Solve-and-robustify: Synthesizing partial order schedules by chaining - Robotics Institute Carnegie Mellon University

Solve-and-robustify: Synthesizing partial order schedules by chaining

Nicola Policella, Amedeo Cesta, Angelo Oddi, and Stephen Smith
Journal Article, Journal of Scheduling: Special Issue on AI Planning and Scheduling, Vol. 12, No. 3, June, 2009

Abstract

Goal separation is often a fruitful approach when solving complex problems. It provides a way to focus on relevant aspects in a stepwise fashion and hence bound the problem solving scope along a specific direction at any point. This work applies goal separation to the problem of synthesizing robust schedules. The problem is addressed by separating the phase of problem solution, which may pursue a standard optimization criterion (e.g., minimal makespan), from a subsequent phase of solution robustification in which a more flexible set of solutions is obtained and compactly represented through a temporal graph, called a Partial Order Schedule (POS). The key advantage of a POS is that it provides the capability to promptly respond to temporal changes (e.g., activity duration changes or activity start-time delays) and to hedge against further changes (e.g., new activities to perform or unexpected variations in resource capacity). On the one hand, the paper focuses on specific heuristic algorithms for synthesis of POSs, starting from a pre-existing schedule (hence the name Solve-and-Robustify). Different extensions of a technique called chaining, which progressively introduces temporal flexibility into the representation of the solution, are introduced and evaluated. These extensions follow from the fact that in multi-capacitated resource settings more than one POS can be derived from a specific fixed-times solution via chaining, and carry out a search for the most robust alternative. On the other hand, an additional analysis is performed to investigate the performance gain possible by further broadening the search process to consider multiple initial seed solutions. A detailed experimental analysis using state-of-the-art RCPSP/max benchmarks is carried out to demonstrate the performance advantage of these more sophisticated solve and robustify procedures, corroborating prior results obtained on smaller problems and also indicating how this leverage increases as problem size is increased.

BibTeX

@article{Policella-2009-10224,
author = {Nicola Policella and Amedeo Cesta and Angelo Oddi and Stephen Smith},
title = {Solve-and-robustify: Synthesizing partial order schedules by chaining},
journal = {Journal of Scheduling: Special Issue on AI Planning and Scheduling},
year = {2009},
month = {June},
volume = {12},
number = {3},
keywords = {Iterative improvement techniques, Scheduling under uncertainty, Constraint-based scheduling},
}