A Python implementation of NSGA-II for the Flexible Job Shop Scheduling Problem with Automated Guided Vehicles (FJSP-AGVs), with two optimization objectives:

• Makespan
• Total energy consumption

This repository is intended as a lightweight research baseline. It includes the core evolutionary solver, benchmark instance parsers, schedule decoding, Pareto-front export, Pareto plotting, and complete Gantt-chart generation for the best-Makespan Pareto solution.

• Multi-objective optimization based on NSGA-II
• Three-part chromosome representation: operation sequence, machine assignment, and AGV assignment
• POX crossover for operation sequencing
• UX crossover for machine and AGV genes
• Mutation operators for operation, machine, and AGV decisions
• Fast non-dominated sorting and crowding-distance-based environmental selection
• Energy modeling for: machine processing, machine idle time, AGV loaded travel, and AGV empty travel
• Batch experiments with Pareto-history export and Pareto plots
• Complete Gantt-chart generation after optimization: machine processing, AGV empty travel, and AGV loaded travel

Example result folders:

• results/Bilge and Ulusoy/Jobset01_Layout1/
• results/Brandimarte_Data/Mk01_Layout/

Example Pareto plot:

Example Gantt chart:

NSGA-II-FJSP-AGVs/
|-- datasets/
|   `-- FJSP_AGVs/
|       |-- Bilge and Ulusoy/
|       `-- Brandimarte_Data/
|-- nsga_fjsp_agvs/
|   |-- NSGA_II.py
|   |-- binary_tournament_selection.py
|   |-- decoder.py
|   |-- environment_selection.py
|   |-- operators.py
|   |-- parser.py
|   `-- problem.py
|-- utils/
|   |-- performance_test.py
|   `-- recorder.py
|-- results/
|-- main.py
`-- README.md

• Python 3.10+
• Required package: numpy
• Optional packages: matplotlib, tqdm

Install dependencies with:

pip install numpy matplotlib tqdm

If you only want to run the optimizer without plotting, numpy is the only required dependency.

Run the default example:

python main.py

Current default configuration in main.py:

• Dataset: Bilge and Ulusoy
• Instance: Jobset01
• Layout: Layout1
• Stopping criterion: max_fe = 10000
• Population size: 100
• Crossover probability: 0.9
• Mutation probability: 0.15

Typical console output:

Stopping criterion: 10000 function evaluations
Final Pareto front (makespan, energy):
Solution 1: makespan = 73.00, energy = 5773.00
Solution 2: makespan = 74.00, energy = 5694.00
Solution 3: makespan = 75.00, energy = 5669.00
Gantt chart saved to: results/Bilge and Ulusoy/Jobset01_Layout1/best_makespan_gantt.png

The exact Pareto front may vary between runs if no random seed is fixed.

After running main.py, the solver will:

• Print the final Pareto front in the console
• Select the individual with the minimum Makespan from the final Pareto front
• Decode that individual again with detailed scheduling information
• Generate a complete 16:9 Gantt chart including: machine processing, AGV empty travel, and AGV loaded travel

Default output path:

results/<dataset>/<instance>_<layout>/best_makespan_gantt.png

To perform repeated runs, save Pareto history, export CSV files, and generate Pareto plots, run:

python utils/performance_test.py

This script can:

• Run a selected dataset multiple times
• Save Pareto-front history for each run
• Merge Pareto points across runs
• Export result tables as CSV files
• Generate scatter plots for Pareto fronts

Example generated files:

• all_pareto_points.csv
• final_pareto_front.csv
• pareto_fronts.png
• combined_pareto_front.png
• history/pareto_history_run_XX.csv

The repository currently includes two benchmark groups:

• datasets/FJSP_AGVs/Bilge and Ulusoy
• datasets/FJSP_AGVs/Brandimarte_Data

Each experiment uses:

• An instance file such as Jobset01.txt or Mk01.txt
• A transport-layout file such as Layout1.txt or Layout.txt

The solver optimizes:

1. Makespan
2. Total Energy

The total energy value includes:

• Machine processing energy
• Machine idle energy
• AGV empty-travel energy
• AGV loaded-travel energy

• nsga_fjsp_agvs/problem.py: problem definition, individual creation, initialization, crossover, and mutation
• nsga_fjsp_agvs/decoder.py: objective evaluation and detailed schedule decoding
• nsga_fjsp_agvs/NSGA_II.py: NSGA-II main loop
• nsga_fjsp_agvs/environment_selection.py: non-dominated sorting and crowding-distance selection
• nsga_fjsp_agvs/operators.py: initialization, crossover, and mutation operators
• utils/recorder.py: Pareto-front reporting and complete Gantt-chart plotting
• utils/performance_test.py: repeated experiments, statistics, CSV export, and Pareto plotting

You can directly modify parameters in main.py or utils/performance_test.py, for example:

• Dataset and layout selection
• max_fe
• Population size pop_size
• Crossover probability cr
• Mutation probability mu
• Number of repeated runs num_runs