基于QPSO-MLSSVM算法的拉曼光谱检测四组分调和油含量

提出了一种运用量子粒子群(quantum-behaved particle swarm optimization，QPSO)算法优化多输出最小二乘支持向量机(multi-output least squares support vector machine，MLSSVM)的新混合优化算法。该算法结合激光拉曼光谱技术可实现对四组分食用调和油中花生油、芝麻油、葵花油和大豆油的快速定量鉴别。采用基线校正去除背景荧光，结合Savitzky-Golay Filters光谱平滑法对原始拉曼光谱进行预处理。构建基于QPSO-MLSSVM混合优化算法的定量分析模型，并采用20个组分组成的预测集对其进行模型校验。实验结果表明，基于QPSO-MLSSVM混合优化算法的定量分析模型对于四组分调和油的预测效果良好，均方差(mean square error, MSE)为0.0241，低于0.05，各油分预测相关系数均高于98%。研究结果充分表明, 应用激光拉曼光谱技术结合QPSO-MLSSVM算法，对四组分调和油中各油分进行快速定量检测可行，具备较强的自适应能力和良好的预测精度，可以满足多组分调和油的成分鉴别。

Raman Spectra Based on QPSO-MLSSVM Algorithm to Detect the Content of Four Components Blent Oil

This paper presents a new hybrid optimization algorithm based on the multi-output least squares support vector machine (MLSSVM) which is optimized by quantum-behaved particle swarm optimization (QPSO). The rapid quantitative identification for the peanut oil,sesame oil, sunflower oil and soybean oil in the four - component edible blending oil can be realized with the algorithm combined with laser Raman spectroscopy. The background fluorescence was removed by baseline correction, and Savitzky-Golay filters spectral smoothing method is used for the pretreation of original Raman spectra. The quantitative analysis model based on QPSO-MLSSVM hybrid optimization algorithm is established, and the prediction set composed of 20 components is used to verify the model. The experimental result shows that it is effective for the prediction of four-component blending oil with the quantitative analysis model based on QPSO-MLSSVM hybrid optimization algorithm, and the Mean Square Error (MSE) is 0.024 1, which is less than 0.05, the correlation coefficients of each component were above 98%. The results show that it is feasible to detect the content of each oil of four-component blending oil by laser Raman spectroscopy combined with QPSO-MLSSVM algorithm, it has strong adaptive ability and good prediction accuracy that can satisfy the multi-component mixed oil component identification.