摘要
针对主元统计方法、基于人工神经网络理论模式识别方法受气体传感阵列交叉敏感特性影响,无法预处理敏感数据,导致识别精度较低的问题,提出基于梯度提升决策树的气体传感阵列识别方法。通过分析气态成分,借助scale函数构造归一化模型,标准化预处理敏感数据,缩小敏感数据与其他数据的差异。采用梯度提升决策树建立高精度的识别模型,修正负梯度误差,并结合SVM-predict识别程序识别六组气体。实验结果表明,该方法在识别气体时,最大误差为0.7μL/L,具有识别精准度高的优势。
The principal component statistics method and pattern recognition method based on artificial neural network theory are affected by the cross sensitivity of gas sensor array and can not preprocess sensitive data,resulting in low recognition accuracy.A gas sensor array recognition method based on gradient lifting decision tree is proposed.By analyzing the gaseous components,a normalized model is constructed with the help of scale function to standardize the preprocessing of sensitive data and reduce the difference between sensitive data and other data.The gradient lifting decision tree technology is used to establish a high⁃precision identification model,correct the negative gradient error,and identify six groups of gases combined with SVM-predict identification program.The experimental results show that the maximum error of this method is 0.7μL/L,with the advantage of high recognition accuracy.
作者
张志业
葛志强
赵小娟
林永江
ZHANG Zhiye;GE Zhiqiang;ZHAO Xiaojuan;LIN Yongjiang(Guoneng(Quanzhou)Thermal Power Co.,Ltd.,Quanzhou 362804,China)
出处
《电子设计工程》
2022年第18期142-145,150,共5页
Electronic Design Engineering
关键词
梯度提升决策树
气体
传感阵列
识别
敏感数据
归一化处理
gradient lifting decision tree
gas
sensor array
identification
sensitive data
normalization treatment