量子遗传-神经网络算法的润滑油动力粘度值可见近红外光谱分析  

Visible and Near Infrared Spectral Analysis of the Lubricating Oil Dynamic Viscosity Based on Quantum Genetic-Neural Network Algorithm

作　　者：刘晨阳 唐兴佳[3] 于涛[3] 王泰升[1] 卢振武[1] 鱼卫星[3] LIU Chen-yang;TANG Xing-jia;YU Tao;WANG Tai-sheng;LU Zhen-wu;YU Wei-xing(R&D Center of Precision Instruments and Equipment,Changchun Institute of Optics,Fine Mechanics&Physics,Chinese Academy of Sciences,Changchun 130033,China;University of Chinese Academy of Sciences,Beijing 100049,China;Key Laboratory of Spectral Imaging Technology,Xi’an Institute of Optics and Precision Mechanics,Chinese Academy of Sci-ences,Xi’an 710119,China)

机构地区：[1]中国科学院长春光学精密机械与物理研究所精密仪器与装备研发中心,吉林长春130033 [2]中国科学院大学,北京100049 [3]中国科学院西安光学精密机械研究所,中国科学院光谱成像技术重点实验室,陕西西安710119

出　　处：《光谱学与光谱分析》2020年第5期1634-1639,共6页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金面上项目(61474156);中国科学院A类战略先导专项(XDA23040101)资助。

摘　　要：润滑油动力粘度是划分润滑油品质的重要依据之一,高铁变速箱润滑油需要进行实时、快速、无损的检测,因此提出一种基于可见近红外光谱微型模块结合量子遗传-神经网络算法对润滑油粘度值进行定量分析的新方法。不仅实现了高铁变速箱润滑油动力粘度的无损快速实时检测,还进一步提高了对润滑油动力粘度预测的精度。微型光谱仪具有性能优良、体积小巧等优势,在便携式光谱无损检测方面用途越来越多。在这里,选用可见短波近红外和近红外波导光栅两种微型光谱模块进行光纤耦合,实现了330~1700 nm可见-近红外波段光谱拼接。首先我们采用该组合微型光谱仪对13种不同粘度的润滑油共78个样本进行光谱扫描得到原始光谱数据。原始光谱经过Savitzky-Golay卷积平滑后,再一阶求导,可以有效地消除基线漂移和背景噪声。然后采用主成分分析和马氏距离相结合的方法来识别浓度界外样本,剔除界外样本3个。最后采用BP(back propagation)神经网络和量子遗传神经网络两种回归算法分别建立定量分析模型,并对比分析了两种算法的性能。量子遗传算法是量子计算和遗传算法相结合地一种概率进化算法,采用量子染色体的形式,利用量子逻辑门进行全局搜索,从而可以利用量子遗传算法优化神经网络地权重和阈值,提高建模效率和精度。分别用BP神经网络算法和量子遗传-神经网络算法进行建模仿真,从75个样本随机抽取10个样本作为预测集,其余65个为建模集。在量子遗传寻优算法中,其种群数目设置为40,终止代数为200,寻优结果表明该算法在训练81代后可快速得到最优解。比较两种建模算法的预测结果,采用量子遗传-神经网络算法相比BP神经网络算法得到的粘度预测结果均方根误差从0.3455降低至0.0294,决定系数从0.8504升至0.9799,可知量子遗传-神经网络算法的预测能�Dynamic viscosity is one of the most important quality factors of lubricating oil.For the safety of high-speed railway,it is necessary to develop a real-time,fast and non-destructive method to monitor the status of the gearbox.Here we propose a new method that utilizes the quantum genetic-neural network algorithm to quantitatively analyze the visible and near-infrared spectra of lubricant acquired by a micro-spectrometer module.The method not only realizes non-destructive rapid real-time detection of the dynamic viscosity of high-speed railway transmission lubricating oil,but also further improves the prediction accuracy of the lubricating oil dynamic viscosity.Thanks to its excellent performance and small size,the miniature spectrometer has been widely used as a portable and nondestructive device.Here,two kinds of micro-spectral modules with visible/short-wave-infrared and near-infrared waveguide gratings are coupled with optical fibers and obtain a wide spectral range from 330 to 1700 nm.Here the integrated waveguide and propagating makes the spectrometer compact and small.In experiment,a total of 78 lubricant samples with 13 different viscosity lubricants were prepared for spectral measurement by the micro-spectrometer.The raw spectral data was pre-processed using the Savitzky-Golay convolution smoothing and the first-order differentiation to eliminate the baseline drift and background noise.Next,principal component analysis and Mahalanobis distance algorithm were used to identify the samples outside the concentration boundary,and three out-of-bound samples were excluded.Finally,the BP neural network and the quantum genetic neural network methods were employed for quantitative analyses and the results are compared,respectively.The quantum genetic algorithm is a probabilistic evolutionary algorithm that combines the advantages of quantum computing and genetic algorithm.It uses the form of quantum chromosomes and quantum logic gates for global searching.Therefore,the quantum genetic algorithm can be used to opti

关 键 词：可见-近红外光谱微型模块 润滑油 动力粘度 量子遗传算法 神经网络算法 

分 类 号：O433.4[机械工程—光学工程]