目的 分析比较5种算法对杏贝止咳颗粒（Xingbei Zhike Granules，XZG）溶解性的模型预测效果，确定最优算法和影响颗粒溶解性的中间体粉末关键物料属性（critical material attributes，CMA），提升XZG的溶解性。方法 以25个制粒前粉末物料属性为输入，颗粒溶化率为输出，对比偏最小二乘（partial least squares，PLS）、决策树（decision tree，DT）、随机森林（random forest，RF）、反向传播神经网络（back propagation neural network，BPNN）和支持向量机回归（support vector regression，SVR）算法的模型拟合与预测效果，同时结合变量重要性与方差膨胀因子（variance inflation factor，VIF），筛选关键物料属性。结果 5个算法中，RF算法的模型拟合与预测效果最好，训练集决定系数（R²）为0.865，测试集R²为0.854，均方根误差（root mean square error，RMSE）为1.375，平均绝对百分比误差（mean absolute percent error，MAPE）为1.153%。筛选出的关键物料属性有休止角（α）、崩溃角（β）、平板角（γ）、吸湿性（H）、含水量（HR）、100～212 μm颗粒含量（F_m）、长度平均径（D₂₁）。结论 RF算法建立的模型更适合预测XZG的溶解性；通过控制制剂成型用主药和辅料的关键物料属性，可以提高XZG的溶解性，为提升中药颗粒剂的溶解性和品质提供新的思路。

Objective To analyze and compare the model prediction effects of five algorithms on the solubility of Xingbei Zhike Granules (XZG, 杏贝止咳颗粒), to determine optimal algorithm and the critical material attributes (CMA) affecting the solubility of particles, and to improve the solubility of XZG. Methods Taking 25 powder material properties before granulation as input, and the solubility of particles as output. The model fitting and prediction effects of partial least squares (PLS), decision tree (DT), random forest (RF), back propagation neural network (BPNN), and support vector regression (SVR) algorithms were compared. At the same time, combined with the importance of variable and variance inflation factor (VIF), the key material properties were screened. Results The RF algorithm has the best model fitting and prediction effect. The training set determination coefficient is 0.865, the test set determination coefficient is 0.854, the root mean square error is 1.375, and the mean absolute percentage error is 1.153%. The CMA screened by had collapse α, β, γ, H, HR, F_m and D₂₁. Conclusion The model established by the RF algorithm is more suitable for predicting the solubility of XZG. The solubility of XZG can be improved by controlling the key material properties of the main drugs and excipients for preparation molding, which provides a new idea for improving the solubility and quality of traditional Chinese medicinal granules.

R283.6

国家工信部2023年产业基础再造和制造业高质量发展专项（TC2308068）