Estimating the SEP Flux for the Upcoming Solar Cycle 25 Using LSTM Network

Estimating space weather parameters for the solar cycle 25, which has already started, is essential to anticipate the behavior of the near-Earth space environment. Artificial Neural Networks have in recent years become very widely used in several scientific fields owing to the advancement in computational power and the availability of big data. In this work, we take advantage of utilizing Recurrent Neural Network models in time-series analysis. We have developed and trained a Long-Short Term Memory (LSTM) model, in order to make long-term predictions of the hourly-averaged energetic proton fluxes at 1AU. We have used as input a combination of solar and interplanetary magnetic field indices (from the OMNI database) from the past four solar cycles and generated predictions of the solar energetic proton fluxes at three energies. So far, we found that the root-mean-square errors for the predictions over a three-month period were 0.0240, 0.0173, and 0.0309, respectively. We also found that the model underestimates the prediction at the highest energy band. We will extend the model architecture in order to estimate the future SEP fluxes over the whole solar cycle.