Study of light curves using machine learning techniques

Abstract

The ability to accurately predict solar activity has become increasingly critical in esti mating the various consequences of Space Weather, such as disruption of communication and navigation systems, damage to spacecraft, and increased radiation exposure for as tronauts and airline passengers. This task of predicting the future of any system relies on collecting and inferring information about the source (The Sun) from its time series. Over time, specialized methods for time series analysis have been invented, Modeling is one of these techniques and serves as a powerful tool to determine the correlations and laws generating the data[23]. To achieve accurate predictions of the amplitude and timing of the next solar cycle, various techniques have been employed, such as neural networks, ARIMA, Fourier techniques, and geomagnetic precursor methods. While some of these methods have shown promising results, the prediction of solar activity remains a complex and challenging task. This is where Gaussian process regression can serve as a powerful tool. It provides a range of models instead of a single rigid model. In this thesis, the application and drawbacks of Gaussian process regression in time series modeling and prediction will be explored. Gaussian process regression is employed to model Sunspots time series and AGN light curves. Maximization of log-likelihood for optimal kernel is adopted and based on the Best kernel results, the 25th Sunspot maximum is predicted. As another exciting possible application, the light travel time lag for a gravitationally lensed quasar is modeled and requires further analysis for extraction of physically significant parameters.