基于群智能优化多分类器的DOM组分检测研究  

Research on DOM Component Detection Based on Multiple Classifiers Optimized by Swarm Intelligence Algorithm

作　　者：付丽辉 戴峻峰 王晓燕 FU Li-hui;DAI Jun-feng;WANG Xiao-yan(Faculty of Automation,Huaiyin Institute of Technology,Huai'an 223003,China;Faculty of Electronic Information Engineering,Huaiyin Institute of Technology,Huai'an 223003,China)

机构地区：[1]淮阴工学院自动化学院,江苏淮安223003 [2]淮阴工学院电子信息工程学院,江苏淮安223003

出　　处：《光谱学与光谱分析》2024年第8期2349-2356,共8页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(62141502);江苏省建设系统科技项目(2019ZD0064);淮安市“1111”工程合作项目(Z413H22507)资助。

摘　　要：可溶解有机物(DOM)是影响生态环境和居民生活安全的重要因素之一。当DOM总量达到一定水平时,将通过水体富营养化引致藻类爆发式生长,使得DOM成分更为复杂、影响更为严峻。常见的DOM检测方法虽然可以对DOM进行定性分析,但对于DOM组分的测定一直存在瓶颈,单传感器难以完成对水体DOM总量与组分的复杂性测试。提出基于等离子共振传感器(SPR)的交叉敏感性的DOM测试方案。利用群智能算法(PSO)训练的BP神经网络(BP),构建三个分类器。利用多模光纤,镀以55~85 nm的7种不同厚度的金膜,构成具有不同最佳折射率测量值的SPR传感探头,使各传感探头的最佳折射率测量值有效分布在1.33~1.43 RIU范围内,保证每个传感头在最佳测量区间内具有较好的灵敏度、线性度,在其他传感头对应的测量区间内,通过波长、谱宽和光强的变化,有着尽可能敏感的交叉响应。结合基于PSO-BP的三分类器的智能算法,通过对DOM的水样制备、DOM成分的测定、折射率的测量、SPR效应的测量、人工智能网络的训练、验证等实验步骤,实现对被测样本的SPR效应的共振波长、谱宽和光强的综合训练,从而完成对里运河(A)、洪泽湖(B)、公园景观湖(C)、校园景观湖(D)四种被测水体的五种DOM组分(酪氨酸类蛋白质、色氨酸类蛋白质、富里酸、溶解性微生物代谢产物、腐殖酸)浓度的有效预测。对四种不同被测水体(A、B、C、D)DOM组分浓度的最高预测率分别为81.2%(酪氨酸)、85%(色氨酸)、82%(色氨酸)、82.6%(色氨酸)。考察了响应参数及分类器个数对预测效果的影响,结果表明,与双分类器和单分类器相比较,三分类器的预测效果最好,对综合水样的五种不同DOM组分浓度的预测正确率分别为81.5%、84%、81%、82%、68.3%,验证了PSO-BP的多分类器及光纤SPR传感器在DOM组分预测中的正确性及可行性。Dissolved organic matter(DOM)is one of the important factors affecting the ecological environment and the safety of residents'lives.When the total amount of DOM reaches a certain level,it will lead to the explosive growth of algae through water eutrophication,which makes the DOM composition more complex and the impact more severe.Although the standard detection methods can qualitatively analyze DOM,there are always bottlenecks in determining DOM components,and it is difficult for a single sensor to complete the complicated test of the total amount and components of DOM in water.Therefore,a DOM test scheme is proposed based on the cross-sensitivity between SPR(Surface Plasmon Resonance,SPR)sensors.Three classifiers are constructed using the BP(Back Propagation,BP)neural network trained by the swarm intelligence algorithm(Particle Swarm Optimization,PSO).The multi-mode fibers are coated with seven kinds of gold films with different thicknesses of 55~85 nm to form the SPR sensing probes with different optimal refractive index measurements to ensure the best refractive index measurement value of each sensing probe is effectively distributed in the range of 1.33~1.43 RIU,and each sensing probe has good sensitivity and linearity in the best measurement range.In the measurement range corresponding to other sensing probes,there are cross-responses as sensitive as possible through the change of wavelength,spectral width and light intensity.Finally,combined with the intelligent algorithm based on three classifiers of PSO-BP,through the experimental steps of water sample preparation of DOM,determination of DOM composition,measurement of refractive index and SPR effect,training and verification of artificial intelligence network,to realize the comprehensive training of the resonance wavelength,spectral width and light intensity of SPR effect on the measured samples.Thus,five DOM components(tyrosine proteins,tryptophan proteins,fulvic acid,soluble microbial metabolites,humic acids)of the inner canal(A),Hongze Lake(B),Park Lan

关 键 词：多分类器 光纤等离子共振传感器 可溶性有机物组分检测 BP神经网络 PSO算法 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]