基于SARIMA-LSTM的零售生鲜品库存需求预测

Retail Raw Goods Inventory Demand Forecast Based on SARIMA-LSTM

针对零售业的生鲜类商品存在易损耗、货架周期短、明显的季节性等特点,提出一种新的组合预测模型。采用SARIMA-LSTM组合模型综合供应商、零售商、用户三方面考虑需求影响因素如准时交货率、综合成本、气温状况、销售金额等,并结合贝叶斯优化算法对LSTM进行超参数选择,将SARIMA的模型预测值和实际值间的误差序列进行修正并得到预测值。实验证明,考虑需求影响因素的组合预测模型比单一传统统计方法的预测精度要高,并对以后零售业库存管理有一定的建设意义。

Aiming at the characteristics of fresh goods in retail industry, such as easy wear and tear, short shelf cycle and obvious seasonality, a new combination forecasting model was proposed. The SARIMA-LSTM portfolio model integrated suppliers, retailers and users demand three aspects to consider factors such as on-time delivery rate, comprehensive cost, the temperature condition, the sales amount, etc., and combined with bayesian optimization algorithm for super LSTM parameter selection, the SARIMA model error between the predicted value and actual value sequence modification and predictive value. The experimental results show that the combined forecasting model considering the demand influencing factors has higher forecasting accuracy than the single traditional statistical method, and has a certain construction significance for the future retail inventory management.