光谱分析法是煤矿气体定量分析的重要技术手段。然而，当前广泛采用的便携式红外气体分析仪在现场测试过程中受环境温度影响较大，为提高气体分析仪的分析精度和适用性，围绕红外气体分析仪恒温加热的必要性、合理加热温度范围、测试数据误差优化模型等问题，开展了理论分析和实验研究。首先，搭建了涵盖温度补偿前、恒温加热测试、温度补偿后以及不同测试环境温度条件的红外气体分析仪实验平台，提出了实验步骤以及相应的分析方法；其次，在环境温度为25～45 ℃时，测试的CO气体体积分数为800×10⁻⁶，通过多次测试发现加热温度为45 ℃时，CO气体分析仪精度最佳，测试值最接近气体体积分数真实值；同时，选用CO（体积分数为10×10⁻⁶~500×10⁻⁶）标准气体进行了精度测试，计算CO气体绝对误差仅为0~9×10⁻⁶，优于标准规定的误差范围；最后，设置−15~45 ℃的变温环境，对加热后的CO气体分析仪进行了温度补偿验证，测试得出环境温度越低，待测气体体积分数越高，则分析仪检测误差越大，最大误差可达110×10⁻⁶。基于此，建立了BP、GA-BP神经网络误差分析模型，将环境温度、测试体积分数和标准气体体积分数与优化模型进行融合。对比发现：采用GA-BP神经网络模型得到的MAE值相较于BP神经网络降低了12.9823×10⁻⁶，且MSE、RMSE计算结果也验证了GA-BP模型在气体分析中的稳定性和可行性。

Spectral analysis is an important technical means for the quantitative analysis of coal mine gas. However, the portable infrared gas sensor widely used at present is greatly affected by the ambient temperature in the field test process. In order to improve the analysis accuracy and applicability of the gas analyzer, we carry out theoretical analysis and experimental research around the necessity of constant temperature heating of infrared gas analyzers, reasonable heating temperature range, and optimization model of test data error. Firstly, an experimental platform of infrared gas analyzer covering pre-temperature compensation, constant temperature heating test, post-temperature compensation and different test environment temperature conditions is constructed, and the experimental steps as well as the corresponding analytical methods are proposed; secondly, in the ambient temperature of 25-45 ℃ interval, the tested CO gas volume fraction is 800×10⁻⁶, through a lot of tests, it is found that the accuracy of CO gas analyzer is best when the heating temperature is 45 ℃, the test value is the closest to the true value of gas volume fraction. At the same time, CO (10×10⁻⁶-500×10⁻⁶) standard gas is selected for the accuracy test, and the absolute error of the calculated CO gas is only 0-9×10⁻⁶, which is better than the error range specified in the standard; then, setting the variable temperature environment of −15- 45 ℃, the temperature compensation verification of the heated CO gas sensor is carried out, the test results show that the lower the ambient temperature, the higher the volume fraction of the gas to be measured, the greater the sensor detection error, the maximum error can reach 110×10⁻⁶; based on this, the error analysis models of BP and GA-BP neural network are established, and the ambient temperature, test volume fraction and standard gas volume fraction are fused with the optimized model. It is found that MAE value obtained by GA-BP neural network model is reduced by 12.982 3×10⁻⁶ compared with BP neural network. The MSE and RMSE calculation results also verify the stability and feasibility of GA-BP model in gas analysis.