Aiming at the problem of low measurement accuracy of non dispersive infrared CO2 sensor affected by temperature,a non dispersive infrared CO2 sensor with single optical path and two channels is designed,and a new method of BP neural network compensation for gas chamber temperature is proposed.In this paper,the influence of temperature on the light source,detector and gas absorption coefficient of the sensor is analyzed,and the BP neural network compensation method for gas chamber temperature is compared experimentally with other compensation methods.The experimental results show that the BP neural network compensation method for gas chamber temperature is superior to other compensation methods.When measuring CO2 in the concentration range of 10% 20%,the maximum relative error is 2.98%,and the repeatability experiment shows an RSD of 1.22%,which provides a reference for the compensation of temperature by non dispersive infrared CO2 sensor.