摘要
为了克服深度学习训练过度耗时的弊端,提出宽度学习系统(BLS),使用岭回归算法求解输出权重,极大地方便了训练过程,对于大样本而言,此做法高效且快速.然而,在处理微波器件这类小样本电磁问题上,则大大限制了模型的拟合能力.为了进一步提升模型的性能,针对小样本的回归问题,将迭代重加权最小二乘算法(IRLS)与宽度学习算法相结合,配合网格搜索法以确定模型的最佳结构.通过对矩形和圆形两种微带天线谐振频率的预测,并与一些主流的算法对比,证实了迭代重加权最小二乘宽度学习系统(IRLS-BLS)的有效性.
In order to overcome the extremely time-consuming drawback of deep learning(DL),a broad learning system(BLS)was proposed as an alternative method.For this model,ridge regression algorithm is used to solve the output weights by default,which greatly facilitates the training process and are really efficient and fast for large samples.However,when dealing with small sample problems such as microwave devices,this method greatly limits the fitting ability of the model.In order to further improve the performance of the model,this paper combines the iterative reweighted least squares(IRLS)algorithm with the broad learning algorithm to deal with small sample regression problems,and uses the grid search method to determine the optimal structure of the model.By predicting the resonant frequencies of rectangular and circular microstrip antennas and being compared with some mainstream algorithms,the effectiveness of the iterative reweighted least squares broad learning system(IRLS-BLS)is verified.
作者
丁伟桐
李鹏飞
袁慧宁
田雨波
DING Weitong;LI Pengfei;YUAN Huining;TIAN Yubo(Ocean College,Jiangsu University of Science and Technology,Zhenjiang 212100,China)
出处
《江苏科技大学学报（自然科学版）》
CAS
北大核心
2022年第5期66-71,共6页
Journal of Jiangsu University of Science and Technology:Natural Science Edition
基金
国家自然科学基金资助项目(61771225)。
关键词
宽度学习系统
岭回归算法
迭代重加权最小二乘算法
网格搜索法
broad learning system
ridge regression algorithm
iterative reweighted least squares algorithm
grid search method
