Presentation Medart

Bassirou NGOM
PhD student
13 septembre 2020
Presentation DS-RT’20
MEDART-MAS
First:
Hello or good morning everyone, My name is Bassirou NGOM, I’m
going to present you my contribution untitled « MeDART-MAS: a meta-
model of data assimilation on real-time multi-agent simulation »
Our presentation outline can be subdivided into five points.
1. We start with introduction and the context of our work
2. After that we have the related work to present a brief literature
over-view
3. We’ll present you our proposition MeDART on the third part then
the implementation on GAMA simulation
4. We perform some tests with a single use cas
5. At the end, we conclude our presentation.
Introduction
Nowadays, real systems are becoming more and more complex, which
makes it difficult to understand them. Modeling is used to understand
these systems. Simulation have been used to study, understand and predict
complex systems behavior.
Data play an important role in the modeling and simulation processes.
Data is used in the models design, validation and to try/test “what if ”
RAPPORT GÉOLOGIQUE 101 1

scenarios. The simulation also uses the data to calibrate the model
parameters in order to reduce the gap between simulation results and
actual observations .
Nevertheless, combining system observations (the data) with running
simulation can increase the accuracy of these models. This combination
can also improve the results of simulation and react quickly to unexpected
phenomenons like road traffic regulation or wildfire .
For that, data assimilation seems to be an interesting way to inject data
into the simulation and the management of the evolution of the system’s
state with real data.
Data assimilation (DA) is a collection of methods that seek to combine

uncertain models with uncertain data to provide the best estimation of the
system state at the given point in time which observations are available.
In other words, data assimilation allows the simulator to connect to
reality, because it allows the simulation to evolve the state of the system
and to estimate the data that the simulator needs. Assimilation has been
used in several fields of science.
In the context of real-time systems, the multi-Agent approach is the
most preferred for modeling complex systems and for understanding the
emergence phenomena [9]. Indeed, with this approach, several agents will
be able to manage the time constraints of the system and participate in its
dynamics at the same time [4] [5]. However, depending on the field,
several assimilation methods have been developed in the literature. These
methods can be split into 3 groups: DA with sequential methods,
variational methods or machine learning methods.
The work that we present in this paper do not focus on a particular assimilation
method. We propose a meta-model to ease data assimilation in real-time multi-agent

simulation.The architecture of the proposed meta-model is designed to be generic
enough to address various simulation domain. The main goal is to help modelers,
whatever their study case and simulator (GAMA, NetLogo, ...), to implement the
appropriate assimilation methods.
Related Work
The literature shows that multi-agent systems (MAS) are used to study
various real systems. In this modeling approach, the system is presented as a set
of intelligent and autonomous interacting entities called agents.
When it is about real- time systems, several contributions introduce a new
type of agent, often called Real-Time Agent. which is more intelligent and
autonomous than other agents in the system. These agents can manage time
constraints and they are responsible for DA into simulation model.
This approach has been widely used in the management of the road traffic
system. Authors use multi-agent based modeling and a stochastic module to
integrate data into the simulation. In this paper, the authors present a
prediction architecture and a control of road traffic based on data collected in
real-time.
• All of these assimilation methods contain two parts
• Wang M. and Hu X. propose in [15] the assimilation of data sensors for multi-
agent simulation of smart environments in real-time. They use particle filter
as assimilation methods. Though in using particle filters, they estimate the
state of the system, then the simulation restart is dynamic to take into account
the new values estimated as an initial condition.

MEDART
Real world
The real world (real system) represented by the data acquisition system consists
of a set of interactive elements. Sensors are deployed in the system and allow
real-time monitoring and collection of system phenomena
Virtual World
The virtual world is a real-time agent-based model. The model represents the
abstraction of the real world. The latter’s dynamics are studied through
autonomous entities called Agent.
The particularity of these agents is the fact that they can participate in the
dynamism of the system.
Assimilation
The assimilation model is the model where transit data from the real world to be
injected into the virtual world or the simulation. With the agent-based approach
used in the virtual world, RTAgents can have an assimilation model. It means
that the assimilation model is a behavior of the RTAgent. RTAgent can interact
with the environment and other agents in the model. It can also connect to the
servers to obtain data from sensors in real time via the assimilation model.
The assimilation model is mainly composed of three parts:
• Data Adaptor: It is an interface connected to the data acquisition

system. It takes into account data from sensors or databases or other data
sources. This interface defines how to connect data from other data
sources to the DA model.
• Estimation Model: This component is a predictive model used to
estimate a series of data at time tsimul when observation data is not
available. It can be composed of several prediction models or
estimation models according to the correction module. In addition, this
component allows synchronization between simulation(tsimul) and
data acquisition model(Tdata). In order to manage the real-time aspect,
when the observation data is not available, it will predict all data D’

required for the simulation at time tsimul based on the previous
observation data.
• Correction : The Correction module corrects and improves the
estimation model by reducing the gap between the predicted data and
actual data.
Implementation on GAMA
GAMA (GIS Agent-based Modeling Architecture) is an open-source

simulation platform developed since 2007. GAMA is a full Integrated
Development Environment(IDE) based on the Eclipse IDE, and allow to
quickly switch between the modeling and simulation perspective
In GAML, you can use the concept of skills to construct species in a

combined manner. Skills are bundles of attributes and actions that can be
shared between different species and inherited by their children. In order
to implement our proposal, we created an Assimilation skill on GAMA.
Simple Use Case

The aim of our work in this chapter is to test the model and show that we
can provide it with real-time data with a simple assimilation model and
have consistent results. The idea is therefore to try to see what type of
model would be easy to implement for test purposes. We then chose data
temperature because of the simplicity of the models used for the
prediction and their implementation in our simulation platform
Lambda
where a and b represent model parameters, ε estimation error. Since D and

D’ have different temporality (Tdata and tsimul),there is a temporality
problem in this equation. To solve this problem, we introduce λ variable,
which represents assimilation frequency and is defined as:

Fig fitting
Fig. 4 compares the data obtained by the assimilation model with respect
to actual measurement data set. For instance, actual data is represented in
blue, and estimated data is represented in red. It should be noted that
estimated data are obtained by fixed λ = 2. The goal of this simulation is
to evaluate if the chosen estimation model could be adapted to the data in
order to make a prediction with less errors. After simulation, an average
error of 0.17 degrees is obtained.
Error Evaluation
We conducted several simulations by changing the frequency of
assimilation(λ) to understand the evolution of prediction errors. For our
tests, we set Tdata = 1 minute and then change the simulation steps. After
each simulation, we calculate the average error of the prediction.

Presentation Medart

Uploaded by

Copyright:

Available Formats

Presentation Medart

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Presentation Medart

Uploaded by

Copyright:

Available Formats

Bassirou NGOM

RAPPORT GÉOLOGIQUE 101 1

Data assimilation (DA) is a collection of methods that seek to combine

RAPPORT GÉOLOGIQUE 101 2

• All of these assimilation methods contain two parts

RAPPORT GÉOLOGIQUE 101 3

The assimilation model is mainly composed of three parts:

• Data Adaptor: It is an interface connected to the data acquisition

RAPPORT GÉOLOGIQUE 101 4

GAMA (GIS Agent-based Modeling Architecture) is an open-source

In GAML, you can use the concept of skills to construct species in a

Simple Use Case

where a and b represent model parameters, ε estimation error. Since D and

RAPPORT GÉOLOGIQUE 101 5

RAPPORT GÉOLOGIQUE 101 6

You might also like