Genetic (Evolutionary) algorithm for process scheduling

This repository is an implementation of modified genetic algorithm, which we call Evolutionary Algorithm. It is adapted to solve P||C_max problem. Besides the implementation of the algorithm, this repo contains utlilities that help test it, such as benchmark generator or simple greedy algorithm

How it works

The algorithm stores line-ups as dynamic arrays of vectors. Each vector in an array represents one processor which contains a total execution time on the processor (first element of every vector) and duration of each process within that processor. The statring line-up is random. The algorithm starts from that data and creates n generations, each consisting of a population of m line-ups (n and m are parameters, which can be adjusted to optimize time or quality of the solution).

New line-ups in a generation are built in a following way based on the best line-up found in the process so far:

1. Processors with the longest and the shortest execution times are searched within the array
2. A random process from the longer processor is put into the shorter processor
3. This process is repeated x times (x is a parameter, which can be adjusted)

While creating the population, algorithm only memorizes the current line-up and the best line-up so far (which is determined by the P||C_max). Once the population is done building, we have the best line-up from that population and if it is better than the base line-up, the new line-up is a base for the next generation.

Pseudocode

1. For size of population:

    1.1. For number of processes:

        1.1.1 Assign process to random processor
    
        1.1.2 Add to execution time of this processor duration of current process

2. For number of generations:

    2.1. For size of population:

        2.1.1. Best new line up is base line-up

        2.1.2. For *x* times (*x* is a parameter, which can be adjusted):
    
            2.1.1.1. Find max and min execution time for every processor

            2.1.1.2. From processor with max execution time and transfer random process to min processor

        2.1.3. If new line-up is better than best new line-up:

            2.1.2.1. Best new line-up is now new line-up

    2.2. If best line-up from new population is better than base line-up:
        
        2.2.1. Base line-up is best new line-up from new population

Results

Graph comparing the performance of the evolutionary algorithm versus the greedy approach and the optimal values. Instances were generated by benchmarkGen.py

Percentage performence comparison.

Authors

• Grzegorz Płaczek (148071)
• Kamil Kałużny (148121)
• Olga Gerlich (148088)