有限时间数据下的零售药店药品销量预测研究

A Research on the Forecast of Drug Sales in Retail Pharmacies with Limited-Time Data

针对开业不久的零售药店及刚上市的新药缺乏长时间销售数据,难以进行AMIMA等时间序列分析,以及难以穷尽所有影响因素以进行因果预测的困难,提出了一种药品销量复合预测模型。该模型由三个部分组成:指数平滑,用于获取销量趋势;分类主成分分析,用于减少冗余信息;后向传播前馈神经网络,用于回归预测。利用AB药房连锁有限公司24个药店的三九感冒灵、51个药店的江中健胃消食片和49个药店的苯磺酸左旋氨氯地平片两年半的销量数据,对半年的销量以月为单位分别进行了预测验证,结果显示:三种药品的预测值与真实值相关系数分别达到0.77、0.84、0.85,标准化均方误差分别为0.48、0.37、0.30,从而说明了模型的有效性。

For newly opened retail pharmacies and new drugs,it is difficult to conduct time series analysis using ARIMA model due to the lack of long-term sales data.It is also difficult to carry out causal prediction because it is hard to take all influencing factors into account.In order to solve these problems,this paper proposes a compound forecasting model for drug sales.The model consists of three parts.The first part is to obtain sales trend via exponential smoothing.The second part is to reduce redundant information through principal component analysis by category.The third part is to forecast sales applying backward propagation feedforward neural network.This paper predicts half-year drug sales of a kind of cold medicines in 24 pharmacies,a kind of stomach medicines in 51 pharmacies and a kind of antihypertensive medicines in 49 pharmacies according to the two-and-a-half-year sales data of AB Company on a monthly basis,respectively.The results show that the correlation coefficients between the predicted values and real values are 0.77,0.84,and 0.85,and the standardized mean square errors are 0.48,0.37,and 0.30.This indicates that it is an effective model.On the one hand,this model will help retail pharmacies to formulate appropriate inventory;on the other hand,it will also help other node companies within the pharmaceutical supply chain to optimize their own decisions accordingly.