Analysis of the Spread of COVID-19: Multiagent-based Simulation by Using Human Mobility Data

Research and Development by

Masaki Onishi, AIST

Corresponding Research Area

Use of ICT and IoT for contact reduction

Analysis of a large-scale Human Mobility in Japan: Trends in mobility patterns based on internal and external prefectural population movements

The GPS-based data is categorized into three types of mobility patterns: (1) prefectural, (2) regional , and (3) inter-regional movements, shown as a percent (%). 11 regions are referred to the Japan Meteorological regions.

(Mobility data: January 2020 to the end of October, shared by Blogwatcher Inc.)

  • Human mobility gradually increased in September, especially during weekends and holidays.
  • Mobility increase in September shows a similar trend in three consecutive holidays in March 2020.

Analysis of a large-scale Human Mobility in Japan: Trends in Mobility Patterns in Nightlife Spots

Visitors in nightlife spots, Susukino in Hokkaido and Kabukicho in Tokyo 17-25 September, 2020 before, during and after 4-day holidays.

As a general trend, the September data shows that Human Mobility decreased in both nightlife spots during 4-day holidays. However, the number of the visitors from the outside of Hokkaido was doubled in Susukino, compared with the regular weekdays.

Multiagent-based Simulation of the virus spreads: By using the nation–wide GPS mobility data in Japan

Simulation of the virus spreads

Sim-1 (blue line) failed to estimate the infected cases after June 2020 due to iterative use of its short-term data. Sim-2 (green line) estimated the explosive outbreak around April 25, 2020, by using the daily mobility data.

Project prospects:

  • Speed up the simulation by increasing the time resolution.
  • Assimilate the mobility data as a key parameter of the simulation along with changeable infection rates (data assimilation).
  • Forecast the virus spreads with the updated data. 

Visualization of the human-to-human infection networks

The visualization shows the forecasts the number of people that an infected case has transmitted the infection to (effective reproduction number).

Project prospects:  

  • Visualize the simulation outcomes with high accuracy.
  • Perform comparative analysis of the confirmed cases and the forecasted cases.

In this project, we have simulated the spreads with long-term mobility data, and visualized the human-to-human infection networks. We will conduct data assimilation along with changeable trends and comparative analysis to evaluate the simulation results for the better.