基于IALO-HBP神经网络的超宽带滤波器逆向建模方法  

Inverse Modeling Approach for Ultra-Wideband Filters Based on IALO-HBP Neural Networks

作　　者：南敬昌[1] 杜晶晶 高明明[1,2] 谢欢 Nan Jingchang;Du Jingjing;Gao Mingming;Xie huan(School of Electronics and Information Engineering,Liaoning Technical University,Huludao 125105,Liaoning,China;Information Science and Technology College,Dalian Maritime University,Dalian 116026,Liaoning,China)

机构地区：[1]辽宁工程技术大学电子与信息工程学院,辽宁葫芦岛125105 [2]大连海事大学信息科学技术学院,辽宁大连116026

出　　处：《激光与光电子学进展》2022年第12期430-438,共9页Laser & Optoelectronics Progress

基　　金：国家自然科学基金(61971210)。

摘　　要：针对双陷波超宽带滤波器使用后向传输(BP)神经网络逆向建模存在精度较低、收敛慢、稳定性不强等问题,提出一种用改进的蚁狮算法(IALO)结合Huber函数优化BP神经网络逆向建模的方法。该方法通过将边界收缩因子连续化,引入动态更新系数以及加入柯西变异来实现对蚁狮算法的改进,并用改进的蚁狮算法优化正向模型的权值,加快建模速度,然后使用Huber函数作为神经网络的评价函数,提高了模型的精度和稳定度。将此方法用于双陷波超宽带滤波器中,实验结果表明,对比BP逆向建模方法,此方法求得的长度、宽度和频率均方误差分别减小了97.44%、99.43%和96.15%,平均运行时间缩短了66.01%,解决了逆向建模的多解问题,提高了设计滤波器的速度和精度。To address the problems of low accuracy, slow convergence, and poor stability in using the backpropagation(BP) neural network for inverse modeling of dual band-notched ultra-wideband filters, this paper proposes an approach to optimizing inverse modeling based on the BP neural network with an improved ant lion optimization(IALO) algorithm and the Huber function. This method improves the ant lion optimization algorithm by serializing the boundary contraction factor, introducing dynamic update coefficients, and adding the Cauchy mutation. Then, the IALO algorithm is applied to optimize the weights of the forward model and thereby speed up the modeling.Subsequently, the Huber function is used to evaluate the neural network. The accuracy and stability of the model are thus improved. This method is used for a double band-notched ultra-wideband filter. Experimental results show that compared with BP inverse modeling, the proposed method reduces the length, width, and frequency mean square errors by 97. 44%, 99. 43%, and 96. 15%, respectively, and shortens the average running time by 66. 01%. The multi-solution problem of inverse modeling is solved, and the speed and accuracy of filter design are improved.

关 键 词：神经网络逆向建模 双陷波超宽带滤波器 改进的蚁狮算法 Huber函数 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]