organized as the following: in the

second section demand-response is introduced in or-

der to deﬁne a generic model for home automation

following by some examples in section 3. Section

4 presents a multi-agent model in a smart grid con-

text. First results are shown in section 5. A discussion

about the model and how to enhance it forms the last

section.

2 GENERIC MODEL FOR HOME

AUTOMATION

Demand-Response includes all intentional modiﬁca-

tions to consumption patterns of electricity to induce

customers that are intended to alter the timing, level

of instantaneous demand, or the total electricity con-

sumption (Albadi and El-Saadany, 2007).

A Smart House is composed of many devices.

They are divided into two categories: those that

includes DR programs and those that don’t. In

our model, demand-response programs are based on

price. For further details about price-based programs,

we recommended the following articles by Albadi et

al. and Cappers et al. (Albadi and El-Saadany, 2008;

Cappers et al., 2010).

We classify devices thanks to simple rules about

their process and the DR programs they can follow in

the following article (Gu

erard et al., 2017).

A device is deﬁned by a set of four categories, it

picks one element of each category:

1. The device is cyclic (i.e. washing machine) or

acyclic (i.e freezer).

2. The device has or doesn’t have batteries.

3. Which data are used: external (i.e. heater in the

room) or internal (i.e internal temperature of the

refrigerator) or both.

4. The device consume, or produce, or both.

Then, a device gets a set of parameters. They are in-

put values that determine how and when a device will

consume or adjust its consumption following a DR

program. The set includes:

Internal data : they correspond to internal data of

the machine, collected by internal sensors.

External data : they correspond to external data of

the machine, in its environment collected by sen-

sors.

Consumer’s preference : these parameters are the

most important. Because the user should not be

hampered in his daily life, its preferences are the

bounds that devices have to reach.

Price : consumers, as described in (Kirschen et al.,

2000), consider both current prices and the prices

of one step into the future. To perform shifting,

devices have to know how to schedule their con-

sumption. In our model, we use current prices,

prices of one step into the future and price trend.

The last two are calculated from derivative func-

tion, norms and a pricer.

There are four general DR programs (QDR, 2006;

Sezgen et al., 2007; Valero et al., 2007):

Reducing Power: customers can reduce their elec-

tricity usage during critical peak periods when

prices are high without changing their consump-

tion pattern during other periods.

Shifting: customers respond to high electricity prices

by shifting some of their peak demand operations

to off-peak periods.

Spinning Reserves: customers use local batteries

and distributed generation combined with loads’

management.

Emergency Cut: when the peak demand is still ac-

tive even after the ﬁrst three actions, an emergency

cut is operated.

For each combination of category-parameters-

program, an automaton is created. A device is an

aggregation of various automaton that describes its

behaviors in function of time, consumer preference

and real-time price.

3 EXAMPLE OF HOME

AUTOMATION

As an example, we will present a cyclic, consumer,

without batteries, with internal parameters’ device: a

dish-washer. In this automaton, a letter of its alphabet

SMARTGREENS 2018 - 7th International Conference on Smart Cities and Green ICT Systems

224

is read every 5 minutes. In order to have a more accu-

rate automaton, we can discretize time in more steps.

Those works will be shown in a further paper.

Depending on the dirtiness of the dishes, the dish-

washer cycle will require heating the water a different

number of times and for different times as shown in

Figure 1. However the states ”cleaning” and ”rins-

ing” (with a consumption respectively at 2100W and

40W) remain the same. Thus, the automaton makes it

possible to model all the cycles with only three states

representing its consumption: StandBy, 2100 and 40

(Figure 2).

It allows a large number of arrangements. Indeed,

according to the conditions (C

V ) the automatons will

create a stack of alphabets elements (L

) describing

the succession of the stages of the cycle. The automa-

ton will unstack from one element to another until the

end of the cycle.

For example, to model the test1 consumption the

automaton take the following stack:

= 40, 2100, 2100, 40, 40, 2100, 40, 40, 40, 40,

40, 40, 40, 40, 2100, 2100, 2100, 40, 40, 40

And to model the test2 consumption the automa-

ton take the following stack:

= 2100, 2100, 40, 40, 2100, 40, 40, 2100, 40, 40,

40, 40, 40, 40, 40, 2100, 2100, 2100, 2100, 40

The presented automaton has a problem in the

case where the average consumption during 5min is

in fact neither 2100W nor 40W . It is then necessary to

either round to the nearest state as done for the 35min-

40min section of the red curve, or to create an au-

tomaton with more states in order to be more accurate

to the real curves, and to take a better discretization

of time (each 2min for example). Another automaton

for the dishwasher is shown in Figure 3).

