Forecasting Sunspot Numbers with Neural Networks

This paper presents a feedforward neural network approach to sunspot forecasting. The sunspot series were analyzed with feedforward neural networks, formalized based on statistical models. The statistical models were used as comparison models along with recurrent neural networks. The feedforward networks had 24 inputs (depending on the number of predictor variables), one hidden layer with 20 or 30 neurons and one neuron on the output layer. The networks were trained using the backpropagation algorithm. As a result, I found that feedforward neural networks are much better forecasters than recurrent neural networks and statistical models. KeywordsNeural networks, Time series analysis, Forecasting, Prediction, Statistical models, Sunspots, Autoregressive models.