Design and Optimization of a High Sensitivity Wide Range U-Shaped Polished Photonic Crystal Fiber Sensor Based on Machine Learning and Genetic Algorithm

Abstract

In this work, a U-shaped polished PCF-SPR sensor is designed, and a machine-learning-assisted optimization strategy is developed to improve the structural design efficiency. The proposed optimization framework integrates a genetic algorithm optimized multilayer perceptron (GA-MLP) surrogate model with genetic algorithm (GA)-based structural parameter optimization. A finite-element-method-based dataset is generated by systematically varying the key structural parameters of the proposed sensor, and a GA-MLP surrogate model is established to predict the sensing performance within the investigated parameter space. The optimized model enables efficient exploration of the sensor structure and achieves accurate prediction of the sensing response. Based on the optimized structural configuration, the proposed sensor exhibits a sensitivity of 5947.375 nm/RIU in the primary detection range and achieves a maximum local sensitivity of 10,100 nm/RIU at a refractive index of 1.40 within the extended refractive index detection range. The optimized results are further verified through full-vector finite element simulations, showing good agreement between the predicted and simulated sensing characteristics. The proposed U-shaped PCF-SPR sensor combined with machine-learning-assisted optimization provides an efficient approach for designing high-performance photonic sensing devices and accelerating the optimization of complex fiber structures.