基于拉曼光谱和多传感器信息融合的养殖水体三氮快速检测机理研究

Three-nitrogen in aquaculture water, that is the ammonia, nitrite and nitrate,is closely related to the aquaculture industry. The excessive three-nitrogen directly produces the neurological toxicity to aquatic animals, which will affect their growth, development and even deaths. However, fast detection of three-nitrogen is still open question for the aquaculture industry since three-nitrogen can transform into one another and be easily affected by a variety of factors such as temperature, dissolved oxygen, pH and so on. In this project, the three-nitrogen is taken as the object of study, three-nitrogen fast detection mechanism and quantitative analysis mechanism based on resonance Raman spectroscopy and liquid core fiber-optics technology are explored to resolve to the question of online detection; by studying aquaculture water key factor for nitrogen and mutual influence relationships, establish the noise reduction and soft computing model to solve the stability problem of three-nitrogen online detection; by researching the criterion on global optimality and the adaptive approach of parameters for multi-source information fusion, to build the accurate detection model for the three-nitrogen in aquaculture water based on Raman spectroscopy and multi-sensors data fusion. Finally, the implementation of this project will provide a theoretical and scientific basis for research on fast on-line detection sensor of three-nitrogen.

养殖水体中过量的氨氮、亚硝酸盐氮、硝酸盐氮（俗称三氮），对养殖对象直接产生毒性，影响其生长、发育，甚至造成大批死亡。三氮之间可以相互转化，同时受水温、溶解氧、pH等多种因素影响，极易变化，三氮在线快速准确检测一直是困扰水产养殖界的公认难题。本项目拟以水产养殖水体中的三氮为研究对象，采用激光共振拉曼技术与液芯光纤技术，探明基于拉曼光纤的三氮快速检测机理及定量分析方法，实现三氮的快速测定；研究水产养殖水体三氮相互转化规律及其关键影响因子作用关系，建立降噪和软计算模型，解决三氮拉曼检测结果稳定性差的问题；探索多源信息融合全局最优判别准则及信息融合模型参数自适应确定方法，构建拉曼光谱和多传感器信息融合模型，提高三氮快速检测的精度。通过上述研究，为三氮在线快速检测传感器研发提供理论和科学依据。

氨氮、亚硝酸盐氮、硝酸盐氮俗称“三氮”，过量的三氮会引起养殖水体的富营养化，破坏水体生态平衡，还会对养殖对象直接产生毒性，影响其生长、发育，甚至造成大批死亡，严重影响水产养殖业产量。本项目拟以水产养殖水体中的三氮为研究对象，采用激光共振拉曼技术与液芯光纤技术，探明基于拉曼光纤的三氮快速检测机理及定量分析方法，实现三氮的快速测定；研究水产养殖水体三氮相互转化规律及其关键影响因子作用关系，建立降噪和软计算模型，解决三氮拉曼检测结果稳定性差的问题；探索多源信息融合全局最优判别准则及信息融合模型参数自适应确定方法，构建拉曼光谱和多传感器信息融合模型，提高三氮快速检测的精度。.课题组通过检测PATP和亚硝酸根离子生成偶氮染料的SERS来间接确定亚硝酸根离子的含量。通过巯基、氨基的强吸附性和偶氮染料在532 nm处的共振作用，两次提高了拉曼强度，有效的抑制了荧光干扰，极大的提高了待测样品拉曼强度和选择性。该方法检测亚硝酸盐的线性范围是0.1-5.0 μmol/L，R2=0.991，在1138 cm-1处的检测限为15.5 nmol/L。在水产养殖水体中亚硝酸盐检测的相对标准误差在1.87~5.30%之间，具有较好选择性和准确性。.针对水产养殖水质氨氮含量极低导致的难以在线测量的难题，同时，水质氨氮也具有变化非线性、多因子影响的特点，提出了一种基于经验模态的粒子群优化极限学习机的氨氮软测量方法。该方法首先利用经验模态法对氨氮进行固有模态分解，获得极限学习机隐含层节点数；然后，利用粒子群算法对极限学习机的输入层权值和偏置进行选择；最终通过易在线测量的水质和气象参数实现氨氮含量的估算。结果表明：提出EMD-IPSO-ELM 相比标准的ELM，EMD-IPSO-ELM 相比标准的ELM，其评价指标MAE、MSE、MAPE 分别降低了11.57%，26.02%和14.41%，该方法可实现水质氨氮的精准估算。同时，论文采用分层软阈值小波降噪对氨氮数据进行预处理，构建的WA-EMD-IPSO-ELM 模型较采用原始数据训练获得的EMD-IPSO-ELM模型，其结果评价指标MAE、MSE和MAPE分别下降了26.27%，48.35%和30.74%，充分证明了数据预处理可有效提高预测精度。