The infrared spectrum characteristic of high temperature gas is an effective way to judge the composition and concentration of gas. Aiming at the problems of complex infrared radiation characteristics and high modeling difficulty of high-temperature gas, a feature extraction algorithm based on interval partial least squares (iPLS) and kernel principal component analysis (KPCA) is studied. Firstly, the characteristic spectral bands with the best prediction ability are determined by pre-screening with iPLS to avoid the loss of useful absorption peak information in the process of single subinterval modeling. Secondly, KPCA is used to reduce the data dimension, retain the key features with high contribution rate, and reduce the complexity of the component prediction model. The simulation results show that after feature extraction by iPLS-KPCA method, the complexity of the prediction model is greatly reduced, and the prediction ability is significantly improved.