download0 view134
twitter facebook

공공누리This item is licensed Korea Open Government License

End-to-End 딥러닝 기법을 적용한 데이터 기반 태풍 진로 예측 연구
Alternative Title
A Research for Typhoon Track Prediction using End-to-End Deep Learning Technique
Korea Institute of Science and Technology Information
Publication Year
funder : 과학기술정보통신부
funder : Ministry of Science and ICT
□ 전지구 위성영상 빅데이터를 활용한 데이터 기반 태풍진로 예측 딥러닝 모델 개발
◦ 위성영상 기반 태풍 중심 탐지 시스템 개발 (GlobeNet)
◦ 시계열 위성영상 예측 시스템 개발 (PSIque)
◦ 위성영상 학습 오토인코더 개발
◦ 수치모델 기반 태풍진로예측 딥러닝 모델 개발 (DeepTC)
□ 대용량 위성영상 분석에 최적화된 분산 딥러닝 프레임워크 설계 및 테스트베드 구성
◦ 분산 Tensoflow 활용 딥러닝 프레임워크 개발
◦ 30노드 이상의 분산 딥러닝 프레임워크 운용

(출처 : 보고서 초록 3p)

IV. Results of the study
◦ Development of tropical cyclone center detection system based on satellite image (GlobeNet)
- Construction of convolution neural network and linear regression application model
- Locations of multiple tropical cyclone centers (up to 6) and confidence probability simultaneous recognition
◦ Development of time-series satellite prediction system (PSIque)
- Construction of sequence to sequence prediction model for SkipConx based memory network
◦ Development of auto encoder learning satellite image
- Development of learning module for latent vector of cloud satellite image
- Improved performance after using as additional information of deep learning model
◦ Development of a numerical model-based tropical cyclone trajectory prediction deep-learning model (DeepTC)
- Establishment of WRF (Weather Research and Forecast) data utilization model using ensemble technique
◦ Development of distributed deep learning framework and development of operation technology
- Multi-node distributed deep learning framework using distributed Tensorflow

(출처 : SUMMARY 7p)
태풍; 진로예측; 딥러닝; 데이터 기반; 종단간 모델; Tropical Cyclone; Trajectory Prediction; Data-driven; End-to-End Model
Files in This Item:
There are no files associated with this item.
Appears in Collections:
7. KISTI 연구성과 > 연구보고서 > 2017
RIS (EndNote)
XLS (Excel)