Optimal route selection in complex multi-stage supply chain networks using SARSA(λ)

Success of any supply chain highly depends on the appropriate use of transportation. It is always a tough decision to select an optimal route in the stages of a supply chain that will balance the reverse condition of cost and time. Q-learning is already getting implied to optimize routes, but it greatly lacks in designing a complete Markov Decision Process (MDP) with enough number of states and actions sets. In this paper, we propose an MDP that can fit the design of a proper multi-stage supply chain network and intend to solve the MDP with SARSA (λ) algorithm. The algorithm was simulated in a java based environment and experimental results show that Q-learning produced higher chunks of rewards but SARSA (λ) was faster in case of convergence speed.