Abstract: Based on the gas system of gas-steam combined cycle power plant from a steelworks, a mechanistic model was established for the multi-stage gas compression system, which essentially consisted of scrubbers, centrifugal compressors and coolers. Adaptive genetic algorithm was applied to estimating the important parameters of the mechanistic model, which cannot be confirmed using the mechanistic model. Since there are many factors having effect on the performance of the multi-stage gas compression system, the mechanistic model may yield inaccurate results. Thus GRBF neural network was employed to correct the error of the mechanistic model. The hybrid model was established by connecting the mechanistic model and GRBF neural network in parallel. Models were applied to the practical gas system, and the results demonstrated that compared with the mechanistic model, the hybrid model has higher accuracy.