The following stack represents consumption of

test2:

2bis

= 2100, 2100, 40, 40, 2100, 40, 40, 600, 40, 40,

40, 40, 40, 40, 40, 2100, 2100, 2100, 2100, 40

The model can be reﬁned to the point of repre-

senting any consumption’s curve. It is then possible,

according to the discretization of the time, to estimate

the various actions of the automaton. In our example,

the input stack is ﬁxed, so there is only one possibility

of consumption.

For another automaton, we obtain a preﬁx tree of

the form presented in Figure 4. It is important to note

that from a preﬁx tree, it is also possible to build the

associated ﬁnite state automaton as shown in Figure

5, so building an automaton for a device can be done

by machine learning.

4 MULTI-AGENT MODEL

A system which consists of large populations of con-

nected agents, or collections of interacting elements,

is said to be complex if there exists an emergent

global dynamic. This behavior results from the ac-

tions of its parts rather than being imposed by a cen-

tral controller. That is a self-organizing collective be-

havior that is difﬁcult to anticipate from the knowl-

edge of local behavior (Boccara, 2004). The complex

system approach is described in the following articles

(Ahat et al., 2013) and (Amor et al., 2014).

Computer modeling and simulation have proven

to be a useful tool, if not essential, to help decision

making in studying and designing complex artiﬁcial

systems (Molderink et al., 2009). Any change in

a Smart Grid involves millions or billions of Euros.

Thus, any change needs a deep study and some sim-

ulations to integrate or to understand all the conse-

quences and any kind of new behaviors, disruptions

in the new grid.

A Smart Grid presents a shared resource among

multiple actors, with divergent interests. A multi-

agent system (MAS) modeling presents the global dy-

namic of the system from individual components and

explores emergent properties associated with this dy-

namic. However, it should be noted that MAS have

a major drawback: one model run does not allow to

conclude about the relationship between model and

results (Weiss, 1999). We will present in the next sec-

tion some results, but previously let us expose our ﬁrst

model.

Our model focuses on microgrid. A microgrid is a

broader view of local consumers, it is a tree structure

representing an eco-district bounded by the upstream

substation. Its goal is to distribute energy from a sub-

station to consumers. It orders an amount of energy

from the T&D network to local consumers.

A local consumer supports the consumption of en-

ergy, which is the distribution of energy among de-

vices under its responsibility. In other words, a local

consumer is deﬁned by the area under the control of

a smart meter or other automation/management con-

troller. Those devices may also produce energy or

storage energy.

The goals of any microgrid are to limit using ex-

ternal sources of energy, to avoid brutal changes in its

consumption curve. The consumption of each device

from each house has to be adapted accordingly to the

microgrid’s behavior. For the ﬁrst simulation, we take

four houses for one microgrid.

In each of the houses there are a number of de-

vices: some have batteries others do not, some are

cyclic other are not etc. These devices will turn them-

Multi-agent Model for Domotics and Smart Houses

225

Figure 1: Consumption’s curve for a dishwasher.

selves on or o f f , or will reduce/increase their con-

sumption according to their automaton. Their behav-

ior will depend on global variables of the microgrid

as well as on local variables according to the previous

section.

Figure 2: Automaton for a dishwasher.

Figure 3: Automaton for a dish-washer (more accurate).

Each agent represents a device with its parameters

and values. The device will choose a consumption

scheme according to his situation. The devices with

a battery can choose to charge or to discharge their

batteries. The devices may choose to defer their con-

sumption if they can. This way each device, in func-

Figure 4: A preﬁx tree automaton (Datta and Mukhopad-

hyay, 2015).

tion of its automaton, has a variety of schemes it can

choose to adopt.

On the UML diagram (see Figure 6) we can see

how the internal and external parameters inﬂuence the

behavior of the devices, whether it is by activating the

battery or delaying consumption.

