A few months after the emergence of a new coronavirus in China, the world is hoping for a treatment or vaccine to combat the Covid-19 pandemic. In the meantime, only health policies appear to be able to limit the spread of CoV-2-CoV-SARS. However, it is still necessary to be able to evaluate its effectiveness. This is the aim of a partnership between the Institut de recherche pour le développement (IRD) and EDF R&D: to model the impact of such policies at the local level.
Game of Thrones, World War Z or the mythical Lord of the Rings. What these film productions have in common is the use of “synthetic crowds”. Computer-generated animations that faithfully reproduce the reality of crowd phenomena. “This is an example of what multi-agent systems (MAS) can do,” explains Mathieu Schumann, an engineer in EDF’s R&D department. This branch of artificial intelligence aims to reproduce the behaviour of a set of agents — processes, robots or even human beings — evolving in a certain environment, interacting with each other.
Many disciplines also exploit these technologies. For example, the energy or telecommunications sectors. And it is these SMAs that are today at the heart of a partnership between EDF’s R&D and the Institut de recherche pour le développement (IRD). The two entities have agreed to combine their research resources to meet the challenges of the fight against Covid-19 and provide local decision-makers with a decision-making tool in this unprecedented crisis situation. Whereas conventional epidemiological models can be used to project the evolution of population compartments (compartments of infected, cured, deceased, etc.), the use of multi-agent modelling allows the heterogeneity of the population and the characteristics of individuals (age, occupation, etc.) to be taken into account at several territorial scales.
“For nearly 15 years, EDF R&D has been developing expertise in these multi-agent systems. This is how we develop solutions to simulate the daily life of people in the home, a bit like in the video game The Sims, to better anticipate energy consumption,” explains Mathieu Schumann. But here, it is the IRD — and more precisely its Ummisco research unit based in Vietnam — that has developed with its partners a model that, in conjunction with epidemiological models, makes it possible to assess the risks of coronavirus propagation on a local scale.
The IRD is a reference in agent-based modeling. For fifteen years now, it has been developing an open source modelling and simulation software suite called GAMA Platform®. And right from the start of the coronavirus crisis, the Institute designed, on this platform, a simulator for disease propagation and evaluation of health policies called COMOKIT — for Covid-19 Modeling Kit.
The power of high-performance computing
The input data for this simulator includes statistical data such as the number of houses in the locality under study, the number of inhabitants per household, the age distribution of these inhabitants and the distribution of assets. “All available information that can help generate a virtual population that is representative of the real population,” explains Mathieu Schumann. “Numerical avatars are then modeled to run the simulation. We don’t use any personal data. »
In order to obtain results quickly — in the coming days or weeks — the simulator exploits the computing power available to EDF R&D. More specifically, the GAIA High Performance Computer (40,000 3.2 PFlops computing cores) which allows parallel calculations to be launched on numerous processors to simulate the interactions of thousands of agents. An essential asset for exploring a large set of parameters in very short times. This is of course sought after in the context of the crisis we are experiencing.
The COMOKIT simulator has been designed as a generic simulator, which can be adapted to any case study. It will enable multiple scenarios to be explored and thus define the most suitable containment perimeter, the impact of school closures, the most effective duration of containment or the most effective measures to be implemented.
A test phase is underway in Vietnam. The objective is to extend these simulations to other countries such as Cameroon or Senegal where the IRD is also present. And contacts have already been established with French local authorities that could be interested in these simulation models.