Nonlinear model predictive control for resource allocation in the management of intermodal container terminals

Nonlinear model predictive control is proposed to allocate the available transfer resources in the management of container terminals by minimizing a performance cost function that measures the lay times of carriers over a forward horizon. Such an approach to predictive control is based on a model of the container flows inside a terminal as a system of queues. Binary variables are included into the model to represent the events of departure or stay of a carrier, thus the proposed approach requires the on-line solution of a mixed-integer nonlinear programming problem. Different techniques for solving such problem are considered that account for the presence of binary variables as well as nonlinearities into the model and cost function. The first relies on the application of a standard branch-and-bound algorithm. The second is based on the idea of dealing with the decisions associated with the binary variables as step functions. In this case, real nonlinear programming techniques are used to find a solution. Finally, a third approach is proposed that is based on the idea of approximating off line the feedback control law that results from the application of the second one. The approximation is made using a neural network that allows to construct an approximate suboptimal feedback control law by optimizing the neural weights. Simulation results are reported to compare such methodologies