SMARTGREENS 2018 - 7th International Conference on Smart Cities and Green ICT Systems

226

Figure 5: The ﬁnite state automaton associated to Figure 5

(Datta and Mukhopadhyay, 2015).

5 FIRST RESULTS

When the regulation is activated in the microgrid, the

devices with a battery try to regulate the system using

one principle: if the global consumption goes beyond

the average value plus a tolerance value that is spe-

ciﬁc to each device, then the devices will set itself on

battery mode in order to get the global value to get

closer to its average value. This way, important vari-

ations are avoided. But if for every device the thresh-

old tolerance value was the same, then they would all

react at the same time, leading to important peaks in

the global consumption (see Figure 7). Therefore, the

variation of the threshold value must be adapted to the

system to stay stable. The same principle is applied

to the deferrable devices and with a load shifting pro-

gram.

The global consumption is approximate to a cyclic

curve. The consumption is higher in the evening and

decreases rapidly after midnight. The model has been

studied running in three different ways, ﬁrst without

any regulation, second with a well-adjusted regulation

and third with a poorly adapted regulation. A day is

divided in 144 time units.

The ﬁgure 7 represents the consumption in func-

tion of the time during a day without regulation.

On ﬁgure 8 we can see how the amplitude of the

peaks are lower with a well-adapted regulation than

with no adaptation. The consumption of the grid stays

closer to its average.

On Figure 9, even though the consumption-mode-

changing-threshold-value (the consumption value of

the grid that makes the device change his mode of

consumption) is different for each device, the re-

sponse of the devices is not adapted to the grid. The

distribution of the threshold values are not adapted

to the variation of the consumption. This gives an

overactive microgrid and therefore an unstable sys-

tem. Here, all the devices react the same way at a

small decrease in consumption leading to an impor-

tant rise of consumption, to which the devices react

with an even greater decrease, leading the system of

balance. The cause of this problem is the bad distri-

bution of the consumption-mode-changing-threshold-

value. The solution to this problem was the cre-

ation of a process in which the consumption-mode-

changing-threshold-value of the device is regularly

updated based on the historic of global consumption.

This way the devices can adapt to a change in the grid.

On the ﬁgure 10, the regulation began after 140

ticks (unite of time). The system went off balance,

but once the regulation activated, the devices adjusted

their threshold values according to the historic of con-

sumption and managed to ﬁnd balance again.

6 DISCUSSION AND FUTURE

WORKS

The simulation of this paper is based on NetLogo.

NetLogo is a language and an IDE (integrated de-

velopment environment) focused on MAS, allowing

to easily create graphs, animations 2D/3D. The main

reason to choose NetLogo is the simplicity for creat-

ing a prototype allowing fast results.

One of the disadvantages of NetLogo for our work

is that the centralization of information about the de-

vice consumption cannot be done. In the last simu-

lation, each device had to take decisions alone, only

based on the global consumption. The best way to

centralize information would be to create an agent

with such a role. JADE allows the attribution of more

speciﬁc tasks for each agent, with a more important

consumption monitoring, and more interactions be-

tween devices. This representation is more realist, it’s

easier to implement a device controller than to imple-

ment a controller on each device.

Table 1: Comparisons between JADE and Netlogo.

Features JADE NetLogo

Utility Post Prototype Prototype

Open source Yes Yes (since v5)

Development complex, longer simply,fast

Task Multi-threads Only one

Synchronization asynchronous synchronous

Object Oriented Yes No

Ontology Yes No

Programs Several Only one

Service Notion Yes No

Language Java NetLogo

This table shows the principal differences between

Multi-agent Model for Domotics and Smart Houses

227

Figure 6: UML diagram of a device with a battery and shifting abilities.

Figure 7: Curve with no regulation.

JADE and NetLogo

On the ﬁgure 11, the control agent analyzes the

current global consumption. This analysis gives the

best distribution strategy improving the optimization,

taking into account:

• User parameters (each home gets difference con-

sumption).

• Weather (outside and inside).

• Electricity market price (current price compared

with the average price).

• The devices list already launched.

Figure 8: Curve with well-adapted regulation.

This agent can launch a device, ask a device to switch

off (only if this device is an acyclic device). During

a device execution, the control agent can change its

consumption, but this change has an impact on the

execution time.

Each consumer device is represented by an

agent, at each creation, an agent begins by declaring

himself at the directory agent and waits for an order

gave by the control agent.

