Quantitative analysis of channel gas is of great significant for the safety of the maintenance staff. This paper proposes a spectral quantitative analysis method based on adaptive boosting kernel ELM for gas component prediction of underground cable channel. The proposed method uses the ensemble learning strategy based on some base models. Each base model are built by the conjunction process, namely, each base model are trained with the weights of training samples which are adjusted based on the performance of the last base model. Then, the outputs of every base models are combined into a weighted sum to obtain the final output of the proposed method. Moreover, the kernel ELM is adopted as the base model, where the kernel function is used to replace the random matrix in ELM for dealing with the nonlinearity of spectral data. A real gas spectral dataset including the methane, the carbon monoxide and the carbon dioxide is used in the experiments.The experiments results verify that the proposed model has higher effectiveness.