近红外光谱和XGBoost的流化床干燥过程水分含量在线检测  

Moisture Content Online Detection in Fluidized Bed Drying Process Based on Near Infrared Spectroscopy and XGBoost

作　　者：何帅 周杰 张福林 穆国庆 HE Shuai;ZHOU Jie;ZHANG Fu-lin;MU Guo-qing(School of Information and Control Engineering,Qingdao University of Technology,Qingdao 266520,China)

机构地区：[1]青岛理工大学信息与控制工程学院,山东青岛266520

出　　处：《光谱学与光谱分析》2024年第12期3347-3352,共6页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(62303257);山东省自然科学基金项目(ZR2023QF008);山东省高等学校青年创新团队发展计划项目(2023KJ119)资助。

摘　　要：含水量对化学和制药颗粒产品的属性(例如稳定性和可压缩性)具有重要影响。传统的流化床干燥过程水分检测是采用传统仪表检测过程中的湿度、温度等表征变量进而推测水分含量,这种方法常存在检测不准确、具有滞后性等,已经很难满足现代生产的需求。近红外(NIR)光谱作为一种新型传感器技术可以从分子层面获取过程信息,操作简单,分析速度快,无需对样本预处理等优点被广泛应用于很多领域。目前NIR光谱分析方法主要基于采集样本做离线检测,难以体现生产过程的实时状态。多数情况下采集的NIR光谱的吸收峰严重重叠,导致NIR光谱的有效信息被各种噪声掩盖。需要采用合适的分析工具进行近红外数据分析和有效信息的提取。传统的算法模型多采用线性或单模型方法,难以有效解决NIR光谱有效信息提取。采用批次颗粒流化床干燥(FBD)过程为检测对象,将近红外光谱应用于流化床制粒干燥过程中,联合XGBoost算法建立颗粒含水量在线测量模型。通过白鲸优化算法获得了模型的最佳参数,进而通过真实的流化床干燥实验验证了该方法的有效性。验证实验中,选取包括水分特征峰且信号较为稳定的波数(4798~9423 cm^(-1))进行建模。采集的4个批次数据中3个独立批次数据作为训练集来训练模型,第4个批次数据用于测试模型。以均方根误差(RMSE)和决定系数(R^(2))两个指标评价对建立的模型并进行评估,结果表明优化后的XGBoost模型在各项指标上表现优于PLS和BP-ANN算法建立的模型。所提出的基于近红外光谱和XGBoost的水分含量在线检测模型为流化床干燥过程水分含量的在线检测提供了新思路。Moisture content significantly impacts the properties(e.g.,stability and compressibility)of chemical and pharmaceutical granular products.The traditional fluidized bed drying process moisture detection uses traditional instrumentation to detect the process of humidity,temperature,and other characterization variables and then infer the moisture content;this method often produces inaccurate detection,has a lag and other shortcomings,it has been difficult to meet the needs of modern production.Near-infrared(NIR)spectroscopy,as a new sensor technology,can be obtained from the molecular level of process information;its operation is simple,has fast analysis speed,and there is no need for sample pre-processing and other advantages,so it is widely used in many fields.However,existing NIR spectroscopic analysis methods are mainly based on offline detection of collected samples,which makes it difficult to reflect the real-time status of the production process.At the same time,in most cases,the absorption peaks of the collected NIR spectra overlap severely,resulting in the effective information of the NIR spectra being masked by various noises.Therefore,it is necessary to use suitable analysis tools for NIR data analysis and effective information extraction.Traditional algorithmic models mostly use linear or single-model methods,which makes it difficult to effectively solve the problem of effective information extraction from NIR spectra.Thus,in this paper,the fluidized bed drying(FBD)process of batch particles is used as the detection object,and near-infrared spectroscopy is applied to the fluidized bed granulation and drying process,which is combined with the XGBoost algorithm to establish an on-line measurement model of moisture content of particles.The Beluga whale optimization obtained the optimal parameters of the model,and then the validity of this approach was verified by the real fluidized bed drying experiments.For the validation experiments,the wave numbers(4798 to 9423 cm^(-1)),which include the characteristic p

关 键 词：近红外光谱 流化床干燥 在线检测 XGBoost 

分 类 号：O433.4[机械工程—光学工程]