https://www.spaceappschallenge.org/2023/find-a-team/taitatumu/?tab=project
Our project presents a comprehensive predictive workflow based on machine learning, offering a solution for predicting geomagnetic storms hitting Earth. The challenge description highlighted the core issue of the existence of occasional faults in Deep Space Climate Observatory (DSCOVR) raw data, which requires additional costs to process. Given the importance of making full utilization of data in the fields of data science and machine learning, we incorporated abnormal values detection and explored the relationship between anomalies and solar activity. By leveraging all available open data and using both "normal" and "abnormal" spectral data from DSCOVR raw data as inputs, our pipeline feeds them into separate models to predict solar wind properties such as density, temperature, and velocity. Subsequently, the predicted data is used to forecast the Planetary K-index (Kp), determining the occurrence and duration of geomagnetic storms. Our system provides early warnings for geomagnetic storms, reducing the risks they pose to communication systems, navigation equipment, power grids, and even space missions. In summary, our project focuses on three key aspects: 1. Geomagnetic storm prediction. 2. Full utilization of open data and detection of abnormal values.