Twenty‐four hour predictions of f0F2 using time delay neural networks

Abstract

The use of time delay feed‐forward neural networks to predict the hourly values of the ionospheric F2 layer critical frequency, f0F2, 24 hours ahead, have been examined. The 24 measurements of f0F2 per day are reduced to five coefficients with principal component analysis. A time delay line of these coefficients is then used as input to a feed‐forward neural network. Also included in the input are the 10.7 cm solar flux and the geomagnetic index Ap. The network is trained to predict measured f0F2 data from 1965 to 1985 at Slough ionospheric station and validated on an independent validation set from the same station for the periods 1987–1990 and 1992–1994. The results are compared with two different autocorrelation methods for the years 1986 and 1991, which correspond to low and high solar activity, respectively.