Forecasting Solar Cycle 26 with Multiple Machine Learning Techniques

doi:10.1088/1538-3873/ae541c

Forecasting Solar Cycle 26 with Multiple Machine Learning Techniques

2026-04-10

0

OA

AI

DOI：10.1088/1538-3873/ae541c

Original

Save

View PDF

Abstract

En 中文

We present a detailed study of the prediction of the future Solar Cycle (SC) 26. We consider the following machine learning techniques (ML): (1) The Nonlinear Auto Regressive eXogenous (NARX) algorithm, which previously showed good performance in predicting SC 25, and (2) Ensemble Learning techniques that combine different learning algorithms with different bias-variance characteristics to improve model accuracy by preventing the common overfitting problems in ML models. In addition, the well-known inverse relationship between the activity rise time and amplitude (Waldmeier effect) is considered in the prediction of SC 26. We feed the models with time series of the sunspot numbers of all previously recorded 24 solar activities. These algorithms were first verified through the prediction of SC 25 before their application to SC 26. The predictive results for SC 25 show a very good match with the available observations. We also present an updated fit of the Waldmeier inverse relationship between the activity rise time and amplitude by considering the time series of 24 solar cycles. The results obtained from different models show that the expected solar peak in terms of sunspot numbers (11 yr, monthly averaged) of SC 26 is 109 (2034 July).

Keywords:

Solar Cycle 26

Machine Learning

NARX Algorithm

Ensemble Learning

Waldmeier Effect

Forecasting Solar Cycle 26 with Multiple Machine Learning Techniques

Abstract

AI Summary

Journal

Organization