| 研究実績の概要 |
A code for an Agent-Based Economic Model (ABEM) was implemented based on latest literature. Lack of numerically efficient codes to simulate millions of agents is a major challenge in agent based simulations. To address this issue, a distributed memory parallel High Performance Computing (HPC) extension was developed. By enhancing the algorithms and data-structures, we could significantly reduce the run-time of the developed HPC extension. As an example, the HPC extension is capable of simulating a small economy with 10 million agents in 18 seconds, which is significantly faster than the best reported in literature. Further, the code is designed to be flexible to integrate with any physics based disaster simulator.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
Most of the work progressed as planned, except the planned tasks with the foreign collaborator. The Agent based model was successfully extended to according to the latest publications in ABEM, and an efficient HPC extension was developed. As planned, the developed HPC extension is capable of simulating a several ten millions of agents within sevaral minutes, which enables large scale ensemble forecasting. It was planned to modify the developed ABEM to simulate an economy after a major natural disaster like earthquakes or floods, under the guidance of a foreign collaborator. However, it has to be postponed to FY 2019, due to changes of foreign collaborator’s plans. the model improvement towards disaster simulation has to be postponed to 2019.
|
| 今後の研究の推進方策 |
Three main tasks are planned for FY2019: 1) modify the developed ABEM to simulate economies hit by a natural disaster; 2) enhance the HPC extension to simulate large economies; 3) and integrate with physics based simulators.
The code developed in FY2018 is based on a model developed for simulating ordinary economic conditions, and modifications are necessary to simulate economies following a natural disaster.
The HPC extension developed in FY2018 is efficient only for small economies with several 10 million agents. Further enhancements are necessary to simulate 1:1 scale model of Japanese economy, which involves more than 130 million agents. We are planning to enhance the code to simulate 150+ million agents by developing efficient algorithms and data-structures.
|
