| 研究課題/領域番号 |
20K19782
|
|---|
| 研究機関 | 東京大学 |
|---|
研究代表者 |
FAN ZIPEI 東京大学, 空間情報科学研究センター, 特任講師 (70835397)
|
|---|
| 研究期間 (年度) |
2020-04-01 – 2022-03-31
|
| キーワード | Mobility Prediction |
|---|
| 研究実績の概要 |
This research aims at making an accurate prediction of human mobility at a metropolitan scale. Both regular and irregular human mobility are taken into consideration to guarantee that our prediction is robust to different real-world situations.
So far, we developed 1) a novel online adaptive ensemble model for traffic volume prediction, and 2) a crowd-context based individual level human mobility prediction using a meta-learning paradigm. Both of the proposed models are tested on real-world data sets, and outperform the state-of-the-art models. Online real time human mobility prediction systems with real world mobile phone user GPS data streams are under development.
We summarized the current achievements as conference papers and submitted to top-tier artificial intelligence conferences.
|
| 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
So far, I have achieved most of the expected achievements as planned, and the research is proceeding smoothly.
For the online fusion part, the online adaptive ensemble model has shown its power in modeling different global states of human mobility. In extension to this, a crowd-context model also shows its flexibility in modeling regular and irregular mobility patterns simultaneously.
For the unprecedented irregular mobility, I am working on a multi-agent based simulation approach on simulating unprecedented irregular mobility, and it will be easily integrated into the ensemble framework in this fiscal year.
|
| 今後の研究の推進方策 |
In this fiscal year, I will focus on two rest problems: 1) irregular human mobility prediction. Particularly, we will study the human mobility model during an extreme natural disaster such as an earthquake. Different situations such as the public transportation system stops on a weekday will be parameterized and the human mobility in response to these will be simulated based on a multi-agent model. 2) Integrate the simulation results (generated offline) with our online adaptive ensemble prediction system.
|
| 次年度使用額が生じた理由 |
1) Upgrade the current computing server, including enlarging the storage space, more RAM, and more GPUs if necessary.
2) Cloud storage and processing data, using Tencent Cloud services for processing the data in China and Google Cloud services for data in Japan or other countries.
3) Human movement sensing devices for collecting more sources of data, including UWB and BLE beacons, and mobile devices collecting the data.
|
