Interval Transportation Problem: Feasibility, Optimality and the Worst Optimal Value

TL;DR

This paper studies an interval transportation problem under data uncertainty, providing conditions for feasibility and optimality, and introduces an exact method for computing the worst optimal value, supported by computational experiments.

Contribution

It offers polynomial-time conditions for weak and strong optimality testing and an exact approach for the NP-hard worst optimal value computation in interval transportation problems.

Findings

Polynomial-time testing for weak optimality.

Exact method for worst optimal value computation.

Competitive performance in computational experiments.

Abstract

We consider the model of a transportation problem with the objective of finding a minimum-cost transportation plan for shipping a given commodity from a set of supply centers to the customers. Since the exact values of supply and demand and the exact transportation costs are not always available for real-world problems, we adopt the approach of interval programming to represent such uncertainty, resulting in the model of an interval transportation problem. The interval model assumes that lower and upper bounds on the data are given and the values can be independently perturbed within these bounds. In this paper, we provide an overview of conditions for checking basic properties of the interval transportation problems commonly studied in interval programming, such as weak and strong feasibility or optimality. We derive a condition for testing weak optimality of a solution in polynomial time by finding a suitable scenario of the problem. Further, we formulate a similar condition for testing strong optimality of a solution for transportation problems with interval supply and demand (and exact costs). Moreover, we also survey the results on computing the best and the worst optimal value. We build on an exact method for solving the NP-hard problem of computing the worst (finite) optimal value of the interval transportation problem based on a decomposition of the optimal solution set by complementary slackness. Finally, we conduct computational experiments to show that the method can be competitive with the state-of-the-art heuristic algorithms.