Proton Flux Forecasting in Astrophysics: A Deep Learning Approach Using Long Short-Term Memory Networks

Research on cosmic ray proton flux serves multiple purposes in the study of space weather and solar activity as well as high-energy astrophysics. The correct prediction of this flux remains fundamental to protect satellite infrastructure, space research activities, and orbit communication networks. An LSTM network-based approach forecasts daily cosmic ray proton flux because of its proven success in time-series forecasting tasks. Daily proton flux measurements exist in the dataset using standardized units (m⁻<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>sr⁻¹s⁻¹GV⁻¹) which maintain both reliability and consistency of data. The main purpose of this study involves both precision enhancement of predictions and temporal trend detection within the proton flux domain. An LSTM-based system was established to develop a dependable tool for space weather risk management. Preprocessing procedures started with missing values handling and were followed by date formatting and addition of time-based features, which included year, month, and day of the year. Additional lagged features were used to incorporate past trends of proton flux data. We normalized the data based on MinMaxScaler methods before partitioning it. Training with early stopping for model prevention of overfitting took place when using the two-layer LSTM model with dropout regularization.The model test incorporated Mean Squared Error together with R2 score to measure its efficiency. The predictive model based on LSTM achieved excellent results which showed an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> score of 0.961 alongside a minimal MSE value. The studies demonstrate LSTM neural networks have strong capabilities for forecasting cosmic ray flux powerfully and reliably.