Optimizing product deliveries within supply chain networks can present a unique set of challenges. Factors such as travel distance and order wait times need to be taken into consideration while keeping profits high. When many manufacturing centers and distributrs are spread out over a large geographic area, determing the best choices for routing products can become even more difficult.
Working with the Applied Intelligence team at Accenture, we modified a product delivery model available on AnyLogic Cloud to demonstrate how Pathmind reinforcement learning can be used to quickly and effectively optimize supply chain simulations.
The supply chain model features a network of three manufacturing centers and fifteen distributors across a map of Europe. The distributors order a random number of goods every 1 to 2 days.
An obvious heuristic such as using the manufacter that is closest to the distribuor that places the order is not the optimal solution. Partial orders are not sent out, so the manufacturing center will wait until it has produced the extra inventory needed to fulfill an order if it does not have enough on hand. That extra time need for production might mean that a manufacturing center further away but with more inventory could complete the order faster.
Taking those factors into consideration, figuring out the best manufacturing center to complete an order becomes more of a challenge. The nearest manufactuer heursitc, as well as random actions, do not produce good results.
Why Pathmind Was Needed
Since Pathmind’s reinforcement learning is better equipped to handle the complexities of the supply chain network, it offers a better solution for making delivery decisions. The AI can also optimize for whichever factor is most important to a supply chain manager, such as lowering wait times or keeping travel distance at a minimum.
The AI policy generated by Pathmind outperformed both the nearest manufacturing center heuristic, as well as random actions. Since Pathmind determined the best solution, there was no need for hours spent manually testing possible alternatives. This success demonstrates how supply chain managers can rely on reinforcement learning to get more from their simulations and discover new paths to efficiency in areas such as delivery routing and beyond.