This repository contains the metaheuristic optimization engine for solving the
TV Advertisement Scheduling Problem, implemented using:

• NSGA-II (multi-objective optimization)
• Genetic Algorithm
• Simulated Annealing (SA)
• Tabu Search (TS)

The goal is to allocate commercials to TV breaks while maximizing profit and minimizing operational penalties, under complex real-world constraints (time windows, budgets, min/max plays, reach requirements, competitor exclusion, break capacity, sequencing rules, etc.).

This work was part of a Master's thesis in Industrial Engineering focusing on intelligent scheduling in broadcast environments.

• NSGA-II with custom sampling, selection, mutation, crossover
• Genetic Algorithm (pymoo-based implementation)
• Simulated Annealing with cooling and neighbor generation
• Tabu Search with tabu list, intensification/diversification

• Custom commercial allocation structure using NumPy structured arrays
• Automatic start/end time calculation
• Repair mechanisms for feasibility
• Built-in dominance sorting & evaluation

Examples:

• No overlapping commercials
• Budget feasibility
• Min/max number of plays
• Reach satisfaction
• No competitor advertisers in the same break
• Release/due time enforcement
• Break capacity respect
• Sequencing consistency

Taguchi-based experiment automation for:

• NSGA-II
• Tabu Search
• Simulated Annealing

• Initial population generator
• Custom dominance sorting
• Constraint checker
• Solution pretty-printing

The TV-ad scheduling problem involves assigning commercials to predefined breaks under constraints such as:

• Commercial budgets
• Required reach
• Release and due times
• Minimum/maximum number of allowed plays
• Break duration limits
• Conflict avoidance (competitors cannot appear in the same break)
• Start/end time sequencing consistency

Each solution represents a full assignment of commercials across all breaks, encoded in a NumPy structured array.

The objective functions:

• Maximize Profit
• Minimize Cost
• Constraints are handled using penalization and repair operators.

• Maximize: Profit – Cost

git clone https://github.com/aliz-f/tv-ad-scheduling-metaheuristics
cd tv-ad-scheduling-metaheuristics

pip install -r requirements.txt

Dependencies include:

• pymoo
• numpy
• loguru
• click

python main.py -d data/sample.json

python genetic_algorithm/genetic_main.py -d data/sample.json

python main.py -d data/sample.json --taguchi sa

python main.py -d data/sample.json --taguchi tabu

Each input JSON file must include:

{
  "commercials": 5,
  "breaks": 10,
  "commercial_duration": [30, 30, 15, 45, 20],
  "commercial_copy": [{"min":1,"max":5}, ...],
  "break_duration": [120, 140, ...],
  "break_length": [{"start": "...", "end": "..."}, ...],
  "price": [[...], [...]],
  "reach": [[...], [...]],
  "budget": [...],
  "required_reach": [...],
  "penalty": [...],
  "due_time": [...],
  "release_time": [...],
  "competitors": [["A","B"], ...],
  "name": [...]
}

Each break has positions, and each position is stored as:

(start_time, end_time, advertiser_obj, copy_index, reserved, price, reach)

The solution supports:

• calculate_profit()
• calculate_cost()
• calculate_cost2() debugging version
• filtered_list (non-empty slots)
• break_free_space
• min_max_played
• used_budget
• obtained_reach
• Automatic time recalculation (reconfig_times())

The Constraints class contains 12 feasibility checks.

Examples:

Repairs are performed via the Reformations class.

Run full Taguchi experiments:

python main.py -d sample.json -t nsga

python main.py -d sample.json -t tabu

python main.py -d sample.json -t sa

Each test produces results under:

result/NSGA/
result/TABU/
result/SA/

Algorithms print:

• Best or final solution (schedule)
• Profit, cost, and final objective
• Constraint violations
• Execution time
• Parameter settings

NSGA-II prints all Pareto-front candidates, sorted by dominance.

This MILP model is one of three main components of the TV scheduling thesis project:

1. Data Generator

   • Generates synthetic instances (breaks, commercials, budgets, reach, etc.)
   • Exports JSON for NSGA-II/metaheuristics and for CPLEX (scaled).

2. MILP Model

   • Reads CPLEX-ready JSON.
   • Solves the scheduling problem exactly (within MIP gap limits).

3. Metaheuristic Algorithms (this repo)

   • NSGA-II, simulated annealing, tabu search.
   • Use the same data to compare against MILP performance.