Research On Prediction Of Optical Properties And Reverse Design Of Optical Fiber Based On Machine Learning And Optimization Algorithm

Orbital Angular Momentum(OAM)mode has important research value in opti-cal fiber communication field due to its unique optical properties.Different from common fiber eigenmodes,OAM mode is susceptible to interference during transmis-sion.Therefore,optical fiber structures which can support long-distance stable trans-mission of OAM mode should be specially designed to avoid the near degeneracy of vector eigenmodes.The traditional optical fiber design method is very dependent on the experience of the researcher.The fiber structural parameters are adjusted and op-timized repeatedly by observing and summarizing the potential relationships between the optical structure and characteristics.The design process is accompanied by a lot of redundant work,and it is difficult to obtain structures with extreme optical perfor-mance,so the cost of research and development is high.In this dissertation,taking OAM transmission fiber as an example,an efficient and accurate optical characteristic prediction neural network is established,and a flexible and practical optical structure reverse design method is proposed combined with the optimization algorithm.The proposed design method avoids the non-uniqueness of reverse design and the single solution problem of tandem neural network,providing a new idea for reverse design of optical structures with extreme performance.The main research work and innovation achievements are listed as fol-lows:1.An optical property prediction method for OAM transmission fiber based on machine learning is proposed.The data set of fiber structure and characteristics used for training neural network model is established,and the classification and prediction neural network model are built respectively for the prediction of optical characteristics of OAM mode.By adjusting the network structure and hyperparameters,the neural network model can predict the optical characteristics of OAM transmission fiber effi-ciently and accurately.2.The optical characteristic prediction network is combined with common parti-cle swarm optimization algorithm and multi-objective particle swarm optimization algorithm,and the fitness function are defined respectively according to different evaluation indexes.The ideal optical fiber structure is designed by iterative optimiza-tion algorithm,and the accuracy of the method is verified by traditional finite element algorithm.The common particle swarm optimization algorithm combined with the neural network prediction models provides a simple and effective way for the reverse design and optimization of optical structures;while the multi-objective particle swarm algorithm offers a more comprehensive evaluation for optical structures and provides more flexible alternative options for reverse design by setting multiple evaluation in-dicators.The proposed design method solves the non-uniqueness problem of reverse design and avoids the single solution problem of tandem neural network,providing an efficiently,accurately and flexibly approach for the inverse design of optical structure with extreme performance.Moreover,this method also has potential application in other scientific fields related to optimization and inverse design.3.Based on the above-mentioned inverse design method,using dispersion and mode effective refractive index difference as evaluation indicators,an optical fiber that can support 22 different OAM modes transmission is designed.The effective re-fractive index difference of all transmission modes is greater than 2.08×10^-4 at 1550nm,which can ensure the stable transmission of OAM modes.In addition,the optical fiber has ultra-low dispersion,and the dispersion of all modes is distributed in the range from-10.98 ps/nm/km to 12.98 ps/nm/km,which further verifies the effective-ness of the reverse design method.