Terahertz biomedicine is getting more and more concentration, especially in the field of spectroscopic research. Its main difficulty lies in how to achieve accurate component analysis of complex biological system as well as effectively avoid water interference . Terahertz spectrum contains the information of molecular vibration. However, its absorption spectrum is weak and overlaps seriously. Therefore, it is difficult to use traditional calibration techniques for quantitative calculation, such as the peak height and peak area. Adopting multivariate correction method makes terahertz spectrum a fast, simple and widely applicable way to carry out quantitative analysis. In this paper, the mixed aqueous salt solution of KCl and NaCl is taken as a typical system to be studied. The concentration of each component varies from 0.1 to 2 mol·L-1, with an interval of 0.1 mol·L-1. Due to the hydrated hydrogen bond of inorganic metal ions, terahertz time-domain spectrum data can be collected to analyze each component quantitatively. Based on theorthogonal experiment principle, we constructed the data set with outstanding structure characteristics to accuratelyextract the hydrogen bond information by spectral analysis. Here, the adaptive algorithm is developed to find the relationship between the spectral data and the concentration, and the variable screening technology is adopted to extract the characteristic information of different components from the original spectral data. Finally, we build the data-driven model between the concentration and the characteristic information. The calculation results show that the prediction errors of KCl and NaCl components are 8.0% and 9.1% respectively, which can effectively meet the requirements of detection accuracy for most applications. Therefore, the new method of data-driven modeling terahertz spectrum analysis can provide a new way for terahertz biomedical research.