Simulation based optimization of a stochastic supply chain considering supplier disruption: An agent-based modeling and reinforcement learning

Abstract

Many researchers and practitioners in the recent years have been attracted to investigate the role of uncertainties in the supply chain management concept. In this paper a multi-period stochastic supply chain with demand uncertainty and supplier disruption is modeled. In the model, two types of retailers including risk sensitive and risk neutral, with many capacitated suppliers are considered. Autonomous retailers have three choices to satisfy demands: ordering from primary suppliers, reserved suppliers and spot market. The goal is to find the best behavior of the risk sensitive retailer, regarding the forward and option contracts, during several contract periods based on the profit function. Hence, an agent-based simulation approach has been developed to simulate the supply chain and transactions between retailers and unreliable suppliers. In addition, a Q-learning approach (as a method of reinforcement learning) has been developed to optimize the simulation procedure. Furthermore, different configurations for simulation procedure are analyzed. The R-netlogo package is used to implement the algorithm. Also a numerical example has been solved using the proposed simulation-optimization approach. Several sensitivity analyzes are conducted regarding different parameters of the model. Comparison of the numerical results with a genetic algorithm shows a significant efficiency of the proposed Q-leaning approach.