MOOPPS: Multi-Objective Production Planning

• MOOPPS System is a decision‐support tool for optimizing multi-objective production scheduling in job-shop and flow-shop settings.
• It integrates a suite of metaheuristic and heuristic algorithms with a flexible, menu-driven GUI to generate and visualize Pareto-optimal schedules.
• The framework is validated on benchmark problems, offering customizable modules and interactive decision support to refine production planning outcomes.

MOOPPS (Multi-Objective Optimization Production Planning System) is a decision-support and optimization system designed for tackling multi-objective production scheduling problems, particularly in job-shop and flow-shop models. MOOPPS implements a suite of metaheuristic and heuristic algorithms with a focus on identifying or approximating the Pareto-optimal set of schedules subject to multiple conflicting objectives. Distinguished by its menu-driven graphical user interface, flexible method database, and interactive decision-support features, MOOPPS enables both automated optimization and interactive schedule selection, as recognized by its European Academic Software Award in 2002 (0809.0961).

1. Formal Problem Definition

MOOPPS targets classical production scheduling environments such as the job-shop and flow-shop problems. The system represents each instance as follows:

• Jobs and Operations: Let $J = \{J_1, \dots, J_n\}$ be the set of jobs. Each job $J_j$ is a sequence of operations $O_{j1}, \dots, O_{j,o_j}$, each assigned to a single machine from the set $M = \{M_1, \dots, M_m\}$.
• Processing Times and Constraints: Processing times $p_{jk} \geq 0$, job release dates $r_j$, and optional due dates $d_j$ are specified. Each operation $O_{jk}$ is assigned a start time $s_{jk}$ subject to release-date, precedence, and capacity constraints.
• Feasible Schedule: The feasible space $\Omega$ comprises all schedules $J_j$0 respecting:
  1. $J_j$1
  2. $J_j$2, for all $J_j$3, $J_j$4
  3. No overlap: for any two operations on the same machine, their assigned intervals do not overlap.

• Multi-Objective Formulation: The objective vector is $J_j$5 with typical objectives including:

  • Makespan $J_j$6
  • Total completion time $J_j$7
  • Maximum tardiness $J_j$8, $J_j$9
  • Number of tardy jobs $O_{j1}, \dots, O_{j,o_j}$0
  • A solution $O_{j1}, \dots, O_{j,o_j}$1 dominates $O_{j1}, \dots, O_{j,o_j}$2 if $O_{j1}, \dots, O_{j,o_j}$3 for all $O_{j1}, \dots, O_{j,o_j}$4 and $O_{j1}, \dots, O_{j,o_j}$5. The Pareto set $O_{j1}, \dots, O_{j,o_j}$6 contains nondominated solutions (0809.0961).

2. System Architecture and User Interface

MOOPPS adopts a three-tier modular design integrating:

• Model Instance Database: Stores scheduling instances (user-defined or imported from standard repositories).
• Method Database: Hosts the metaheuristics, heuristics, and selection routines.
• Graphical User Interface (GUI):
  • Exposes full menu-driven parameterization and instance selection.
  • Supports parameter entry for methods (population sizes, cooling rates, neighborhoods, iteration limits).
  • Offers "optimization" mode (solution generation) and "interactive decision-making" mode (final selection).
  • Features outcome-space scatter plots and Gantt chart visualization of schedules (0809.0961).

A typical workflow involves selecting problem instances, configuring methods, executing runs, inspecting Pareto-approximate sets ($O_{j1}, \dots, O_{j,o_j}$7), and interacting with solution visualizations.

