Optimizing Deep Learning Models with Improved BWO for TEC Prediction.

Area of Science:

• Geophysics and Space Science
• Artificial Intelligence
• Computational Science

Background:

• Accurate prediction of total ionospheric electron content (TEC) is vital for space weather and wireless communication.
• Deep learning models show promise for TEC prediction but require extensive hyperparameter optimization.
• Existing swarm intelligence algorithms like Beluga Whale Optimization (BWO) can be prone to local minima.

Purpose of the Study:

• To address the challenge of hyperparameter optimization in deep learning models for TEC prediction.
• To propose an improved swarm intelligence algorithm, Firefly Assisted Multi-strategy Beluga Whale Optimization (FAMBWO), to overcome limitations of existing methods.
• To develop and evaluate an automated machine learning framework incorporating FAMBWO for enhanced TEC prediction.

Main Methods:

• Developed FAMBWO by integrating multi-strategy improvements into the Beluga Whale Optimization (BWO) algorithm.
• Benchmarked FAMBWO against 11 state-of-the-art swarm intelligence algorithms on 30 benchmark functions.
• Proposed the FAMBWO-MA-BiLSTM framework, utilizing FAMBWO for hyperparameter optimization of a MA-BiLSTM model for TEC prediction.

Main Results:

• FAMBWO demonstrated superior convergence speed and solution quality compared to 11 other algorithms on 30 benchmark functions.
• The FAMBWO-MA-BiLSTM framework significantly outperformed models optimized via grid search, random search, Bayesian optimization, and the original BWO.
• FAMBWO-optimized MA-BiLSTM achieved demonstrably better predictive performance for TEC.

Conclusions:

• FAMBWO is an effective and improved swarm intelligence algorithm for optimization tasks.
• The proposed FAMBWO-MA-BiLSTM framework offers a robust solution for automated hyperparameter optimization in deep learning-based TEC prediction.
• This work advances the accuracy and efficiency of TEC prediction models through advanced optimization techniques.