Dynamic control of multicommodity fleet management problems

Dynamic fleet management problems with multiple equipment types and limited substitution can be modeled as dynamic, multicommodity network flow problems. These problems are further complicated by the presence of time windows on task arcs (a task, or load, can be handled at different points in time) and the need for integer solutions. In this paper, we formulate the problem as a dynamic control problem, and show that we can produce solutions within four to five percent of a linear relaxation. In addition, we can solve the ultra-large problems that arise in certain applications; these problems are beyond the capabilities of state-of-the-art linear programming solvers.