基于多任务学习的光信道性能多变量预测方法  

A multivariate prediction method for optical channel performance based on multi-task learning

作　　者：孙晓川 曹荻非 郝明祥 李志刚 李莹琦 SUN Xiaochuan;CAO Difei;HAO Mingxiang;LI Zhigang;LI Yingqi(College of Artificial Intelligence,North China University of Science and Technology,Tangshan,063210,P.R.China;Hebei Key Laboratory of Industrial Intelligent Perception,Tangshan,063210,P.R.China)

机构地区：[1]华北理工大学人工智能学院,河北唐山063210 [2]河北省工业智能感知重点实验室,河北唐山063210

出　　处：《重庆邮电大学学报（自然科学版）》2023年第5期854-862,共9页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：河北省高等学校科学技术研究项目(ZD2021088)。

摘　　要：针对高速光骨干网络中信道传输损伤对信道性能造成影响的问题,提出基于多任务学习的光信道性能多变量预测方法。构建基于交叉递归图理论的光信道性能影响因素的确定方法,从定性和定量的角度实现对光信道性能的同信道以及跨信道影响因素的有效筛选;利用多个筛选的影响因素,构建基于多任务学习的深度回声状态网络多变量预测方法,实现不同光信道状态参数的特征共享,进而高效准确地完成特定光信道的性能预测任务。仿真结果表明,提出的方法在保证运算效率的前提下,相较于其他先进深度学习模型,预测精度平均可提升约39.8%。In view of the impact of channel transmission damage on channel performance in high-speed optical backbone networks,a multi-variable prediction method for optical channel performance based on multi-task learning is proposed.A method for determining the factors affecting optical channel performance based on cross-recursive graph theory is constructed,which effectively screens the same-channel and cross-channel influencing factors of optical channel performance from both qualitative and quantitative perspectives.Using multiple screened influencing factors,a multi-variable prediction method based on deep echo state network for multi-task learning is constructed,which achieves feature sharing of different optical channel state parameters,thereby efficiently and accurately completing the performance prediction task of specific optical channels.Simulation results show that the proposed method can improve the prediction accuracy by about 39.8%on average compared to other state-of-the-art deep learning models,while ensuring operational efficiency.

关 键 词：光纤和光通信 交叉递归图理论 多任务学习 深度回声状态网络 

分 类 号：TN913.7[电子电信—通信与信息系统]