A Hybrid Model for Forecasting Sunspots Time Series Based on Variational Mode Decomposition and Backpropagation...

Area of Science:

• * Astrophysics and space weather.
• * Time series analysis and forecasting.
• * Computational intelligence and machine learning.

Background:

• * Sunspot number variations significantly influence Earth's climate, agriculture, communications, and natural disasters.
• * Accurate sunspot prediction is vital for mitigating potential impacts.
• * The chaotic nature of sunspot time series presents a forecasting challenge.

Purpose of the Study:

• * To propose a novel hybrid model for forecasting monthly mean sunspot numbers.
• * To address the chaotic characteristics of sunspot time series data.
• * To improve the accuracy of sunspot number predictions.

Main Methods:

• * Variational Mode Decomposition (VMD) was employed to decompose the sunspot time series into intrinsic mode functions (IMFs).
• * A backpropagation (BP) neural network was optimized using the firefly algorithm (FA) for initializing weights and thresholds.
• * Individual prediction models were established for each IMF, with final predictions obtained by combining component forecasts.

Main Results:

• * The proposed VMD-FA-BP hybrid model demonstrated higher prediction accuracy compared to standalone BP, FA-BP, and EMD-BP models.
• * Simulation results validated the effectiveness of the hybrid approach in forecasting sunspot time series.
• * The method successfully captured the complex dynamics of sunspot number fluctuations.

Conclusions:

• * The developed VMD-FA-BP model offers a robust and accurate method for sunspot number forecasting.
• * This approach can be effectively applied to predict complex time series data with chaotic characteristics.
• * Enhanced sunspot prediction capabilities can aid in preparedness for climate and technological impacts.