Project Sponsored by:

1
Hyper-Heuristic Approaches to a Dynamic Workforce
Scheduling Problem
Project Sponsored by
Stephen Remde, Keshav Dahal, Peter
Cowling, MOSAIC Research Group
2
Overview

• Introduction
• Problem Description
• Hyper-heuristic Approaches
• Experimental Results
• Conclusions
• Future Work
• Question

3
Introduction

• Dynamic Workforce Scheduling
• Generalisation of Resource Constrained Project
  Scheduling Problem
• Scheduling when activities/tasks should start and
  which resource should do them
• RCPSP
• Much work in literature on the RCPSP
• NP-Hard
• Differences
• Notion of Skill/Location

4
The Problem

• Skills
• Tasks
• Start/End Locations
• Skill requirements
• Time windows
• Priority
• Resources
• Home location
• Skill competencies
• Time windows
• Travel speeds

5
Hyper-Heuristic Approaches

• Heuristics generally tailored to a specific
  problem
• Hyper-heuristics can work with any problem given
  the right set of Low Level Heuristics
• To test our insert task low level heuristic we
  used a local search
• Method tested against a Genetic Algorithm

6
Test Problem

• 20 different skills
• 200 resources
• Possessing between 1 and 15 skills
• 1000 tasks
• Requiring between 1 and 4 skills
• Over a 3 day period

7
Experimental Results
8
Experimental Results
9
Conclusions

• LS heuristics produce fitter solutions faster
  than the GA method tested against
• Over 80 increase in Fitness
• Time only a fraction of the GA
• LS produces solutions with less travel
• Due to the fact travel is not directly considered
  by the GA

10
Future Work

• Restricting Sets of Resources
• Low Level Heuristics
• Task Selectors
• Resource Selectors
• New Hyper-heuristics

11
Questions?