Download presentation
Presentation is loading. Please wait.
1
柴田一成(京都大学) Kazunari Shibata (Kyoto University)
太陽地球環境ビッグデータとICT を活用した実用的宇宙天気予報スキームの開発 Development of a space weather forecasting system using solar-terrestrial big data and ICT 柴田一成(京都大学) Kazunari Shibata (Kyoto University)
2
実際はようこう/LASCOのムービーで。
太陽と宇宙天気について、プレゼンターの柴田先生が一番慣れてる&調子が乗ってくるスタート
3
What is space weather? High energy particles X−ray/UV radiation
Disturbance of outer space environment driven by solar activity and resultant hazardous effects on human activities What comes from sun X−ray/UV radiation High energy particles Solar wind and coronal mass ejection Consequences Geomagnetic storm and aurora Power plant failure Satellite damage Radiation dose to astronauts Failure of satellite communication and navigation (GPS) Impact on climate (long term effect)
4
Space weather data: big and diverse (サイズが大きいだけじゃ無くて多種多様である、ということを英語でどう表現する?
Geomagketic field Radiation belt Solar images (e.g., Solar Dynamic Observatory produces 4000x4000 pixels images every 12 seconds. ~1.5TB / day) Ionospheric disturbances CHAIN 電波
5
Our goal To develop a space weather forecasting system that uses the solar-terrestrial big data and the advanced informatics such as machine learning and image processing. Integrate the big data of solar-terrestrial observations provided by various institutes with various formants to enable real-time and user-customized forecasting. To extract the quantitative index of the characteristics of the image data, such as solar magnetic field distribution, that are useful for space weather forecasting. To apply machine learning technologies to make the space weather forecasting system from past data. The system is further optimized by using genetic algorithms according to available data and user requirements.
6
Overview of the research plan
1) Integrate SW Big data 2) Feature selection by image processing 3) Forecast algorithm by machine learning sun solar wind radiation belt ionosphere Tuning by evolutionary computing (GA?) and feedback geomagnetism Accurate and customized forecast Needs satellite anomaly SW forecast users
7
1) Integrate space weather big data
Improve the existing meta-data base “IUGONET” for real-time Construct a new data base for satellitte anomaly and combine it to IUGONEET Strengthen connection to the other international SW big-data systems such as ESPAS and NASA’ virtual observatory => Provide SW big data for real-time forecast Allow user-customized forecast? Satellite anomaly (JAXA, private sector etc) IUGONET: Network of ground-based observations for SW
8
2) Feature selection by image processing *ここでいう画像認識技術は英語でどう表現すべき?
Flare-productive sunspot: simple, stable Non Flare-productive sunspot: complex, rapidly evolving How to quantify? coronal image 石井 Quantify the complexity of solar features, such as sunspot magnetic field and twisted magnetic fields Iteration and feedback with experienced solar physicists This allow us to use the various image data for machine learning
9
3) Forecasting system by machine learning
Generate 1-dimensional “feature vector” that characterize the solar-terrestrial condition (input from (2)) Generate a map from the feature vector to the forecast target (such as solar X-ray flux) using machine learning (such as support vector machine) The map can work as a forecaster of a time series variable (e.g., solar X-ray flux) from the available input data (such as solar magnetic fields) Time Feature vector forecast target t = 0 : [x1, x2, x3, ・・・, xn] → t = 1 : [x1, x2, x3, ・・・, xn] → t = 2 : [x1, x2, x3, ・・・, xn] → : t = N : [x1, x2, x3, ・・・, xN] → Tuning by selecting optimal set of the elements of the feature vector by genetic algorithm Y0 Y1 Y2 YN 精度向上対策 ・入力データ欠損 入力データ精度ばらつき対策
10
研究代表者グループの活動 研究全体の統括(柴田、特定助教) (2) ユーザーニーズの調査とフィードバック(石井、Berger、特定研究員)
(3) 社会経済及び地球環境への影響調査(水野、山敷、特定研究員) (4) 情報の公開(柴田、石井、Berger、特定助教)
11
Team Structure PI group K. Shibata (Kyoto U) M. Ishii (NiCT)
T. Berger (NOAA) Y. Mizuno (Kyoto WU) Y. Yamashiki (Kyoto U) CREST研究員 1) Big data T. Iyemori (Kyoto U) M. Nose (Kyoto U) A. Saito (Kyoto U) S . Ueno (Kyoto U) A. Asai (Kyoto U) H. Matsumoto (JAXA) R. Bernd (ドイツ) CREST研究員 3) Mechine learning S. Nemoto (BBT) T. Muranushi (RIKEN) S. Tsuruta (Tokyo Elec. U) CREST 研究員 2) Image processing H. Isobe (Kyoto U) A. Asai (Kyoto U) K. Kondo (Kyoto U) M. Kobashi (Hyogo PU) CREST研究員 分野・担当まで書く? ここの詳細は不要な気がするが。
12
アピールポイント 宇宙利用は急速に拡大している分野である。特に衛星測位や衛星地球観測は地理空間情報活用やInternet-Of-Thingsの技術との親和性が高く、様々なイノベーションが期待できる。宇宙天気予報はこれらの宇宙利用拡大にとって必須の情報であり、世界中で研究が加速している。 我々の予報システムは、遺伝的アルゴリズム(GA)と組み合わせたアクティブラーニングや 人工知能的手法(ディープラーニング/深層学習)を用いて、科学技術データや センサーデータの時系列画像から時系列数値の抽出、時系列画像から時系列画像の抽出等の、ビッグデータを用いた汎用予報エンジンとして利用できる可能性がある
13
説明ポイント 1) Big data available?
Currently, most of the data are taken by research institutes such as universities as scientific observations. Therefore, the data are mostly open and available online, but the problems are lack of common data-format and real-time availability 2) ツールやシステム、サービスとして公開できる範囲 => 根本さん他に確認 革新的アプリ技術と言えるの何か?その基本となる独創的アイディアは何か?
Similar presentations
© 2024 slidesplayer.net Inc.
All rights reserved.