改进粒子群优化算法和双分支网络的耦合效率预测  

Coupling efficiency prediction of improved particle swarm optimization algorithm and double branch network

作　　者：赖春红 陈先勤 姜小明[2] 陈帅 王悦 刘思创 邹林熹 LAI Chunhong;CHEN Xianqin;JIANG Xiaoming;CHEN Shuai;WANG Yue;LIU Sichuang;ZOU Linxi(School of Optoelectronic Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China;School of Life Health Information Science and Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)

机构地区：[1]重庆邮电大学光电工程学院,重庆400065 [2]重庆邮电大学生命健康信息科学与工程学院,重庆400065

出　　处：《重庆邮电大学学报(自然科学版)》2025年第1期76-84,共9页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家自然科学基金项目(62105050)。

摘　　要：针对混合波导结构耦合效率分析中软件仿真所需时间长、资源占用大的问题,提出神经网络模型对混合波导结构耦合效率进行预测,减少时间资源的占用。为解决多变量问题中不同自变量间步长差距较大导致单一网络特征提取能力受限的问题,提出针对不同变量使用双分支网络进行特征提取,使用改进后的粒子群优化算法对双分支网络超参数进行全自动优化,进一步提高模型预测精度。在混合波导结构耦合效率数据集上的实验表明,提出的改进粒子群优化-双分支网络的均方误差比支持向量机算法与循环神经网络分别降低了88.17百分点、21.17百分点,比标准粒子群优化算法降低了16.81百分点,验证了模型以及改进算法的有效性和优越性。To address the issues of long simulation time and high resource consumption in software-based analysis of coupling efficiency in hybrid waveguide structures,this paper proposes a neural network model to predict coupling efficiency,thereby reducing time and resource usage.To solve the problem of significant step size differences between independent variables in multivariable problems,which limits the feature extraction capability of a single network,a dual-branch network is introduced to extract features for different variables.An improved particle swarm optimization(PSO)algorithm is applied to automatically optimize the hyperparameters of the dual-branch network,further improving model prediction accuracy.Experiments on the hybrid waveguide structure coupling efficiency dataset show that the proposed improved PSO-dual-branch network reduces the mean squared error by 88.17 percentage points compared to the support vector machine(SVM)algorithm,21.17 percentage points compared to the recurrent neural network(RNN),and 16.81 percentage points compared to the standard PSO algorithm,validating the effectiveness and superiority of the proposed model and optimization algorithm.

关 键 词：粒子群优化算法 双分支神经网络 混合波导结构耦合效率 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]