https://github.com/lucadiba/timetabling-problem-solver

A C++ metaheuristics solver for the ETP (examination timetabling problem)
https://github.com/lucadiba/timetabling-problem-solver

Last synced: 10 days ago
JSON representation

A C++ metaheuristics solver for the ETP (examination timetabling problem)

• Host: GitHub
• URL: https://github.com/lucadiba/timetabling-problem-solver
• Owner: LucaDiba
• License: mit
• Created: 2020-01-14T13:29:08.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2020-01-17T23:43:23.000Z (almost 5 years ago)
• Last Synced: 2024-11-08T18:19:14.753Z (2 months ago)
• Language: C++
• Homepage:
• Size: 106 KB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Examination Timetabling Problem Solver

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

A C++ metaheuristics solver for the ETP (examination timetabling problem).

# Data representation

- Vector of exams: `v[i]` is the timeslot assigned to exam `i`

- Vector of timeslots: `v[t]` is a list of exams scheduled on timeslot `t`

# Metaheuristic

## 1. Multi-start: Genetic Algorithm

Solution → chromosome  

Exam (element of the solution) → gene

#### Initial population

- Population size: max(#exams * 0.1, 10)

- Population: random chromosomes (feasible solution):

  1. exams are sorted by number of conflicts

  2. exams are placed in random timeslots

  3. restart from scratch when stuck

 

#### Evolution

1. Sorting chromosomes according to the fitness score (`1/penalty`)

2. Offspring generation

  - Elite Chromosomes: 20%

  - Crossover (adapted for feasibility): 30%

  - Mutation: 30%

  - Inversion (adapted for feasibility): 20%

#### Stopping condition

(used > 80% of total time) OR ( (used > 50% of t.t.) AND (no improving solutions from more than 8% of t.t) )

## 2. Neighborhood Search: Simulated Annealing

#### Initialization

- Initial solution: the best one found by GA

- Initial temperature: `(F(x_AVG) - F(x*)) / ln(0.5)`

- Cooling rate: `0.99`

- Plateau: 10 * `n_exams` * `n_timeslots`

#### Neighbour

1. Pick a random exam

2. Pick a random timeslot (different from current one)

3. Move the exam to that timeslot if is feasible to do it

4. Move it if `rand(0, 1) < p`, where `p` is a function of the candidate and the current penalties

# Posible improvements

- Use more C++ syntax instead of C's one

- Multi-thread solution generation (and the entire problem itself)

- Incremental penaly computation when moving examx