基于WOA-WNN-LSTM算法的红外CO痕量气体压力补偿与时序浓度分析  

Pressure Compensation of Industrial Ambient Gases and Their Prediction Based on Infrared Spectroscopy

作　　者：田富超[1,2,3] 张海龙 苏嘉豪 梁运涛 王琳[1,2,3] 王泽文 TIAN Fu-chao;ZHANG Hai-long;SU Jia-hao;LIANG Yun-tao;WANG Lin;WANG Ze-wen(Graduate School,China Coal Research Institute,Beijing 100013,China;State Key Laboratory of Coal Mine Safety Technology,China Coal Technology and Engineering Group Shenyang Research Institute,Shenfu Demonstration Zone 113122,China;Department of Safety Engineering and Engineering,School of Emergency Management and Safety Engineering,China University of Mining and Technology(Beijing),Beijing 100083,China)

机构地区：[1]煤炭科学研究总院研究生院,北京100013 [2]中煤科工集团沈阳研究院有限公司煤矿灾害防控全国重点实验室,辽宁沈抚示范区113122 [3]中国矿业大学(北京)应急管理与安全工程学院安全工程与工程系,北京100083

出　　处：《光谱学与光谱分析》2025年第4期994-1007,共14页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(52174229,52174230);辽宁省自然科学基金项目(2022-KF-23-03,2023-MS-355)资助。

摘　　要：红外光谱分析是工业环境气体定量分析的重要手段,当前红外气体检测仪的测量精度受环境压力变化影响较大,导致检测数据在不同压力条件下偏离实际气体浓度。为提高红外气体传感器的精度,选择了鲸鱼优化算法(whale optimization algorithm,WOA)和小波神经网络(wavelet neural network,WNN)相结合的压力补偿算法,并结合长短期记忆法(long short-term memory,LSTM)对补偿后的数据进行预测。通过搭建工业环境气体压力补偿实验平台,使用高精度配气仪配置100~900 ppm标准CO气体,在80~120 kPa范围内进行数百组重复实验,发现CO气体传感器在负压条件下测量值小于标气浓度,正压条件下测量值大于标气浓度,并随压力变化呈线性关系,绝对误差最高为86 ppm。将传感器数据使用小波神经网络进行误差降低,初步补偿后的CO误差降至26 ppm,但由于参数可移植性较差,个别数据误差较大。进一步使用鲸鱼优化算法优化小波神经网络的参数后,补偿效果显著提升,传感器测量值与真值之差保持在0.004%以内且数据稳定。最终结合LSTM进行气体浓度预测,预测值与实际值之间的均方根误差(RMSE)均小于0.1,平均绝对误差(MAE)均小于0.020,实验结果表明,WOA-WNN-LSTM算法能够有效提高红外气体传感器的测量精度,成功消除环境压力对测量结果的影响,为工业环境气体检测提供了更为可靠和精准的解决方案。Infrared spectroscopy is one of the important means of quantitative analysis of industrial environmental gases.Still,the current infrared gas detector's measurement accuracy is greatly affected by ambient pressure changes,resulting in the detection data deviating from the actual gas concentration under different pressure conditions.To improve the accuracy of the infrared gas sensor,this paper chooses a pressure compensation algorithm combining the Whale Optimization Algorithm(WOA)and Wavelet Neural Network(WNN).It combines it with Long Short-Term Memory(LSTM).Memory(LSTM)to predict the compensated data.By building an experimental platform for gas pressure compensation in industrial environments,using a high-precision gas dispenser to configure 100~900 ppm standard CO gas,and conducting hundreds of repetitive experiments in the range of 80~120 kPa,it is found that the measured value of the CO gas sensor is less than the concentration of the standard gas under negative pressure conditions,and more than the concentration of the standard gas under positive pressure conditions,and the absolute error is linearly correlated with the pressure change,with the highest absolute error of 0.5 ppm.A linear relationship was found,with an absolute error of up to 86 ppm.The sensor data was used to reduce the error using a wavelet neural network,and the initial compensated CO error was reduced to 26 ppm.Still,the individual data error was large due to poor parameter portability.After further optimizing the parameters of the wavelet neural network using the whale optimization algorithm,the compensation effect was significantly improved.The difference between the sensor measurement and the true value was kept within 0.004%,and the data were stable.The root mean square error(RMSE)between the predicted and actual values is less than 0.1,and the mean absolute error(MAE)is less than 0.020.The experimental results show that the WOA-WNN-LSTM algorithm can effectively improve the measurement accuracy of the infrared gas sensors and success

关 键 词：红外光谱分析 环境压力补偿 鲸鱼优化算法 小波神经网络 时序浓度预测 

分 类 号：O657.3[理学—分析化学]