【作者】 李彦鹏；

【导师】 周建成； 薛谊；

【作者基本信息】 东南大学 ， 材料与化工（专业学位）， 2025， 硕士

【摘要】 橘皮（Citri Reticulatae Pericarpium,CRP）与青皮（Citri Reticulatae Pericarpium Viride,CRP-V）作为产量巨大的柑橘类加工副产物,其高值化利用不足构成资源浪费与产业升级的挑战。两者因采收期不同而呈现显著的化学成分差异,为研究天然产物化学多样性及其与功能（如香气、抗氧化活性）的关系提供了理想模型。然而,传统研究在热敏性成分保护、全谱化学信息获取及复杂构效关系解析方面面临瓶颈。本研究旨在通过整合先进技术,构建从差异化提取到应用评价的全链条研究体系,系统挖掘CRP与CRP-V的应用潜力,特别是在烟草增香提质方面的应用,为农业副产物的高效、精准利用提供科学支撑。具体研究内容如下:1.采用超临界CO₂萃取（Supercritical Fluid Extraction with CO₂,SFE-CO₂）技术,结合响应面法对CRP和CRP-V的SFE-CO₂提取工艺进行系统优化,确定了最佳工艺条件（CRP:35.45℃,29.76 MPa,乙醇30.35 ml/50 g;CRP-V:30.99℃,26.40 MPa,乙醇31.88 ml/50 g）,获得了具有高预测精度（R²>0.99）的模型,CRP和CRP-V的得率分别达到1.413%和1.941%。通过感官评价证实,CRP-SFE-CO₂提取物在卷烟应用中展现出优于传统水提和醇提产物的香气品质。2.利用超高效液相色谱-高分辨质谱和气相色谱-质谱技术,对SFE-CO₂、乙醇提取和水提取三种方法获得的CRP与CRP-V提取物进行了全组分化学成分分析。结果表明,SFE-CO₂法在捕获丰富且具代表性的挥发性有机物和独特的非挥发性有机物方面具有显著优势,其提取物呈现出与传统溶剂提取物截然不同的成分分布。同时,通过统计分析清晰揭示了CRP与CRP-V之间的显著化学组成差异。3.进一步地,为解析成分与气味的关系,本研究构建了一个标准化的气味预测数据集,并开发了一种基于消息传递神经网络（Message Passing Neural Networks,MPNN）的多标签气味分类模型。该模型在处理气味描述符模糊性、多标签复杂性和数据稀缺性方面表现出良好性能,并通过注意力机制实现了模型的可解释性,揭示了潜在的结构-气味关联。将此模型应用于预测CRP-SFE-CO₂提取物挥发性有机物的气味特征,结果显示SFE-CO₂提取物具有最丰富的预测香气标签,与感官评价结果高度吻合,验证了模型的有效性与实用价值。综上所述,本研究成功优化了CRP与CRP-V的SFE-CO₂提取工艺,系统阐明了提取方法与原料成熟度对化学成分谱的深刻影响,并创新性地构建了可解释的MPNN模型用于天然产物气味预测。研究成果不仅深化了对CRP与CRP-V化学成分的科学认知,也为这两种农业副产物在烟草等领域的高值化、精准化应用提供了关键的理论依据、技术方法和数据支持,展现了整合现代分析技术与人工智能促进农副产物高效利用的潜力。更多还原

【Abstract】 Citri Reticulatae Pericarpium（CRP）and Citri Reticulatae Pericarpium Viride（CRP-V）,as high-volume citrus processing byproducts,face challenges in valorization,representing both resource waste and a hurdle for industrial upgrading.Their significant differences in chemical composition,arising from distinct harvest times,offer an ideal model system for investigating the chemical diversity of natural products and its relationship with functional properties,such as aroma and antioxidant activity.However,traditional research methods encounter bottlenecks in preserving thermosensitive compounds,acquiring comprehensive chemical profile information,and elucidating complex structure-activity relationships.This study aimed to construct an integrated research pipeline,from differential extraction to application evaluation,by incorporating advanced technologies.The goal was to systematically explore the application potential of tangerine peel and green tangerine peel,particularly for tobacco aroma and quality enhancement,thereby providing scientific support for the efficient and precise utilization of these agricultural byproducts.The specific research contents are as follows:1.Supercritical CO₂ extraction（SFE-CO₂）technology,combined with Response Surface Methodology,was employed to systematically optimize the SFE-CO₂ extraction process for CRP and CRP-V.The respective optimal conditions were determined（CRP:35.45℃,29.76 MPa,30.35 ml/50 g;CRP-V:30.99℃26.40 MPa,31.88 ml/50 g）,yielding models with high predictive accuracy（R2>0.99）.The yields for CRP and CRP-V reached 1.413%and 1.941%,respectively.Sensory evaluation confirmed that the CRP SFE-CO₂ extract exhibited superior aroma characteristics in cigarette applications compared to traditional aqueous and ethanolic extracts.2.Ultra-High Performance Liquid Chromatography-Orbitrap Mass Spectrometry and Gas Chromatography-Mass Spectrometry hyphenated techniques were utilized for the comprehensive chemical characterization of tangerine peel and green tangerine peel extracts obtained using SFE-CO₂,ethanol extraction,and water extraction methods.The results indicated that the SFE-CO₂ method offers significant advantages in capturing a rich and representative profile of volatile organic compounds and unique non-volatile organic compounds,yielding extracts with chemical fingerprints distinct from those obtained by traditional solvent extraction.Furthermore,the study clearly revealed significant differences in chemical composition between tangerine peel and green tangerine peel attributable to maturity stage variations,particularly in the diversity of non-volatile organic compounds and the constitution of volatile organic compounds.3.To elucidate the relationship between chemical components and odor,this research constructed a standardized odor prediction dataset and developed a multi-label odor classification model based on Message Passing Neural Networks（MPNN）.The model demonstrated good performance in addressing the semantic ambiguity of odor descriptors,the complexity of multi-label classification,and data scarcity.Model interpretability was achieved via attention mechanisms,revealing potential structure-odor relationships.Application of this model to predict the odor characteristics of volatile organic compounds in the CPR-SFE-CO₂ showed that the SFE-CO₂ extract possessed the richest predicted aroma profile,correlating strongly with sensory evaluation results,thus validating the model’s effectiveness and practical utility.In summary,this research successfully optimized the SFE-CO₂ extraction process for tangerine peel and green tangerine peel,systematically elucidated the profound impact of extraction methods and material maturity on chemical profiles,and innovatively constructed an interpretable MPNN model for natural product odor prediction.The findings not only deepen the scientific knowledge of the chemical constituents of tangerine peel and green tangerine peel but also provide a critical theoretical foundation,technical methodologies,and data support for the valorization and precise application of these agricultural byproducts in fields such as tobacco.The work highlights the potential of integrating modern analytical techniques and artificial intelligence to promote sustainable resource utilization.更多还原