为了准确、快速地检测和预测甲烷气体的浓度, 设计了基于红外差分吸收法的甲烷浓度检测系统.为了降低系统部件不稳定带来的影响, 检测系统采用双气室结构, 气室的输入和输出接口处通过渐变折射率透镜连接到传输光纤, 以降低光强的损耗.系统对甲烷检测结果的平均误差为0.007 5.基于粒子群优化的误差反向传播神经网络算法构建了甲烷预测模型, 以浓度在0.2%~2.0%范围内的甲烷气体为研究对象.在样本训练过程中, 预测模型的精度达到10－4, 实际输出值与期望值线性回归的相关系数为0.998 8, 最大相对标准偏差为0.248%.实验结果表明, 在甲烷浓度预测中, 相对于误差反向传播神经网络预测模型, 粒子群优化误差反向传播神经网络的预测性能更优.

In order to accurately and quickly detect and predict the concentration of methane gas, a methane concentration detection system based on infrared differential absorption method was designed. The detection system adopted a double-chamber structure to reduce the influence of system component instability, and the input and output interfaces of the gas chamber were connected to the transmission fiber through a graded-index lens to reduce the loss of light intensity. The average error of the detection system is 0.007 5. An error back propagation neural network algorithm based on particle swarm optimization was used to construct a prediction model with methane gas in the range of 0.2%~2.0%. In the process of sample training, the accuracy of the prediction model reaches 10－4, the correlation coefficient between the actual output value and the expected linear regression is 0.998 8, and the maximum relative standard deviation is 0.248%. The experimental results show that the prediction performance of particle swarm optimization error back propagation neural network is better than that of error back propagation neural network prediction model in methane concentration prediction.