Fast and accurate calculation of atmospheric transmittance is essential for infrared atmospheric remote sensing and satellite data assimilation. However, fast radiative transfer models show significant errors in strong water vapour absorption channels (e.g., near 6.7 μm). An important reason is that the numerical instability encountered during the regression of transmittance coefficients when dealing with lower and middle atmosphere. To address this, this study proposes an optimized scheme to calculate atmospheric transmittance vertical profiles for the RTTOV (Radiative Transfer for TOVS) fast transmittance algorithm. The method introduces a physically motivated transmittance threshold to sub-select training samples and employs cumulative transmittance-based weighting factor within a weighted least squares regression to recalibrate the transmittance coefficients. It aims to optimize the calculation scheme for transmittance coefficients of the Hyperspectral Infrared Atmospheric Sounder-II (HIRAS-II) instrument onboard Fengyun 3F satellite (FY-3F). The method is assessed by calculations on the training profile datasets provided within the RTTOV model framework. By comparing transmittance and brightness temperature calculations at 6.7 microns from this method with those from a line-by-line model and observations from HIRAS-II, the results show that the accuracy of the forward model for the 6.7 μm absorption channel is significantly enhanced by applying a threshold-based noise reduction method. This improvement enhances the stability and reliability of the transmittance calculations for this strong absorption band. Further accuracy enhancements are obtained by incorporating weighting corrections into the calculations of transmittance coefficients. The root mean square error (RMSE) and bias of the observation minus background (OMB) time series for FY-3F HIRAS-II demonstrate that the transmittance coefficient calculation scheme with weighting factor correction improves the forward model accuracy, which is more consistent with RTTOV simulation results. The OMB bias at the 6.7 μm absorption peak channel performs better than that of RTTOV, while the OMB bias on both sides of the 6.7 μm absorption peak channel remains consistent with RTTOV.