Research Article Open Access

A Hybrid Deep Learning Model for Space Radiation Dose Rate Prediction

Abstract

The prediction of space radiation dose rates holds significant importance for space science research. In this paper, a hybrid neural network-based approach for forecasting space radiation dose rates is proposed, utilizing a dataset of 4,174,202 in orbit measurements collected over a 12-month period from satellites. During data pre-processing, a first derivative wavelet transform is applied to retain trend information and perform noise reduction. In model design, the FDW-LSTM model is introduced, combining the First Derivative Wavelet (FDW) transform with Long Short-Term Memory (LSTM) networks. Experimental results demonstrate a coefficient of determination (R²) of 0.97 between the predicted values and actual measurements for the FDW-LSTM model. Compared to the Mean Absolute Deviation (MAD), 3-Sigma Rule (3σ), and Quartile methods, the FDW-LSTM model yields an average increase of 0.2 in R². Additionally, compared to the predictions of the GRU and RNN neural networks, the FDW-LSTM model achieves an improvement of 0.12 and 0.54 in R², respectively.

To read the full article Download Full Article | Visit Full Article