2020 Fiscal Year Annual Research Report
Development of an HPC enhanced system for finding fine-grained lifeline recovery plans and evaluating their long term economic performance
Project/Area Number |
18H01675
|
Research Institution | The University of Tokyo |
Principal Investigator |
|
Project Period (FY) |
2018-04-01 – 2022-03-31
|
Keywords | economic simulations / agent based model / earthquake disaster / big-data / machine learning |
Outline of Annual Research Achievements |
Analyzing the data from e-Stat, Bank of Japan, and System of National accounts of Japan, we generated the basic input data for simulating the Japanese economy. This input data set could reproduce the past records to reasonable degree. The agent based simulators parallel computing capabilities are further enhanced by improving its shared memory computing capabilities such that super computers like Fugaku, which has high CPU core count and limited amount of system memory, can be efficiently utilized. The developed HPC-enhanced economic simulator is coupled with the Integrated Earthquake Simulator so that the ground motion due to a given earthquake wave, the corresponding seismic response of each individual building in a large area, and the national economic following the simulated disaster can be seamlessly simulated. We improved the recovery plan generating code to generate better recovery plans for discrete problems. Further, the computational capability of the recovery plan generator was enhanced by implementing distributed parallel extension to search for solutions on parallel fronts. The code for recovery plan generation under network constraint also was improved to reduce runtime and generate more optimal solutions. Using the developed system, we conducted a preliminary simulation of an earthquake disaster in Kobe region and simulated the economy. Further, we developed a LSTM based machine-learning model to improve the decision making capabilities of the agents.
|
Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
We completed all the planned items successfully, except the utilization of data from Teikoku DataBank. The data from Teikoku DataBank could not be utilized due to incompatibilities with the data from e-Stat. As planned, we successfully simulated an earthquake disaster in Kobe area and the national economy following the simulated disaster.
|
Strategy for Future Research Activity |
The developed system will be further tightly integrated with IES utilizing PACT so that economic aftermaths of a major earthquake disaster can be realistically simulated considering each building’s repair cost, repair time, probability of collapse, etc. Further, we plan to generate a more consistent input data set based on statistics from government portals so that the Japanese national economy can be simulated to a higher accuracy compared to the results of this fiscal year. We plan to analyze the progress of Japanese economy due to the Nankai Trough Giant Earthquake Model provided by the Cabinet office. Due to the limitations of availability of data, only the Osaka and Kobe area, involving about 1.8 million buildings, is to be simulated.
|
Research Products
(3 results)