http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Detecting Exoplanets from the Shape of Data: A TDA-based Approach to Kepler Light Curves
정대원 계명대학교 자연과학연구소 2025 Quantitative Bio-Science Vol.44 No.1
Detecting exoplanets from stellar light curves remains a challenging task because of the presence of noise, complex temporal patterns, and significant class imbalances. In this study, we propose a novel classification framework that applies topological data analysis (TDA) to a Kepler light curve time series to extract robust noise-resistant features. By computing persistence diagrams from time-delay embedded flux data and vectorizing them into entropy, landscape, and amplitude descriptors, we capture the essential topological structures related to planetary transits. These features are combined with the synthetic minority over-sampling technique (SMOTE) to mitigate class imbalances and are evaluated using several machine learning classifiers. The study results showed that the TDA-based features, particularly when paired with Random Forest and SVC, significantly improved the sensitivity and F1-score for exoplanet detection, outperforming traditional models trained on raw flux data. This approach highlights the potential of topological methods to enhance time series classification in astrophysical applications.
정대원,최해진,이선구 한국항공우주학회 2015 한국항공우주학회 학술발표회 논문집 Vol.2015 No.11
우리나라 지상시스템은 위성 관제와 탑재체자료 처리를 위해서 1990년대부터 개발되기 시작하였다. 과학위성, 저궤도위성 및 정지궤도 지상시스템의 초기 개발 시에는 국제기술협력 등이 필요하였으나, 초기 기술 확보 이후에는 국내 독자 기술로 개발되기 시작하였다. 우리나라는 지금까지 총 14기의 지상시스템을 개발하였으며, 본 논문은 이러한 경험을 바탕으로 발전 전망 및 방향을 제시한다. 관제시스템은 재구성, 재사용, 자동화 기술을 기반으로 하는 다중위성 관제시스템 개발이 필요하다. 처리시스템은 다양한 탑재체자료 처리 기술의 확보와 처리정확도 향상을 위한 검보정 기술의 접목이 필요하다. 또한, 탑재체자료 양의 증가로 인한 고속처리 기술 확보와 처리된 탑재체자료를 사용자 간에 사용하고 통합하는 상호운영성 연구가 필요하다. 끝으로, 다양한 종류의 많은 위성을 운영할 수 있도록 국내외에 분포된 안테나 등 국가 지상시스템 인프라 망 구축이 필요하다. Our country"s ground systems have started to be developed for mission control and payload data processing since 1990s. International technology cooperations were needed in the early development phase of ground system for science experiment satellite, LEO satellite and GEO satellite and then they have been developed as domestic own technology since acquiring early technology. Our country has developed total 14 ground systems until now, this paper suggests prospect and direction on ground system development in the base of such development experiences. Mission control system is needed to develop multi-satellite mission control system in the base of technology of re-configure, re-use and automation. Processing system is needed to acquire processing technology for kinds of payload sensor data and study interoperation to integrate and use outputs which are processed between users. Finally, national ground system infrastructure is needed to operate kinds of lots of satellites at worldwide area.
정대원,황욱연 한국품질경영학회 2024 품질경영학회지 Vol.52 No.4
Purpose: Leveraging on the contemporary machine learning algorithms, we would like to improve the prediction performance of the existing MLR(MultipleLinearRegression)modeltopredictthebloodhemoglobinlevels. Methods: The GBDT(Gradient Boosting Decision Trees) such as the XGBoost(Extreme Gradient Boosting), the LightGBM(Light Gradient Boosting Machine), and the CatBoost(Categorical Boost), the RF(Random Forests), and the MLP(Multi-Layer Perceptron) are adopted to build the new prediction models. Results: The machine learning algorithms provide prediction performance better than the existing prediction model. Conclusion: The proposed prediction models can be considered as an alternative better than the existing prediction model.