Abstract: In this paper, the mathematical model of a neural network compensating method was presented, the maximum and minimum distance method was used to determine the number of clusters, RBF neural network center and expansion constants were calculated by using the clustering algorithm. The experimental results showed that the compensating model design was reasonable and achieved the desired requirements, which provided theoretical basis for detecting low carbon monoxide gas concentration accurately.