Sunspot interval prediction based on fuzzy information granulation and extreme learning machine

• Published in: Journal of astrophysics and astronomy Vol. 41; no. 1
• Main Author: Lingling, Peng
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 01.12.2020; Springer Nature B.V
• Subjects: Artificial neural networks; Astronomy; Astrophysics and Astroparticles; Data analysis; Empirical analysis; Granulation; Machine learning; Mathematical models; Observations and Techniques; Physics; Physics and Astronomy; Prediction models; Solar physics; Space weather; Sunspot numbers; Sunspots; sunspot; complementary ensemble empirical mode decomposition; point prediction; Interval prediction; fuzzy information granulation; extreme learning machine
• ISSN: 0250-6335; 0973-7758
• DOI: 10.1007/s12036-020-09649-4

Sunspot prediction is an important task for space weather and solar physics. Traditional point forecast may not be sufficiently satisfactory and reliable. To quantify the uncertainty of point prediction, a hybrid interval prediction model has been proposed for sunspot forecasting. Three major steps are taken: (1) the complementary ensemble empirical mode decomposition (CEEMD), to decompose the sunspot sequence into a series of modal components, (2) the fuzzy information granulation (FIG), to extract the minimum, average and maximum value of each window, and (3) the extreme learning machine (ELM), to conduct point prediction and interval prediction, superimposing the prediction values of all components as the final forecast results. The empirical study focus on the 13-month smoothed monthly sunspot number recorded by Solar Influences Data Analysis Center (SIDC) and show that the mixed model with the filtered CEEMD is more effective than the unfiltered one. It also enables us to track changes of the sunspot number with fast calculating speed and high accuracy both in point prediction and interval prediction.