基于支持向量回归的光声光谱信号温度和湿度校正方法

Temperature and humidity correction method for photoacoustic spectroscopy signal based on support vector regression

非共振光声光谱技术可用于在线检测变压器油中溶解气体,保障变压器安全运行。然而在室外环境下,背景信号易受到环境温度、湿度的影响,必须对这些因素进行校正,以提高仪器在线检测的稳定性和可靠性。背景信号的温度和湿度校正因子一般通过实验室测量得到,但测量过程中由于外部环境干扰,采集样本会出现较大波动,因此需要具有一定鲁棒性的回归算法计算校正因子。研究了基于支持向量回归的背景信号的温度和湿度校正方法,选择乙炔作为研究对象,在实验室内利用湿度发生器产生不同浓度水汽,同时利用温度传感器测量光声池温度,回归乙炔背景信号校正因子,并采用体积分数分别为0、5×10^(−6)和2×10^(−5)的乙炔和空气混合气体进行了验证。研究结果表明,对于体积分数为5×10^(−6)和2×10^(−5)的乙炔混合气体,所提方法和最小二乘法校正结果趋势相同,但最小二乘法校正信号存在趋势性偏离,而所提方法对背景信号校正具有更好的重复性和稳定性,优于最小二乘法。

Photoacoustic spectroscopy(PAS)technology is very useful in online detection of dissolved gases in transformer oil to ensure the safety of transformers.However,in the outdoor environment,the measured background signal of the instrument is easily affected by the ambient temperature and humidity,and the correction of these factors must be considered to improve the stability and reliability of the online measurement of PAS instrument.The background correction factor is generally obtained by regression of different humidity and temperature measured in the laboratory,but the environment often changes during measurement procedure,so a robust regression algorithm is needed to obtain accurate correction factor.The temperature and humidity correction method based on the support vector regression is studied in this work.Acetylene is selected as the research object.Humidity air with different concentrations is generated by using humidity generator in laboratory,and the temperature of the PAS cell is measured by temperature sensor,and then the background signal correction factor of acetylene is regressed from the measured data.Furthermore,the method is verified by acetylene air mixture with acetylene volume fraction of 0,5×10^(−6) and 2×10^(−5).The results show that the proposed method and the least squares regression method have the same trend for acetylene mixture with volume fraction of 5×10^(−6) and 2×10^(−5).Whereas the least squares regression has tendency in residuals,and the proposed method has better repeatability and stability for temperature and humidity correction than the least square method.