AI-page_4_Covid-19

Contributions of our AI team to Covid-19 situation

Simulating Covid-19 – Infection and Mortality in given Populations by use of Multi-Agent-Simulations

Summary:
Reaching far beyond the possibilities of calculated results via (partial) differential equations, MAS combined with aZAs, allow for a far more adequate representation of real existing populations. The agents being constructed, reflect the specific population to be investigated, both in individual details and in percentages.
Demographic structures, socio-cultural parameters (e.g. the preferred spacial distance in greetings and communication), degrees of compliance to rules imposed, vulnerability in health etc. have been built into the agent-population.
Then the agents have been put into a ’natural environment‘ (traffic- and transportation networks, shops, box offices, restaurants etc.) where they interacted. Various probabilities of infections (both starting from a single agent or from multiple ones) have been assigned plus individual times of incubation and contagiousness. The results in infection numbers and deaths were output after having passed a multilayered ANN, acting as an additional transferfunction for ‚unknown parameters‘. 
In short, after cycles representing about 10 days in a model region resembling Lombardy and Milano, our figures were and still are within a range of 96% +/- 2% of the numbers, reported by Italian authorities until now.
Similar, but slightly less accurate results have been obtained for Spain (93% +/- 3%). We still try to figure out, if we missed some parameter, more significant in Spain than in Italy or what else can explain the lower accuracy.
Proceeding from those MASes, we currently offer companies simulations about the consequences of infections at specific departments, plus estimates about the probability of infections per work- or office-environment.

To share some of the results:
YES, masks (even the simpler ones) are important IF – and only if – the employees know how to use them (adjust the flexible metal strip in the top center section!).
The masks should be company-provided AND there should be rules on how and when to use them.
Additionally, our findings show a non-linearity between ‚working hours‘ and ‚risk of infection or being infected‘. Thus we recommend, whenever possible, just 4 hour shifts with additional 4 hours in remote working.
We also simulated the use of UV-C-beamers inside the air conditioning, a modification that should not be too difficult to implement. (Don’t forget about the ozone!, ask for further details).
Regarding spacial requirements, 9-12 square meters seemed to be an acceptable (hight of the rooms at least 2.75m) compromise between risk and feasability. There is also non-linearity between smaller space and the risk of infection

In case you are interested in simulations of your company-situation, please don’t hesitate to contact us.