• A cyclic device executes the order until the task is

ﬁnished.

SMARTGREENS 2018 - 7th International Conference on Smart Cities and Green ICT Systems

228

Figure 9: Curve with ill-adapted regulation.

Figure 10: Curve with autoadapted regulation.

• An acyclic device executes the order until the con-

trol agent asks to stop.

When a device stops, the agent returns to the control

agent the electric consumption used. After using an

agent can be replayed.

Each edge represents interaction between two

agents. The AID is an object of JADE, representing

the agent address. The control agent needs to have the

device AID for sending a launch request.

The JADE model will replace the NetLogo model

in order to provide more results. The JADE model

will be based on a real microgrid near Paris, France,

named Le-Perray. This project aimed to combine

BIM model, MAS model and deep learning process to

optimize energy consumption, production and lower

the price for all users.

Figure 11: Control agent explication.

Figure 12: Interaction between home agents.

ACKNOWLEDGEMENTS

Two of the authors are in second and last year in

an engineering school (France, the same degree as

Msc). They work in a half-time curriculum with an

associated professor about their subject (respectively

a multi-agent model and a generic automaton model

for smart devices). This paper concludes their ﬁrst

year in this curriculum.

Multi-agent Model for Domotics and Smart Houses

229

REFERENCES

Ahat, M., Amor, S. B., Bui, M., Bui, A., Gu

erard, G., and

Petermann, C. (2013). Smart grid and optimization.

American Journal of Operations Research, 3:196–

206.

Albadi, M. H. and El-Saadany, E. (2007). Demand response

in electricity markets: An overview. In 2007 IEEE

power engineering society general meeting.

Albadi, M. H. and El-Saadany, E. (2008). A summary of de-

mand response in electricity markets. Electric power

systems research, 78(11):1989–1996.

Amin, S. M. (2011). Smart grid: Overview, issues and op-

portunities. advances and challenges in sensing, mod-

eling, simulation, optimization and control. European

Journal of Control, 17(5-6):547–567.

Amor, S. B., Bui, A., and Gu

erard, G. (2014). A context-

free smart grid model using complex system ap-

proach. In Proceedings of the 2014 IEEE/ACM 18th

International Symposium on Distributed Simulation

and Real Time Applications, pages 147–154. IEEE

Computer Society.

Boccara, N. (2004). Modeling complex systems. Springer

Verlag.

Cappers, P., Goldman, C., and Kathan, D. (2010). De-

mand response in us electricity markets: Empirical

evidence. Energy, 35(4):1526–1535.

Datta, S. and Mukhopadhyay, S. (2015). A grammar infer-

ence approach for predicting kinase speciﬁc phospho-

rylation sites. PloS one, 10(4):e0122294.

erard, G., Pichon, B., and Nehai, Z. (2017). Demand-

response: Let the devices take our decisions. In Pro-

ceedings of the 6th International Conference on Smart

Cities and Green ICT Systems, pages 119–126.

Kirschen, D. S., Strbac, G., Cumperayot, P., and

de Paiva Mendes, D. (2000). Factoring the elastic-

ity of demand in electricity prices. IEEE Transactions

on Power Systems, 15(2):612–617.

Molderink, A., Bosman, M., Bakker, V., Hurink, J., and

Smit, G. (2009). Simulating the effect on the energy

efﬁciency of smart grid technologies. In Winter Sim-

ulation Conference (WSC), Proceedings of the 2009,

pages 1530–1541. IEEE.

QDR, Q. (2006). Beneﬁts of demand response in electricity

markets and recommendations for achieving them. US

Dept. Energy, Washington, DC, USA, Tech. Rep.

Sezgen, O., Goldman, C., and Krishnarao, P. (2007). Op-

tion value of electricity demand response. Energy,

32(2):108–119.

Valero, S., Ortiz, M., Senabre, C., Alvarez, C., Franco, F.,

Gabald, A., et al. (2007). Methods for customer and

demand response policies selection in new electricity

markets. IET generation, transmission & distribution,

1(1):104–110.

Weiss, G. (1999). Multiagent systems: a modern approach

to distributed artiﬁcial intelligence. The MIT press.

SMARTGREENS 2018 - 7th International Conference on Smart Cities and Green ICT Systems

230