| Demand forecasting and inventory decisions are two essential components of retail operations.Accurate demand forecasting forms the basis for effective inventory decisions,as only with precise forecasting can retailers create better inventory plans.Moreover,effective inventory decisions can reduce warehouse occupancy,minimize total inventory costs,and enhance capital flow for retailers.In this thesis,we take a Shandong retailer as the research object.By investigating the current situation of the retailer’s operation,we find that the retailer relies too much on manual experience in merchandise demand forecasting and inventory decision making,which leads to many problems in merchandise inventory management due to the huge errors in merchandise demand forecasting.Therefore,this thesis aims to improve the accuracy of retail merchandise demand forecasting,and make innovative improvements to address the deficiencies of missing value filling and multidimensional forecasting.As a result,the forecast accuracy is effectively improved and the inventory cost of enterprises is reduced.We first propose a method for filling missing values in merchandise sales data using timeseries regular matrix decomposition in data processing.This method outperforms traditional missing value filling algorithms.Second,we utilize the temporal convolutional network algorithm to tackle the challenge of multidimensional forecasting for retailers with a large number of Stockkeeping Units(SKUs),allowing for multi-item forecasting with a single model.Finally,based on the results of merchandise demand forecasting,we propose different inventory decisions for different products by classifying the large number of products.By conducting experiments on the historical sales data of this enterprise,the results of this thesis show that the method proposed in this thesis has higher prediction accuracy compared with traditional prediction models. |
