中红外光谱技术的乳化溢油检测方法研究  

Research on Emulsified Oil Spill Detection Methods Based on Mid-Infrared Spectroscopy Technology

作　　者：李心怡 孔德明[1] 宁晓东 崔耀耀 LI Xin-yi;KONG De-ming;NING Xiao-dong;CUI Yao-yao(School of Electrical Engineering,Yanshan University,Qinhuangdao 066004,China;School of Information Science and Engineering,Yanshan University,Qinhuangdao 066004,China;School of Mechanical and Electrical Engineering,Shijiazhuang University,Shijiazhuang 050035,China)

机构地区：[1]燕山大学电气工程学院,河北秦皇岛066004 [2]燕山大学信息科学与工程学院,河北秦皇岛066004 [3]石家庄学院机电学院,河北石家庄050035

出　　处：《光谱学与光谱分析》2025年第3期631-636,共6页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(62173289)资助。

摘　　要：快速、准确获取乳化溢油的种类和含油量等信息对于海上溢油污染的监测具有重要意义。中红外光谱技术是一种简单、高效的检测手段,可通过光谱特征峰的位置和强度来表征物质分子的结构信息。然而,目前将中红外光谱技术应用于乳化溢油的检测还尚未有成熟的研究成果。基于此,选取92#、95#、98#三种具有代表性的汽油样本,建立乳化汽油体系;选取5#、7#、10#三种具有代表性的白油样本,建立乳化白油体系。利用中红外光谱技术获得乳化溢油样本的光谱数据并进行预处理,然后选用线性判别分析(LDA)算法实现乳化溢油的油种鉴别。在此基础上利用竞争性自适应重加权采样法(CARS)和随机森林(RF)分别选择出与含油率呈线性和非线性关系的特征波长,既降低了数据维度又丰富了特征数据的多样性。然后选用极端梯度提升(XGBoost)、一维卷积神经网络(1D-CNN)、支持向量回归(SVR)作为基学习器,偏最小二乘回归(PLSR)作为元学习器,构建两层的Stacking集成学习模型来预测乳化溢油中的含油率。Stacking集成学习模型获得的乳化汽油和乳化白油的测试集决定系数分别为0.9824和0.9873,均方根误差分别为0.0410和0.0340。与XGBoost、1D-CNN、SVR、PLSR相比,Stacking集成学习模型具有更好的稳定性和准确性。上述研究结果表明,基于中红外光谱技术结合LDA和Stacking集成学习的检测方法,能有效实现乳化溢油的定性与定量分析,从而为乳化溢油领域的研究提供了新思路。Quick and accurate acquisition of information,such as emulsified oil spills types and oil content,is of great significance for monitoring offshore oil spill pollution.Mid-infrared spectroscopy is a simple and efficient detection method that can characterize the structure information of substance molecules by the position and intensity of spectral characteristic peaks.However,applying infrared spectroscopy technology to detect emulsified oil spills has not yet yielded mature research results.Based on this,this paper selected three representative gasoline samples,92#,95#,and 98#,to establish an emulsified gasoline system and three representative white oil samples,5#,7#,and 10#,to establish an emulsified white oil system.The spectral data of emulsified oil spill samples were obtained by mid-infrared spectroscopy,and pretreatment was carried out.Then,a Linear Discriminant Analysis(LDA)algorithm was used to identify oil species from emulsified oil spills.Based on this,the Competitive Adaptive Reweighted Sampling(CARS)and Random Forest(RF)methods were used to select the feature wavelengths with linear and non-linear relationships with oil content,respectively.This reduces data dimensionality and enriches the diversity of feature data.Then,use eXtreme Gradient Boosting(XGBoost),1D Convolutional Neural Network(1D-CNN),Support Vector Regression(SVR)as the base learners,and Partial Least Squares Regression(PLSR)as the meta-learner to build a two-layer Stacking integrated learning model to predict the oil content in emulsified oil spills.The test set determination coefficients of emulsified gasoline and emulsified white oil obtained in the Stacking integrated learning model were 0.9824 and 0.9873,respectively,and the root mean square errors were 0.0410 and 0.0340,respectively.Compared to XGBoost,1D-CNN,SVR,and PLSR,the Stacking integrated learning model has better stability and accuracy.The above research results indicate that the detection method based on mid-infrared spectroscopy technology combined with LDA and Stacking

关 键 词：中红外光谱 乳化溢油 线性判别分析 特征波长 Stacking集成学习 

分 类 号：O433.4[机械工程—光学工程]