现代应用物理2025，Vol.16Issue(1)：151-157,7.DOI:10.12061/j.issn.2095-6223.202412005
面向高功率微波反演的高效时序神经网络算法研究
A High-Efficiency Time Series Model for High-Power Microwave Inversion Base on Neural Networks
摘要
Abstract
The columnar plasma array(CPA)is an excellent high-power microwave(HPM)protection method,but not any HPM is sufficient to excite the electromagnetic shielding effect of CPA.In this paper,a new time series model named improved iTransformer(iiTransformer)based on Encoder-Decoder framework and multivariate attention mechanism is proposed.This model aims to mathematically model the complex nonlinear processes between HPM and CPA,and achieve high-power microwave inversion.The algorithm simulation and data acquisition of complex processes are completed by the finite element method(FEM),and HPM inversion is achieved using the iiTransformer model and ResNet-18 model,respectively.In the iiTransformer,Encoder and Decoder respectively perform multi-head self attention processing on the data and target sequence to extract the relationship dependencies of multiple variables in multiple channels.The research results show that the designed iiTransformer has strong representation learning and nonlinear fitting abilities,not only strong generalization ability but also good robustness.For iiTransformer,the loss on the training set is 1.637 × 10-7,the loss on the validation set reaches 1.51×10-8,and the accuracy on the test set is 99.978％,far higher than the accuracy of the ResNet-18 model.关键词
高功率微波防护/柱状等离子体阵列/时间序列模型/Encoder-Decoder框架/multivariate attention机制Key words
high-power microwave protection/columnar plasma array/time series model/encoder-decoder framework/multivariate attention mechanism分类
电子信息工程引用本文复制引用
董纯志,黄志祥,冯乃星..面向高功率微波反演的高效时序神经网络算法研究[J].现代应用物理,2025,16(1):151-157,7.基金项目
国家自然科学基金资助项目(62271001) （62271001）
安徽省自然科学基金资助项目(2308085Y39,2022AH030014) （2308085Y39,2022AH030014）
