Abstract
We have developed an approach to acquire complicated user optimization criteria and use them to guide iterative solution improvement. The effectiveness of the approach was tested on job shop scheduling problems. The ill-structuredness of the domain and the desired optimization objectives in real-life problems, such as factory scheduling, makes the problems difficult to formalize and costly to solve. Current optimization technology requires explicit global optimization criteria in order to control its search for the optimal solution. But often, a user's optimization preferences are state-dependent and cannot be expressed in terms of a single global optimization criterion. In our approach, the optimization preferences are represented implicitly and extensionally in a case base. Experimental results in job shop scheduling problems support the hypotheses that our approach (1) is capable of capturing diverse user optimization preferences and re-using them to guide solution quality improvement, (2) is robust in the sense that it improves solution quality independent of the method of initial solution generation, and (3) produces high quality solutions, which are comparable with solutions generated by traditional iterative optimization techniques, such as simulated annealing, at much lower computational cost.
Introduction
In ill-structured domains like factory job shop scheduling, finding an "optimal" solution is notoriously difficult. Traditional optimization methods require an explicit global cost function, such as minimizing total delay. However, real-world experts often make complex, state-dependent trade-offs that are difficult to formalize. For instance, a manager might accept a slight delay in one order to significantly reduce work-in-process (WIP) inventory, provided machine utilization remains high. Capturing these nuanced preferences via traditional rule-based systems is labor-intensive and brittle. This paper introduces CABINS, a system designed to acquire and reuse subjective user preferences through case-based reasoning (CBR).
Methodology: The CABINS Approach
CABINS (Case-based Acquisition of user Preferences for Solution Improvement) bridges the gap between rigid algorithms and human expertise. It consists of three modules: an initial schedule builder, an interactive repair module for knowledge acquisition, and an automated repair module for reuse.
The core innovation lies in its case representation. Rather than a mathematical formula, preferences are stored as discrete cases containing the context of the schedule, the repair tactic applied, and the user's evaluation of the outcome.
Case Representation and Workflow
- Acquisition: A user manually repairs a sub-optimal schedule. CABINS records the state (global features like resource utilization and local features like waiting time), the tactic used (e.g., swapping or shifting activities), and the user's judgment (Acceptable/Unacceptable).
- Reuse: For new problems, CABINS identifies similar past cases to select a repair tactic. It then predicts whether the resulting state would be acceptable to the user based on historical evaluations.
Experimental Evaluation
The system was tested on 60 challenging job shop scheduling problems involving 10 jobs across 5 machines. These benchmarks featured substitutable resources and tight due dates to simulate the "ill-structured" nature of real-world manufacturing.
Key Results and Insights
CABINS demonstrated several critical advantages over traditional methods:
- Quality and Efficiency: It produced schedules comparable in quality to Simulated Annealing (SA) but with a 4-5x speedup. By leveraging experience, CABINS avoids the "blind" search characteristic of SA.
- Robustness: The system improved solution quality regardless of how the initial schedule was generated (e.g., simple Earliest Due Date rules vs. complex constraint-based methods).
- Preference Adaptation: CABINS successfully adopted specific user biases. When trained on different "user profiles"—one prioritizing only tardiness and another balancing tardiness with inventory—the system's performance mirrored those specific goals.
Conclusion
CABINS demonstrates that explicit mathematical cost functions are not strictly necessary for optimization in complex domains. By storing extensional examples in a case base, the system implicitly learns the trade-off logic of human experts. While initial knowledge acquisition requires human effort, the resulting system is efficient, robust, and capable of capturing the subjective nuances of real-world decision-making. Future work will explore how these case bases can evolve to accommodate shifting organizational priorities over time.
Summary updated