In this paper, we proposed an adaptive multi-band joint concentration inversion algorithm, which combines the transmittance stable range and the spectral width threshold to adaptively select the effective band of the measured gas. The nonlinear least squares fitting method is used to invert the concentration of each effective band and analyze the residual to obtain the concentration inversion results and their weights of each effective band. The accurate quantitative analysis of the concentration of the measured gas is realized by weighted averaging. The algorithm verification experiment is carried out. The results show that the stability coefficient of the adaptive multi-band joint concentration inversion algorithm is 0.9976. Compared with the traditional single-band and multi-band concentration inversion algorithms, the root mean square error of the inversion results is reduced by 64.44% and 41.52%, the mean relative error is reduced by 65.97% and 46.72%, and the mean absolute error is reduced by 66.32% and 47.74% respectively. It can be concluded that the inversion accuracy and stability are significantly improved.