Applying Evolution Strategies to a University Timetabling System

1
Applying Evolution Strategies to a University
Timetabling System

• LAWRENCE TECHNOLOGICAL UNIVERSITY
• Thomas George
• Chanjin Chung

2
The Problem

• Large Number of Possible Courses
• Limited Resources
• Rooms
• Professor Availability
• Student Constraints
• Degree Requirements
• Availability
• Personal Preferences

3
Curriculum Input Data

• Allowable Combinations of Periods
• Number of Class Rooms
• Course ID and Wildcard
• or
• Course ID and Available Periods

4
Student Constraints

• Student ID
• Course ID
• Available Periods
• Maximum Number of Courses is a System Limit (no
  input)

5
Objective

• Maximize Enrollment
• Promotes Highest Revenue
• Increases Graduation Rate
• Maximize Selection of Student Preferences
• Allows Students to take required courses
• Improves Student Satisfaction
• Maximize Class Size
• Efficient Use of Resources

6
Sample Output

• Total Enrollment
• Course ID
• Course Schedule
• Course Enrollment
• Status (lt 5 results in cancellation)

7
Crossover
8
Crossover
9
Crossover
10
Crossover
11
Mutation

• Probability of Mutation selected in interface.
• Each course in schedule may be mutated
• Mutation based on valid available schedules

12
Mutation

• Given that a course is to be mutated
• Randomly select from list of available schedules
• Available schedules determined by hard
  constraints such as required lecture hours

13
Mutation
14
PARAMETERS

• Population Size
• Crossover Rule
• Mutation Delta
• 1st Mutation Rate
• Window Size

15
Algorithm
16
Demonstration
17
Future Work

• Currently Implementing Web Based System to
  Capture Actual Student Data
• Adding the ability to Process Additional
  Constraints in the Implementation
• Evaluating Impact of using Different Pseudo
  Number Generators on Data

18
Why an Evolution Strategy?

• The initial focus of our project involved study
  of evolutionary methods
• Use of Strings in Java
• Faster than Binary Manipulation in Platform
• Human Readable for Debugging
• Simplifies implementation of Hard Constraints
• Note Adjusting mutation rates and crossover
• method combines some of GAs advantages with ES