Demand forecasting and capacitated lot - sizing problem under time-varying demand: A case study of schneider electric Vietnam Co., LTD

Abstract

For companies assembling products in general and companies specializing in high-tech electrical equipment lines in particular, ensuring that demand is forecasted accurately, and the assembly lines always operate smoothly is a prerequisite factor affecting the ability of responsiveness to their customers. The timely supply of necessary components to assemble each product line is also an indispensable daily responsibility. This study proposed an efficient combination of the Demand Forecasting (DF) model and the Capacitated Lot-sizing Problem (CLSP) under time – varying demand in the case of Schneider Electric Vietnam Co., Ltd (SEV). In the first stage, demand prediction was implemented based on the Long Short-term Memory (LSTM) method and Support Vector Machine (SVM) algorithm with the primary input data of historical sales data for a sample SKU. Next, the results of this stage will be validated to select the better method and obtain the final predicted demand. In the second stage, in addition to demand forecasting, the input data is supplemented with a Bill of Materials (BOM), inventory status data, ordering cost, holding cost, supply capacities, and minimum order quantity. The Exact method was utilized to solve the lot-sizing problem in terms of limited supply by using CPLEX software. The results of the research indicate that the performance of LSTM is superior to that of SVM no matter what metrics are represented in this case. Finally, the outputs obtained are optimal lot size, minimum total cost, and time to order each component to conduct an effective assembly.