SVR算法的生物混液左旋多巴含量紫外光谱建模

UV Modeling of Levodopa Concentration Based on SVR Algorithm

以酪氨酸和左旋多巴混合溶液中左旋多巴的紫外光谱数据为研究对象,首先用Kennard-Stone算法对样品集进行分割;然后使用ν-SVR和ε-SVR算法对核进行建模,以建立不同的核。紫外光谱定量分析模型中左旋多巴含量的功能;最后,采用粒子群算法对参数进行了优化,并与传统的PLS算法进行了比较。实验结果表明,由ν-SVR,ε-SVR和PLS建立的左旋多巴含量校正模型具有较高的准确性,预测性能略有不同。在预测集实验中,PLS,ν-SVR(RBF)和ε-SVR(RBF)算法的预测均方根误差分别为1.755、0.826和0.68。实验已经证明了使用紫外光谱法快速测定左旋多巴含量的有效性以及基于径向基函数的ε-SVR的建模优势。

Taking the concentration of levodopa in the mixed solution of tyrosine and levodopa as the research object,the Kennard-Stone algorithm is used to divide the sample set,and the ν-SVR and ε-SVR models are combined. Finally,particle swarm optimization algorithm is used to optimize the parameters. The UV quantitative analysis model of levodopa concentration with different kernel functions is established. The results show that the correction models of levodopa concentration established by the three methods of ν-SVR,ε-SVR and PLS have high accuracy,and the prediction performance is slightly different. In the prediction set experiment,the RMS errors of PLS,ν-SVR(RBF)and ε-SVR(RBF)are 1.755,0.826 and 0.68,respectively. The experimental results verify the validity of using UV spectrum to determine the concentration of levodopa,and the modeling advantage of ε-SVR based on radial basis function.