8. Deep Learning for Mid-term Forecast of Daily Index of Solar 10.7 cm Radio Flux

For the mid-term forecast of daily index of solar 10.7 cm radio flux with deep learning method, a neural network based on the classical multi-layer perception model is proposed. The network contains only one hidden layer with 90 neutrons, and an autoregressive model of time series is implemented non-parametrically. In the forecast, the historical daily indices as well as the historical forecast error are considered. The model gives the forecast of next 27 days with the values of past 27 days. The networks are trained and validated with historical data over 50 years, and the result clearly shows that compared to the traditional methods, the mean relative error is significantly reduced. Unlike most of the previous studies, in which the model parameters need to be rolling updated, while with this model the parameters are fixed after the training. The proposed model greatly simplifies the daily operation of forecast, and is extremely advantageous to the promotion in other applications.