Control of Dual-Sourcing Inventory Systems Using Recurrent Neural Networks

A key challenge in inventory management is to identify policies that optimally replenish from multiple suppliers. To solve such optimization problems, managers need decide what quantities order each supplier, given the net and outstanding orders, so expected backlogging, holding, sourcing costs are jointly minimized. Inventory problems have been studied extensively for over 60 years, yet even basic dual-sourcing which orders an expensive supplier arrive faster than a regular remain intractable their general form. In addition, there emerging develop proactive, scalable algorithms can adjust recommendations dynamic demand shifts timely fashion. this work, we approach dual neural network--based lens incorporate information on dynamics its replenishment (i.e., control) into design of recurrent networks. We show proposed network controllers (NNCs) able learn near-optimal commonly used instances within few minutes CPU time personal computer. demonstrate versatility NNCs, also they control with empirical, non-stationary distributions challenging tackle effectively using alternative, state-of-the-art approaches. Our work shows high-quality solutions complex be obtained deep neural-network approaches directly account process. As such, our research opens up new ways efficiently managing complex, high-dimensional dynamics.