Sunspot Occurrence Forecasting With Metaheuristic Optimized Recurrent Neural Networks

Abstract

Solar activity plays an important role when considering terrestrial communication. Solar activity plays an even more important role when working with communication systems relying on artificial satellites. Electromagnetic emissions, known as solar flares, can disrupt communications and damage important infrastructure. However, as powerful solar flares are often preceded by observable occurrences of sunspots, a robust system prognosis system can be leveraged to help improve chances of minimizing damage to infrastructure. This work explored the potential of recurrent neural networks coupled with an introduced modified metaheuristics algorithm to tackle the increasingly pressing challenge of forecasting solar activity based on historical data. Due to the heavy reliance of neural networks on proper parameter selections as well as adequate architectural structure, the introduced optimizer is leveraged for the optimization of these control parameters. The proposed approach is evaluated on a real-world dataset that is publicly available. The outcomes are compared to several well-established optimization metaheuristics, and the outcomes show great promise for tackling this increasingly important topic as we head into the peak of the current solar cycle.