We report on a three year project that aimed at exploring the limits of stochastic and robust modelling in a large scale real live environment.
Our industrial partner operates warehouses at different locations within the same city where goods are picked to orders and delivered. Due to size restrictions of some of the warehouses their supply is largely stored at other locations and a shuttle service consisting of several trucks is used to bring in the necessary supply on time. The task is to find a schedule for the trucks and an assignment of pallets to these so that the probability of a shortage at any of the warehouses is minimized; time for optimization is limited to roughly twenty minutes. In our approach, time discretized flow models for the trucks and each type of article are coupled with linear constraints. A piecewise linear cost function, related to the expected number of pallets that have yet to be transported, models the stochastic inventory management aspects. We employ Lagrange decomposition in combination with a bundle method to solve the system. We present offline results on half a year of online data and report on first experiences with this approach when simulating the online environment with real data.
Last modified: Thu Nov 4 21:43:51 CEST 2004