This talk will describe a project for a utility company that wants to improve its repair dispatching operations. Customer calls for service sometimes require immediate response or responses within time windows. Yet these calls are very difficult to predict. The historical pattern of demands explains monthly patterns pretty well, but the unpredictable day-to-day variations can be quite large. Given the uncertainty in the prediction, and the fact that the workforce has many skill levels--including constraints on job matching--it is necessary to simulate the scheduling or dispatch of the workers to the predicted jobs.
We describe a stochastic programming formulation of the scheduling problem that attempts to do an optimal assignment in the face of uncertainty. In deciding how many service technicians to staff, we consider a probabilistic range of potential demands, and choose the level of staffing that minimizes total expected cost with uncertainties covered by overtime.
|