Related Experiment Video
Updated: Jul 6, 2025
11:08
Fabrication And Characterization Of Photonic Crystal Slow Light Waveguides And Cavities
Published on: November 30, 2012
18.9K
Enhanced grey wolf algorithm for automatic tuning of an ensemble neural network in predicting PCF optical properties
Optics Express
|January 5, 2024
Summary
Researchers developed an enhanced Grey Wolf Optimization (ACD-GWO) algorithm to optimize ensemble neural networks for predicting photonic crystal fiber (PCF) optical properties accurately and rapidly.
Area of Science:
- Optics and Photonics
- Computational Intelligence
- Materials Science
Background:
- Photonic crystal fibers (PCFs) possess unique optical properties determined by their structural characteristics.
- Accurate prediction of these properties is crucial for optical device design and optimization.
- Existing methods may lack efficiency in hyperparameter and architecture tuning for predictive models.
Purpose of the Study:
- To introduce an enhanced Grey Wolf Optimization algorithm (ACD-GWO) for automated hyperparameter and architecture adjustment.
- To develop an ensemble neural network capable of precise and swift prediction of PCF optical properties.
- To demonstrate the superiority of the proposed method over conventional machine learning models and numerical simulations.
Main Methods:
- Implementation of an enhanced Grey Wolf Optimization algorithm (ACD-GWO) incorporating adaptive strategies, chaotic mapping, and dimension-based approaches.
- Development of an ensemble neural network model trained using the ACD-GWO optimized framework.
- Comparative analysis against random forest and feedforward neural network models, and numerical simulation software.
Main Results:
- The ACD-GWO optimized ensemble neural network achieved high accuracy in predicting PCF optical properties (effective refractive index, effective mode area, dispersion, confinement loss) with a mean squared error of 3.78 × 10-6.
- Significantly reduced computational time: 2.27 minutes for training and 0.08 seconds for prediction.
- Demonstrated superior performance compared to random forest and feedforward neural networks.
Conclusions:
- The ACD-GWO algorithm effectively optimizes ensemble neural networks for predicting PCF optical properties.
- The developed model offers a faster and more accurate alternative to traditional numerical simulations.
- This advancement opens new avenues for optical device design, performance optimization, and research in optics.
