Enhanced grey wolf algorithm for automatic tuning of an ensemble neural network in predicting PCF optical properties.

Abstract

The structural characteristics of photonic crystal fibers (PCFs) determine their optical properties. This paper introduces an enhanced Grey Wolf Optimization algorithm termed ACD-GWO, which proposes adaptive strategies, chaotic mapping and dimension-based approaches and integrates them into the Grey Wolf Optimization framework. The aim is to achieve efficient automatic adjustment of hyperparameters and architecture for ensemble neural networks. The resulting ensemble neural network demonstrates accurate and rapid prediction of optical properties in PCFs, including effective refractive index, effective mode area, dispersion, and confinement loss, based on the PCF's structural characteristics. Compared to random forest and feedforward neural network models, the ensemble neural network achieves higher accuracy with a mean squared error of 3.78 × 10-6. Additionally, the computational time is significantly reduced, with only 2.27 minutes required for training and 0.08 seconds for prediction, which is much faster than numerical simulation software. This will provide new possibilities for optical device design and performance optimization, driving cutting-edge research and practical applications in the field of optics.