目的 基于近红外光谱（NIRS）技术，提出一种移动块标准偏差（MBSD）与移动F检验相结合的双算法耦合策略，实现制药过程混合终点的精准判定。方法 以低质量比（1∶19）模型药（对乙酰氨基酚-糊精）体系为研究对象，采用NIRS技术开展混合过程在线光谱采集，通过重力感应触发模式保证采集数据的一致性；对原始光谱进行SG卷积平滑、SNV变换及一阶导数预处理并优化关键参数，结合MBSD法监测光谱波动趋势、移动F检验法开展方差齐性统计分析，构建双算法耦合的动态协同判定模型，设置趋势稳定、统计稳定及持续时间三重约束条件，实现混合终点的自动化精准判定；同时采用高效液相色谱（HPLC）法进行离线取样验证，参照相关技术指导原则评价混合均匀性。结果 确定SG卷积平滑最优参数为窗口点数11、二阶多项式，MBSD法最优窗口大小为15；双算法耦合策略有效规避了单一算法的“伪均匀”误判问题，6批模型药的协同判定终点集中在14～15 min，该时间点离线HPLC检测结果显示，所有批次样品质量分数的RSD均≤5%，且单一样品质量分数均在均值± 10.0%范围内，在线判定结果与离线验证结果高度吻合。结论 MBSD与移动F检验双算法协同判定策略，可实现制药混合过程的秒级响应与无损实时监控，兼具高灵敏度与强鲁棒性，有效提升混合终点判定的准确性，可适配制药工业实时放行检验的发展需求，为药物混合过程的在线质量监控提供可靠的技术方法。

Objective Based on near-infrared spectroscopy (NIRS) technology, a dual-algorithm coupling strategy combining moving block standard deviation (MBSD) and moving F-test was proposed to accurately determine the end point of the mixing process in pharmaceutical manufacturing. Methods A low-dose (1∶ 19) model drug system was used as the research object. Online spectral data collection during the mixing process was conducted using NIRS technology, and the consistency of the collected data was ensured through a gravity induction trigger mode. The original spectra were preprocessed with SG convolution smoothing, SNV transformation, and first-order derivative, and the key parameters were optimized. The MBSD method was used to monitor the spectral fluctuation trend, and the moving F-test was used to conduct variance homogeneity statistical analysis. A dynamic collaborative determination model based on the dual-algorithm coupling was constructed, and three constraint conditions, namely trend stability, statistical stability, and duration, were set to achieve precise and automated determination of the mixing end point. At the same time, offline sampling verification was performed using high-performance liquid chromatography (HPLC), and the homogeneity of the mixture was evaluated in accordance with relevant technical guidelines. Results The optimal parameters for SG convolution smoothing were determined to be a window size of eleven and a second-order polynomial, and the optimal window size for the MBSD method was fifteen. The dual- algorithm coupling strategy effectively avoided the “pseudo-uniformity” misjudgment problem of a single algorithm. The collaborative determination end points of the six batches of model drugs were concentrated at 14 to 15 minutes. The HPLC detection results at this time point showed that the RSD of the mass fraction of all batches of samples was ≤ 5%, and the mass fraction of each sample was within ±10.0% of the mean. The online determination results were highly consistent with the offline verification results. Conclusion The dual-algorithm collaborative determination strategy of MBSD and moving F-test can achieve second-level response and nondestructive real-time monitoring of the pharmaceutical mixing process, with high sensitivity and strong robustness. It effectively improves the accuracy of the determination of the mixing end point and can meet the development needs of real-time release testing in the pharmaceutical industry, providing a reliable technical method for online quality monitoring of the drug mixing process.

R283.6

国家“重大新药创制”科技重大专项资助项目（2018ZX09201010-001-004）