9:30 AM - 9:45 AM
[MGI29-03] Response anomaly detection in count rate derived from terrestrial radiation measured by ultraviolet space telescope detector
Keywords:Anomaly detection, Earth magnetosphere, Machine learning
Since the discovery of the Earth's radiation belts in the late 1950s, many satellites have made plasma measurements to understand the production, transport, and loss processes of energetic particles. Particles in the radiation belt are trapped by the Earth's intrinsic magnetic field and enter the upper atmosphere connected to the radiation belt via magnetic field lines. We have used data from the Hisaki spectroscopic planet observatory satellite to detect counts of energetic particles originating from the Earth's radiation belt at altitudes of about 1000 km in the exosphere. The spectral region on the MCP (microchannel plate) detector is limited to the center, and dark counts in the outer regions can be used to monitor radiation.
We Analyzed the 2014-2018 observational data and detected several sharp increases in the dark count rate (2-5 times the normal rate) that were discernible to the human eye. We attempted to solve the response anomaly detection problem in order to detect not only large fluctuations but also small increases that are difficult to detect visually and to probe the causes of the fluctuations. Linear multiple regression was applied to the dark count time series using satellite geographic parameters (latitude, longitude, geocentric distance, and local time) as explanatory variables. Data points outside the confidence interval of the resulting prediction model were detected as anomalies. Next, to determine whether the anomalies were caused by solar wind disturbances or not, a regression was performed using SYM-H, which is used as a space weather indicator, and observation data (electrons, protons, X-rays, and magnetic field strength) from the GOES satellite orbiting the earth at an altitude of about 36,000 km (geostationary orbit) as new explanatory variables. SHAP (SHapley Additive exPlanations) values were then calculated for each anomaly, and the contribution of the above features was evaluated.
We Analyzed the 2014-2018 observational data and detected several sharp increases in the dark count rate (2-5 times the normal rate) that were discernible to the human eye. We attempted to solve the response anomaly detection problem in order to detect not only large fluctuations but also small increases that are difficult to detect visually and to probe the causes of the fluctuations. Linear multiple regression was applied to the dark count time series using satellite geographic parameters (latitude, longitude, geocentric distance, and local time) as explanatory variables. Data points outside the confidence interval of the resulting prediction model were detected as anomalies. Next, to determine whether the anomalies were caused by solar wind disturbances or not, a regression was performed using SYM-H, which is used as a space weather indicator, and observation data (electrons, protons, X-rays, and magnetic field strength) from the GOES satellite orbiting the earth at an altitude of about 36,000 km (geostationary orbit) as new explanatory variables. SHAP (SHapley Additive exPlanations) values were then calculated for each anomaly, and the contribution of the above features was evaluated.