目的 通过预测杏贝止咳颗粒（Xingbei Zhike Granules，XZG）颗粒脆碎度，结合最优算法，识别中间体的关键物料属性（critical material attributes，CMAs）。方法 以6个中间体物料的共60个物性参数作为输入，颗粒脆碎度作为输出，使用偏最小二乘回归算法（partial least squares，PLS）、决策树算法（classification and regression tree，CART）、广义路径追踪算法（generalized path seeker，GPS）、多元自适应回归样条算法（multivariate adaptive regression splines，MARS）、随机森林算法（random forest，RF）和树网随机梯度提升算法（TreeNet）共6种机器学习算法建立预测模型，根据模型拟合效果与预测误差确定最佳算法，筛选中间体的CMAs。结果 基于GPS算法建立的预测模型表现最佳，可准确预测出XZG的颗粒脆碎度。其训练集决定系数（R²_c）为0.981，测试集决定系数（R²_p）为0.966，训练集均方根误差（root mean square error of calibration，RMSEC）为0.976，测试集均方根误差（root mean square error of prediction，RMSEP）为1.304，平均相对预测误差（average relative prediction error，ARPE）为4.72%，低于5%。共筛选出6个中间体的15个CMAs。结论 基于中间体物性参数构建的预测模型，为干法制粒的颗粒脆碎度预测提供了一个新的思路；可以筛选出影响干法制粒颗粒脆碎度的关键物料属性，有助于提升产品质量。

Objective To identify critical material attributes (CMAs) of intermediates by predicting the granule friability of Xingbei Zhike Granules (XZG, 杏贝止咳颗粒) combined with the optimal algorithm. Methods Sixty physical parameters of six intermediate materials were selected as input variables, and granule friability was chosen as the output variable. Six machine learning algorithms, including partial least squares (PLS), classification and regression tree (CART), generalized path seeker (GPS), multivariate adaptive regression splines (MARS), random forest (RF), and TreeNet, were employed to establish predictive models. The optimal algorithm was selected based on model fitting performance and prediction error to screen CMAs from the intermediates. Results The prediction model constructed with the GPS algorithm exhibited the best performance for accurately predicting the granule friability of XZG. The model yielded a coefficient of determination for calibration (R²_c) of 0.981, a coefficient of determination for prediction (R²_p) of 0.966, a root mean square error of calibration (RMSEC) of 0.976, a root mean square error of prediction (RMSEP) of 1.304, and an average relative prediction error of 4.72%, which is lower than 5%. A total of 15 CMAs from six intermediate materials were identified. Conclusion The predictive model based on intermediate material attributes offers a novel approach for predicting granule friability in dry granulation processes. It effectively identifies critical material attributes affecting granule friability, thereby contributing to improved product quality.

R283.6

国家工信部产业基础再造和制造业高质量发展专项(TC2308068)