Using artificial neural network as a meta-modeling technique in supply chains

This paper aims to develop a framework for inventory levels estimation in multi-level supply chains. The framework utilizes artificial neural network (ANN) as a meta-modeling technique for discrete event simulation (DES) to estimate the on-hand inventory levels. A DES models the relationships between the supply chain levels. Then the output of the simulation model is analyzed by an ANN model that estimates the on-hand inventories. In this paper, we assume a four-level supply chain (retailer (R), distributor (DC), manufacturing plant (MP), and supplier (S)), in which each level employs a periodic review inventory system (T, S). Two decision variables are considered: inventory target (S) and time between orders (T). The manufacturer capacity is taken into account as well. A simulation experiment is conducted by varying the values of T and S for the R, DC, and MP, and the manufacturing capacity of the MP. The simulation results are then analyzed using an ANN model. The mean absolute difference between the simulation output and the ANN model (mean absolute error (MAE)) is used to evaluate the ANN model. The results showed that the ANN is a powerful meta-modeling technique for analyzing simulation outputs, since the MAE is less than 6%.