Nano-Flare Analysis using Machine Learning Techniques

Coronal heating is one of the long-standing problems in solar physics. So far, two primary mechanisms have been proposed to explain how the corona is heated, namely small-scale magnetic reconnection and wave dissipation. To estimate the contribution of small-scale magnetic reconnection, so called nano-flares, to heat the corona is crucial to solve the coronal heating problem. To reach this goal, we apply machine learning techniques Deep Learning (DL) and Genetic Algorithm (GA) to nano-flare analysis. For both methods, we carry out a one-dimensional hydrodynamic simulation of a coronal loop heated by nano-flares. We observe its temporal variations of EUV and soft X-ray intensities using Hinode/XRT and SDO/AIA in a pseudo-manner. For DL method, we carry out the simulation and pseudo-observation thousands of times with various nano-flares. Using these data, we train Deep Neural Network (DNN) to output the energies and occurrence times of nano-flares as a response to the input of light curves. For GA method, we randomly make genes which have information of energies and occurrence times of nano-flares. With these information as input, we carry out the simulations and pseudo-observations. We evaluate the validity of each gene by comparing pseudo-observed light curves to those of actual observations. Genes make the next generation which reconstruct observed light curves better according to the validities. After repeating these procedures many times, we obtain the gene with the nano-flare information which can reconstruct observed light curves best. Using these methods, we analyze nano-flares which cannot be detected by pre-existing method.