| 曹晓琼1，徐勇将1,2.基于拉曼光谱结合机器学习的热氧化葵花籽油
品质分析模型构建[J].中国油脂,2026,51(5):.[CAO Xiaoqiong1， XU Yongjiang1,2.Construction of model for quality analysis of thermally oxidized sunflower seed oil based on Raman spectroscopy combined with machine learning[J].China Oils and Fats,2026,51(5):.] |
| 基于拉曼光谱结合机器学习的热氧化葵花籽油
品质分析模型构建 |
| Construction of model for quality analysis of thermally oxidized sunflower seed oil based on Raman spectroscopy combined with machine learning |
| |
| DOI：10.19902/j.cnki.zgyz.1003-7969.250223 |
| 中文关键词: 热氧化油 拉曼光谱 峰强比 机器学习 品质预测 |
| 英文关键词:thermally oxidized oil Raman spectroscopy peak intensity ratio machine learning quality prediction |
| 基金项目:国家自然科学基金（32172136) |
|
|
| 摘要点击次数: 147 |
| 全文下载次数: 5 |
| 中文摘要: |
| 旨在为热氧化油脂品质的快速、准确、无损评价和预测提供参考，以在120~210 ℃下加热一定时间的葵花籽油作为热氧化油，采用拉曼光谱对热氧化油进行表征，通过提取与总极性组分含量（TPC）、过氧化值（POV）、酸值（AV）、自由基含量（spins）、氧化甘油三酯（ox-TGs）含量密切相关的拉曼光谱特征峰强比（PIRs），并结合加热时间、加热温度等参数作为输入变量，构建了基于支持向量回归（SVR）、反向传播神经网络（BPNN）、随机森林回归（RF）和梯度提升决策树回归（XGBoost）的预测模型，通过多元评估指标体系评估筛选最佳模型。结果表明：不同热处理条件下油样的拉曼光谱均在720、842、964、1 077、1 260、1 302、1 435、1 653、1 742、2 850、2 889、3 009 cm-1处有特征峰值，随着加热时间的延长，所有温度下964、1 260、3 009 cm-1处碳碳双键峰强与1 435 cm-1处峰强比均呈下降趋势，且在210 ℃时下降最快；4种预测模型中， SVR预测热氧化油品质指标模型整体表现较好，性能偏差比（RPD）均大于 1.5。综上，PIRs与品质指标之间存在高度相关性，通过SVR模型结合PIRs与加热时间和加热温度预测热氧化油品质的方法具有较佳的预测效果，能实现多指标同步预测，为油脂加工过程的实时质量控制提供了高效技术手段。 |
| 英文摘要: |
| Aiming to provide a reference for the establishment of a fast, accurate and non-destructive quality evaluation and prediction model for thermally oxidized oils, sunflower seed oils heated at 120-210 ℃ for certain durations were used as the thermally oxidized oil samples, and Raman spectroscopy was employed to characterize the thermally oxidized oils. Raman peak intensity ratios (PIRs) closely related to total polar compounds content(TPC), peroxide value (POV), acid value (AV), free radical content (spins), and oxidized triglycerides (ox-TGs) content were extracted. Together with heating time and heating temperature as input variables, prediction models based on support vector regression (SVR), backpropagation neural network (BPNN), random forest regression (RF), and gradient boosting decision tree regression (XGBoost) were constructed. The optimal model was selected through evaluation using a multi-dimensional assessment index system. The results showed that the Raman spectra of oil samples under different heat treatment conditions all exhibited characteristic peaks at 720, 842, 964, 1 077, 1 260, 1 302, 1 435, 1 653, 1 742, 2 850, 2 889 cm-1and 3 009 cm-1. With prolonged heating time, the PIRs of the carbon-carbon double bonds at 964, 1 260 cm-1 and 3 009 cm-1to the band at 1 435 cm-1 showed a decreasing trend under all temperatures, with the fastest decline observed at 210 ℃. Among the four prediction models, the SVR model for predicting quality indicators of thermally oxidized oils performed relatively well overall, with residual prediction deviation (RPD) values all greater than 1.5. In conclusion, there is a high correlation between PIRs and the quality indicators. The method using the SVR model combined with PIRs, heating time, and heating temperature to predict the quality of thermally oxidized oils demonstrates favorable predictive performance, enables simultaneous prediction of multiple indicators, and provides an efficient technical means for real-time quality control in oil processing. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
