近年来国内中医药事业迎来了发展窗口期，其发展也被提升到民生的高度。中药是中医学科主要的治疗手段，中药房也是中医医院的核心医技部门。由于中药饮片用量大、品种多、库存成本高等特点，一直是中药房库存管理的重点和难点。而涵盖了饮片的采购计划、贮藏养护等内容的库存管理工作，也不仅与采购与管理成本控制密切相关，其库存环节也与到临床治疗效果有关。 在对三家中医医院中药房库存管理情况进行调研的基础上，本文选择T医院中药房作为案例医院，该医院为三甲综合医院，中药房承担着全院门诊、病房、代煎的饮片供应和调剂任务，饮片库存管理业务较有代表性。本文从需求预测着手，探究常用饮片的需求变化规律，以从源头减少库存及相应成本为库存管理优化目标。具体研究方法为通过抽取T医院销量排名前10的饮片品种出库数据作为研究样本，经过预处理后将其整理成训练及测试数据集，进而使用较适用于时间序列分析的经典ARIMA模型族，以及RNN-LSTM为代表的机器学习算法，分别进行拟合训练及建模，并在近期场景下对未来需求量做出预测。将预测结果与实际值对比表明，两类模型预测方法都可以呈现较好的准确度并可相互交叉验证，对其各自的适用范围、可拓展性、建模工作量及难度等方面进行了比较，总体而言在饮片需求预测中，应用机器学习算法更为简便。根据上述需求预测结果，进一步设定安全库存以应对需求中存在的不确定性中，及超出预期的情况。在安全库存的基础上，通过建立Max/Min补货策略设定再订货点及最高库存水平，并将其与同期实际值相比较，发现可显著降低采购等库存成本及相关库存管理成本，从而构成一套较完整的需求预测与库存管理优化方法，可供医院中药房制定库存计划及优化库存管理时参考。
An opportunity window of Traditional Chinese Medicine has been emerged in recent years. The development of TCM has been leveled up to the extent of people’s standard of living. TCM therapy including herbal slices is the main cure of this discipline as well, and its pharmacy is the core medical unit in TCM hospital. The inventory management of Chinese crude drug consists of not only regular planning, repurchase, keeping and maintenance，but the control of holding and management cost. Even the clinical cure effect has something to do with this process, thanks to its unique characteristic. Based on the field study of inventory management of three TCM medical facilities, this thesis selects a Three-A grade hospital as case study objective, of which displine characteristic is just TCM. This hospital pharmacy takes the responsibility of crude drugs supply and distribution of both out-patient and in-patient unit as well as boiling medicine service. When it comes to TCM crude drugs, its inventory management business is reprsentative.The objective of this study is to discover and find out possible change rules in demand of common TCM species through demand forecast to contain relative cost of inventory from origin. In light of the consumption amount of top 10 herbal slices in T hospital, outbound data could be exported from HIS. Hence time series analysis could be conducted after being pre-processed into dataset, which could be divided into training set and validation set. The classic model such as classic ARIMA and machine learning algorithm such as RNN-LSTM are applied in the analysis, which are both fit in training set and tested in validation set. Then, demand forecast, especially in short term could be made in this method. According to comparison in range of application, feasibility, modeling process and difficulty, machine learning method is recommended in this scenario of demand forecast overall.To tackle the intrinsic uncertainty in demand which could be beyond expectation, safety stock could be set in reasonable level according to the prediction result. Reordering point and maximum stock level could be calculated by replenishing stock strategy such as Max/Min model. Both direct and indirect cost of inventory could significantly be reduced by comparison between the scenario of optimized and previous record. A synergetic model integrating demand forecast and reordering strategy therefore could be referenced in TCM hospitals for inventory planning and optimization.
