基于深度学习与Box-Behnken响应面法的柴达木枸杞甜菜碱提取工艺优化研究

Optimization of extraction process of betaine from Lycium barbarum L.from Qaidam Basin based on deep learning and Box-Behnken response surface methodology

本文对柴达木枸杞甜菜碱提取工艺进行了优化。以柴达木枸杞为原料,采用超声提取法,以超声时间、温度、物料比以及乙醇浓度为考察因素,分别使用Box-Behnken响应面法以及深度神经网络模型两种方法对提取工艺进行优化并验证,得到的相对标准差分别为2.73%和2.06%。其中,使用深度神经网络模型的相对误差较小,且得到的最优预测条件在实验验证中的甜菜碱得率优于Box-Behnken响应面法的预测条件。最终得到最佳工艺条件为:超声时间为46 min、提取温度为71℃、料液比为1∶24、乙醇浓度为83%。该结果表明,通过深度神经网络模型优化得到的提取工艺稳定可靠,得到的提取工艺能够提高柴达木枸杞甜菜碱的得率,为后续的生产提供参考。

In this paper,the extraction process of betaine from Lycium barbarum L.from Qaidam Basin was optimized.Ultrasonic extraction method was used with Lycium barbarum L.from Qaidam Basin as the raw material,and ultrasonic time,temperature,material-to-liquid ratio,and ethanol concentration were as the factors for optimization.Box-Behnken response surface methodology and deep neural network model were employed to optimize and validate the extraction process.The validity of the two optimization methods was verified,with relative standard deviations of 2.73%and 2.06%for Box-Behnken response surface methodology and deep neural network model,respectively.The latter method exhibited smaller relative error and provided optimal prediction conditions that yielded higher betaine yield than those predicted by Box-Behnken response surface methodology.The optimal extraction conditions were determined as follows:ultrasonic time of 46 min,extraction temperature of 71℃,material-toliquid ratio of 1∶24,and ethanol concentration of 83%.These findings demonstrated that the optimized extraction process obtained through deep neural network model was stable and reliable,and could improve the betaine yield of Lycium barbarum L.from Qaidam Basin,providing a reference for subsequent production.