The Unreliable Job Selection and Sequencing Problem

TL;DR

This paper investigates a stochastic scheduling problem where jobs are selected and sequenced on a machine prone to failure, aiming to maximize expected profit, and introduces algorithms for solving it efficiently.

Contribution

It formulates the Unreliable Job Selection and Sequencing Problem, analyzes its complexity, and proposes novel algorithms for optimal solutions.

Findings

The problem is NP-hard in general.

Polynomial-time solutions exist for special cases.

Proposed algorithms efficiently solve large instances.

Abstract

We study a stochastic single-machine scheduling problem, denoted the Unreliable Job Selection and Sequencing Problem (UJSSP). Given a set of jobs, a subset must be selected for processing on a single machine that is subject to failure. Each job incurs a cost if selected and yields a reward upon successful completion. A job is completed successfully only if the machine does not fail before or during its execution, with job-specific probabilities of success. The objective is to determine an optimal subset and sequence of jobs to maximize the expected net profit. We analyze the computational complexity of UJSSP and prove that it is NP-hard in the general case. The relationship of UJSSP with other submodular selection problems is discussed, showing that the special cases in which all jobs have the same cost or the same failure probability can be solved in polynomial time. To compute optimal solutions, we propose a compact mixed-integer linear programming formulation, a dynamic programming algorithm, and two novel stepwise exact algorithms. We demonstrate that our methods are capable of efficiently solving large instances by means of extensive computational experiments. We further show the broader applicability of our stepwise algorithms by solving instances derived from the Product Partition Problem.