基于卷积神经网络的成品油红外光谱分类分析  

Classification and Analysis of Infrared Spectrum for Refined Oil Products Based on Convolutional Neural Network

作　　者：马松浩 王竞宇 张晓雪[2] 宋权威 王丽荣[5] 齐晗兵[2] Ma Songhao;Wang Jingyu;Zhang Xiaoxue;Song Quanwei;Wang Lirong;Qi Hanbing(China National Petroleum Corporation Northeast Sales Daqing Branch;School of Civil and Architectural Engineering,Northeast Petroleum University;State Key Laboratory of Petroleum Pollution Control;CNPC Research Institute of Safety&Environment Technology;Technology Research Institute,No.6 Oil Production Plant,Daqing Oilfield)

机构地区：[1]中国石油天然气股份有限公司东北销售大庆分公司 [2]东北石油大学土木建筑工程学院 [3]石油石化污染物控制与处理国家重点实验室 [4]中国石油集团安全环保技术研究院有限公司 [5]大庆油田第六采油厂工艺研究所

出　　处：《油气田环境保护》2024年第4期42-47,共6页Environmental Protection of Oil & Gas Fields

基　　金：黑龙江省省属本科高校基本科研业务费东北石油大学青年科学基金(2019QNL-14)。

摘　　要：针对石油类产品在运输及使用过程中泄漏污染引起的环境问题,文章以4种成品油为研究对象,基于卷积神经网络(CNN)对油品红外光谱进行分类分析,为石油类产品泄漏追踪溯源。实验测量了4种成品油及其混合物共387组红外透射光谱,采用Savitzky-Golay多项式平滑法(S-G)、标准正态变换(SNV)和多元散射校正(MSC)3种方法对光谱数据进行预处理,分别建立了预处理前后CNN分类模型。研究结果表明:预处理后的光谱数据建立的CNN模型分类精度均高于原始数据,其中SNV预处理的光谱数据表现出最佳分类精度为0.9744,损失值为0.2579。该研究结果说明基于神经网络结合红外透射光谱的检测方法对成品油品种分类是可行的,且该项成果为后续实现石油类污染物高效、快速检测提供理论支持。For the environmental problems caused by the leakage and pollution of petroleum products in the process of transportation and use,the infrared spectrum for four kinds of refined oil products was classified and analyzed based on convolutional neural network(CNN),so as to trace the source of oil products leakage.In this paper,387 groups of infrared transmission spectra of four kinds of refined oil products and their mixtures were measured.Three methods including Savitzky-golay polynomial smoothing(S-G),standard normal transformation(SNV)and multiple scattering correction(MSC)were used to preprocess the spectral data.The classification models of CNN before and after preprocessing were established.respectively.The results showed that the accuracy of the CNN classification models established by the preprocessed spectral data was higher than that of the original data,and the spectral data preprocessed by SNV showed the best model classification accuracy of 0.9744,with a loss value of 0.2579.The results showed that the detection method based on CNN combined with infrared transmission spectrum was feasible for the classification of refined oil varieties,and it provided theoretical support for the subsequent realization of efficient and rapid detection of petroleum pollutants.

关 键 词：卷积神经网络(CNN) 成品油 红外光谱 光谱预处理 透射率 

分 类 号：X8[环境科学与工程—环境工程] TE991[石油与天然气工程—石油机械设备]