Forecasting Ionospheric foF2 Based on Deep Learning Method

In this paper, a deep learning long-short-term memory (LSTM) method is applied to the forecasting of critical frequency ionosphere F2 layer (foF2). Hourly values foF2 from 10 ionospheric stations in China and Australia (based on availability) 2006 2019 are used for training verifying. While 2015 exclusive verifying accuracy. The inputs LSTM model sequential data previous values, which include local time (LT), day number, solar zenith angle, sunspot number (SSN), daily F10.7 flux, geomagnetic Ap Kp indices, geographic coordinates, neutral winds, observed value at moment. To evaluate ability model, two different neural network models: back-propagation (BPNN) genetic algorithm optimized backpropagation (GABP) were established comparative analysis. parameters forecasted under quiet disturbed conditions during activity maximum (2015) minimum (2019), respectively. results these models compared with those international reference (IRI2016) measurements. diurnal seasonal variations 4 analyzed 8 selected verification stations. reveal that presents optimal performance all series foF2, while IRI2016 has poorest performance, BPNN GABP between them.