Improved GWO and its application in parameter optimization of Elman neural network

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Optimization

Background:

• Traditional neural networks struggle with complex optimization due to gradient descent limitations.
• Developing advanced optimization algorithms is crucial for improving neural network performance.

Purpose of the Study:

• To propose an improved grey wolf optimizer (SGWO) for enhanced neural network structure optimization.
• To introduce a novel prediction method, SGWO-Elman, by applying SGWO to Elman networks.
• To mathematically analyze SGWO convergence and experimentally validate its optimization and prediction capabilities.

Main Methods:

• Enhanced grey wolf optimizer (SGWO) with circle population initialization, information interaction, and adaptive position updates.
• Application of SGWO to optimize the structure of Elman neural networks.
• Mathematical convergence analysis of SGWO using Markov chain theory.
• Comparative experiments to evaluate SGWO's optimization and SGWO-Elman's prediction performance.

Main Results:

• SGWO demonstrated a global convergence probability of 1, functioning as a finite homogeneous Markov chain.
• SGWO exhibited superior optimization performance on complex, multi-dimensional functions compared to standard methods.
• SGWO effectively optimized Elman network structures, leading to accurate prediction performance in the SGWO-Elman model.

Conclusions:

• The proposed SGWO algorithm offers robust optimization capabilities for complex problems.
• The SGWO-Elman model provides accurate predictions, outperforming traditional methods for Elman network optimization.