In order to effectively monitor and control the concentration of SF6 gas in power system,a SF6 gas sensor is designed based on the principle of non-dispersive infrared(NDIR).However,in the actual detection process,the difference between the ambient temperature and pressure has a great impact on the detection accuracy of the SF6 gas sensor,so it is necessary to adopt appropriate methods to eliminate the measurement error caused by the environment.In this paper,GWO-BP neural network was used for real-time compensation for the measuring deviations caused by changing temperature and pressure,and then compared with other compensation approaches.Analysis shows that the measurement accuracy of concentration data after compensation is less than ±15 ppm in a range of 0 to 2000 ppm,with the full range accuracy up to 0.75 %.It effectively improves the sensor measurement accuracy and reduces the volume and cost of the sensor meanwhile.