| This dissertation considers methods for cooperative control problems for teams of mobile, autonomous servers with geographically diverse customers having stochastic arrival times. The problems are generally combinatorially complex; may feature uncertain state information and/or constraints on control commands, and require fast control decisions to be returned in a dynamic, on-line setting. In each of the three problem applications in this dissertation, the control command constraints and server dynamics can be represented by requiring the servers to move on graphs. These control problems are generally too complex to apply globally optimal solution methods in on-line applications, or for exact analysis. Instead, heuristic control methods are usually required, and systems are evaluated empirically via simulation.; The heuristic controllers developed for these three problems have in common a receding horizon method, which determines control commands by solving an optimization problem predicting the value of various states that could be reached at the end of a limited window of time into the future. The optimization problem is re-solved to determine new control commands as the system state changes. The receding horizon controllers are designed to engender cooperation among the servers, and are specialized to achieve good performance based on quality-of-service criteria specified for each application. Two applications, the graph traveling repairman problem and the truckload pick-up and delivery problem, are types of vehicle routing problems. The third application is the elevator dispatching problem, which is difficult enough that previous heuristic control methods are generally specialized for a particular archetypal passenger arrival pattern.; In addition to the controllers, a graph-based model of the elevator system has been developed in order to account for the system dynamics. This model can predict travel times to various floors for the elevator cars, and captures the state of passenger queues. This dissertation also presents simulators for each of the three control problems. The simulators allow the empirical evaluation of these complex systems with various controllers. The receding horizon controllers perform well in simulations, and in the elevator dispatching problem the receding horizon controller can outperform some other heuristic controllers in all of the archetypal passenger arrival patterns. |
