Real-Time Optimization of Containers and Flatcars for Intermodal Operations

We propose a dynamic model for optimizing the ows of atcars that considers explicitly the broad range of complex constraints that govern the assignment of trailers and containers to a atcar. The problem is formulated as a logistics queueing network which can handle a wide range of equipment types and complex operating rules. The complexity of the problem prevents a practical implementation of a global network optimization model. Instead, we formulate a global model with the speci c goal of providing network information to local decision makers, regardless of whether they are using optimization models at the yard level. Thus, our approach should be relatively easy to implement given current rail operations. Initial experiments suggest that a atcar eet that is managed locally, without the bene t of our network information, can achieve the same demand coverage as a eet that is 10 percent smaller, but which is managed locally with our network information. We consider the problem of managing a eet of railroad atcars over a network in a real-time setting. A local terminal manager must determine how to assign a set of trailers and containers to the available atcars that can move this equipment. The challenge is that there are several dozen types of atcars, and many types of trailers and containers. Each atcar is best designed for moving a certain con guration of trailers and containers. For example, some atcars do not have the proper connections to hold a container. Other atcars might hold four 45 foot containers, but only two refrigerated 45 foot containers. Other atcars can only hold two containers, and it does not matter if they are refrigerated or not. The yard manager needs to nd groups of trailers and containers with the same destination that best utilize the hitch capacity of each atcar. The problem of assigning trailers and containers (or, more simply, boxes) to atcars at a yard to maximize hitch utilization is itself a challenging problem. It can be solved greedily or using an optimization procedure (see, for example, Feo & Gonzaliez-Velarde (1995)). However, both approaches are myopic and ignore the ability of the destination to properly utilize the equipment. For example, we may only have containers at a yard, but we have a atcar that can hold a trailer (a special hitch is needed to accommodate trailers). If the container is going to a destination that originates a lot of trailers, then we want to favor atcars that can carry trailers. In this paper, we focus on methods that contribute network information into local yard decisions. It is often the case that di erent atcars can be used to move a set of trailers and containers to a destination which, from the perspective of the local yard operation, appear equally e cient. For example, a yard manager might use a atcar that can move either trailers or containers to move two containers to their destination. That destination, however, may have little use for atcars that move trailers, whereas another yard desperately needs this equipment type. It is the goal of this paper to introduce network level information to improve the decisions made at a local level so that they better optimize an entire network. Railroads were the rst transportation industry to use optimization models to describe their networks and the ow of empty cars. However, these models were quite simple. The rst work that uses a space{time network is by White & Bomberault (1969) (see also Shan (1985), and the survey by Assad (1987)). Chih (1986) develops a multicommodity network ow model for railcar distribution that uses decomposition techniques to reduce the ows of multiple car types to single commodity network ows. Joborn (1995) presents a multicommodity network ow model that considers capacity restrictions on trains for repositioning of empty freight cars.