To improve the spectral matching accuracy of the star simulator light source system, firstly, a star simulator light source system based on digital micromirror is designed and built. Secondly, the spectral fitting of the genetic algorithm is performed according to the regional spectrum calibrated by wavelength, The results show that the scheme exits a certain matching error between the fitted spectrum and the target spectrum. Finally, in order to improve the accuracy of spectral matching, an error feedback and accuracy improvement method is proposed to divide the region into two-dimensional wavelength and energy. The experiment simulates light sources with color temperatures of 2550 K, 4766 K, 6576 K, and 8910 K. The results show that, compared with the feedback method of one-dimensional division in the wavelength direction, the maximum error of spectral matching decreases by 55.7%, 50.6%, 45.2%, and 42.2%, respectively, which significantly improves the spectral matching accuracy of the star simulator light source system. The study aims to compensate for the angle measurement error caused by the spectral match error, which improves the star sensor's calibration accuracy.