基于KNN-XGBOOST堆叠模型在PCB RFID天线阻抗预测的研究OA北大核心CSTPCD

In view of the time-consuming antenna impedance required in the traditional processes of antenna simulation and modeling,an antenna impedance prediction method on the basis of KNN-XGBOOST-based model is proposed.Most of the existing studies focus on the single prediction algorithms.These studies aim to find out the algorithms with better prediction effect by comparative analysis.Initially,a substantial data of PCB(printed circuit board)RFID(radio frequency identification)antenna impedance design is collected with simulation software ANSYS.Subsequently,leveraging eight influential features including antenna length and frequency,a stacked ensemble learning model is constructed on the basis of constructing the base model with algorithms KNN(K-nearest neighbor)and XGBOOST(eXtreme Gradient Boosting),and the meta-model with linear regression.In the experiment,fine-tuning of the model´s hyperparameters is implemented via cross-validation and grid search techniques,so as to ensure that the model can reach the optimal predictive performance.The experimental results demonstrate that the root mean square error(RMSE)of the KNN-XGBOOST-based model is reduced by 30%～70%,and its R-squared(R2)is increased by 10%in comparison with those of the KNN-based model and XGBOOST-based model.When predicting the real and imaginary parts of PCB RFID antenna impedance,the KNN-XGBOOST-based model exhibits higher accuracy rate and lower prediction error,which verifies its application value in the optimization of electromagnetic simulation design.