基于RadTransformer神经网络的辐射场强预测  

The Prediction of Radiation Field Intensity Based on RadTransformer Neural Network

作　　者：张戟[1] 陈颖[1] ZHANG Ji;CHEN Ying(Clean Energy Automotive Engineering Center,Tongji University,Shanghai 201804,China)

机构地区：[1]同济大学新能源汽车工程中心,上海201804

出　　处：《佳木斯大学学报（自然科学版）》2024年第9期1-5,65,共6页Journal of Jiamusi University：Natural Science Edition

摘　　要：研究板级辐射干扰时,由于PCB内部信号频率极高、走线结构复杂,难以精准分析干扰传播路径和提取路径上的寄生参数,导致建模时考虑因素不全,仿真的辐射场强值不够准确。因此,利用多头自注意力神经网络自研RadTransformer模型,结合CST软件,对车载指示灯系统的辐射场强进行快速准确的预测。首先依据国家标准GB/T 18655-2018进行辐射发射试验,同时搭建对标实际测试环境的仿真模型,获取预测数据集。选取与辐射场强相关的三个重要参数:共模电流、传递函数、天线系数,利用RadTransformer模型提取它们对辐射场强的注意力特征值,预测指定频率下的辐射场强值。研究结果表明,该模型预测的方法兼具准确性与快速性,预测时间仅用时0.1 ms,平均误差值仅为2.0,相比于基准模型平均误差值减少了30%。When studying board level radiation interference,the accurate analysis of interference propagation path and extraction of parasitic parameters on the path are challenging due to the extremely high signal frequency and complex routing structure of PCB.As a result,factors in modeling are incompletely considered and the simulated radiation field intensity lacks accuracy.To address this issue,we utilize a self-developed RadTransformer model with multi-head self-attention neural network combined with CST software to swiftly and accurately predict the radiation field intensity of vehicle indicator system.Firstly,we conduct radiation emission tests according to national standard GB/T 18655-2018 and establish a simulation model that replicates the actual test environment for benchmarking purposes.Subsequently,three crucial parameters related to radiation field intensity-common mode current,transfer function,and antenna coefficient-are selected.By employing the RadTransformer model,their characteristic values regarding attention towards radiation field intensity are extracted for predicting it at specified frequencies.The research findings demonstrate that the prediction method exhibits both accuracy and efficiency as it only takes O.lms for predictions while maintaining an average error value of merely 2.0-a reduction by 3o%compared to that of the benchmark model.

关 键 词：汽车电子系统 电磁兼容 辐射干扰 神经网络 RadTransformer 

分 类 号：U463.65[机械工程—车辆工程]