机动车尾气CO检测中神经网络多环境因子在线修正算法研究

摘要

The influence of temperature,humidity and pressure on the measurement of exhaust gas CO concentration after pretreatment is analyzed.An on-line correction algorithm with multi-environment factors of neural network for the vehicle exhaust CO detection has been proposed.First,the exhaust gas sample data has been trained offline to build the BP neural network model,and then the real-time measured temperature,humidity,pressure and decimal absorption value of the samples have been put into the model for its online correction.Then the corrected CO concentration has been achieved,so the measurement error of the NDIR sensor caused by environmental changes has been solved.Through the prototype experiment,the simulation experiment and the comparison with SEMTECH-EcoStar,the maximum relative deviation of the CO with the concentration from 0 to 0.2％ is 4.8％when the temperature range is from 30 to 50℃,relative humidity is from 25 to 40％,the pressure is from 95 to 115 kPa.The experiments have been carried out in the vehicle field to get the correction factor between 0.8 and 1,which verifies the necessity and reliability of the method and provided effective technical support for the detection of the CO concentration of the high-temperature exhaust gas from motor vehicles.%分析了温度、湿度、压力对预处理后尾气CO浓度测量的影响,提出一种机动车尾气CO检测神经网络多环境因子在线修正算法,首先采用尾气样本数据离线训练得到BP神经网络模型,然后将实时测得的样品气温度、湿度、压力及小数吸收值代入到模型进行在线修正,得到修正后CO浓度,解决了NDIR传感器因环境变化所带来的测量误差影响.通过标样实验、模拟实验,并和SEMTECH-EcoStar对比检测结果,在样品气温度30～50℃、相对湿度25～40％、压力95～115 kPa、CO浓度0～0.2％范围内的最大相对偏差为4.8％.车载外场实验,得到修正因子在0.8 ～1之间,验证了方法的必要性和可靠性,为机动车尾气的CO浓度的准确检测提供有效技术支持.