In　the　electromagnetic　(EM)　optimization　algorithm　experiment,　it　isn＇t　easy　to　optimize　the　sample　points　far　away　from　the　optimization　target.　In　order　to　solve　this　problem,　this　paper　combines　the　neuro-transfer　function　(neuro-TF)　and　sensitivity　information　of　the　sample.　This　paper　simulates　the　output　of　microwave　devices　at　high　frequencies　by　building　surrogate　models.　Sensitivity　information　can　improve　the　accuracy　and　robustness　of　surrogate　models.　This　method　enables　the　feature　model　to　ensure　a　certain　accuracy　under　poor　conditions.　©　2023　Applied　Computational　Electromagnetics　Society　(ACES).