3. Metaheuristic Algorithms and Method Database

MOOPPS integrates the following algorithmic modules:

Method	Key Features	Customizable Parameters
Priority-rule heuristics	Active-schedule generation, typical rules: SPT, EDD	Rule selection
Multi-point Local Search (Hillclimber)	Pool of solutions with neighborhood moves	Pool size, move type
Multi-Objective Evolutionary Algorithms (MOEA)	Elitist archive, permutation-based crossover/mutation	Population size, p_crossover, p_mutation, archive_size
MOSA (Multi-Objective Simulated Annealing)	Acceptance by dominance or simulated annealing criterion	T_init, T_final, α_cool, moves_per_T
PILS (Pareto Iterated Local Search)	Intensification/diversification via neighborhoods/perturbations	Neighborhood types, termination criterion
AIM (Aspiration Interactive Method)	Interactive selection/post-processing	Not algorithmic, but for selection

The PILS approach implements variable neighborhood search with perturbation, using exchange, forward-shift, and backward-shift neighborhoods and a simple archiving mechanism for the current Pareto approximation. Metaheuristics are plug-and-play within the MOOPPS framework, and new modules can be integrated (0809.0406, 0809.0961).

4. Pareto Front Construction and Visualization

All metaheuristic methods maintain an up-to-date global archive $O_{j1}, \dots, O_{j,o_j}$8 of nondominated solutions. Solutions are updated upon candidate evaluation:

• If the candidate is dominated, it is rejected.
• Otherwise, the archive removes any dominated solutions and appends the new one.

The GUI displays:

• Scatter Plots: Each point $O_{j1}, \dots, O_{j,o_j}$9 with coordinates $M = \{M_1, \dots, M_m\}$0.
• Zooming, panning, and selection within objective space.
• Gantt Chart Visualization: For each schedule, jobs and operations are displayed on a time axis, with machine rows and job allocations over time.

LaTeX code snippets for objective-space scatter plots and Gantt charts are provided in the user documentation for reproducible research (0809.0961).

5. Interactive Decision Support: AIM Module

After optimization, the Aspiration Interactive Method (AIM) facilitates decision-maker selection:

• Aspiration Levels: The user defines aspiration levels $M = \{M_1, \dots, M_m\}$1 for each objective.
• Interactive Filtering: $M = \{M_1, \dots, M_m\}$2 is highlighted, the rest are hidden.
• Iterative Narrowing: The operator iteratively adjusts aspiration levels via GUI sliders or inputs until only a single nondominated schedule remains.
• Schedule Selection: The final selection $M = \{M_1, \dots, M_m\}$3 can be exported and further analyzed in the Gantt module.

This "cutting-plane" approach is non-compensatory and transparently visual, requiring no a priori scoring (0809.0961).

6. Performance, Parameterization, and Practical Usage

Experimental results on standard flow-shop and job-shop scheduling benchmarks confirm that MOOPPS and its PILS implementation produce high-quality Pareto-approximations with a small, easily-configured parameter set (0809.0406). Typical default parameters include:

• Neighborhood count for PILS: $M = \{M_1, \dots, M_m\}$4
• Perturbation block size: $M = \{M_1, \dots, M_m\}$5
• Objective function evaluations: e.g., $M = \{M_1, \dots, M_m\}$6 ($M = \{M_1, \dots, M_m\}$7), up to $M = \{M_1, \dots, M_m\}$8 ($M = \{M_1, \dots, M_m\}$9).

Quality metrics such as average and worst-case additive $p_{jk} \geq 0$0-deviations (D1, D2) quantify approximation quality relative to true Pareto fronts. On a $p_{jk} \geq 0$1 benchmark, PILS achieves D1=0.0227 and D2=0.0897 versus 0.3171 and 0.4658 for a multi-objective simulated annealing baseline (0809.0406). Average run-time scales exponentially with problem size, with ~800,000 evaluations for $p_{jk} \geq 0$2.

7. Extensions and Customization

The system architecture allows adaptation to novel scheduling settings:

• Representations and objective definitions are customizable.
• New neighborhood or perturbation operators can be incorporated for domain-specific challenges.
• Extensions such as hybridization with permutation-based genetic crossovers or large-step Markov perturbations are proposed.
• AIM supports dynamic preference articulation and interactive re-optimization, enhancing post-analysis flexibility (0809.0406, 0809.0961).

MOOPPS remains a reference platform for research and applied studies in multi-objective scheduling, providing both technical rigor and interactive explorability.