An iterative exact algorithm for the weighted fair sequences problem

TL;DR

This paper introduces an iterative exact algorithm for the NP-hard weighted fair sequences problem, significantly improving solution speed and success rate over previous methods through a new mixed-integer programming model and valid inequalities.

Contribution

The paper presents a novel iterative exact solution algorithm with enhanced mixed-integer programming models and valid inequalities for the weighted fair sequences problem, outperforming existing approaches.

Findings

Solved 404 of 440 instances to optimality within five minutes

Outperformed previous methods by solving more instances faster

Achieved up to two orders of magnitude speedup

Abstract

In this work, we present a new iterative exact solution algorithm for the weighted fair sequences problem, which is a recently introduced NP-hard sequencing problem with applications in diverse areas such as TV advertisement scheduling, periodic machine maintenance and real-time scheduling. In the problem we are given an upper bound on the allowed solution sequence length and a list of symbols. For each symbols, there is a positive weight and a number, which gives the minimum times the symbol has to occur in a feasible solution sequence. The goal is to find a feasible sequence, which minimizes the maximum weight-distance product, which is calculated for each consecutive appearance of each symbol in the sequence, including the last and first appearance in the sequence, i.e., the sequence is considered to be circular for the calculation of the objective function. Our proposed solution algorithm is based on a new mixed-integer programming model for the problem for a fixed sequence length. The model is enhanced with valid inequalities and variable fixings. We also develop an extended model, which allows the definition of an additional set of valid inequalities and present additional results which can allow us to skip the solution of the mixed-integer program for some sequence lengths. We conduct a computational study on the instances from literature to assess the efficiency of our newly proposed solution approach. Our approach manages to solve 404 of 440 instances to optimality within the given timelimit, most of them within five minutes. The previous best existing solution approach for the problem only managed to solve 229 of these instances, and its exactness depends on an unproven conjecture. Moreover, our approach is up to two magnitudes faster compared to this best existing solution approach.