We have developed an integrated framework of iterative revision integrated with knowledge acquisition and learning for optimization in ill-structured domains, and implemented it in the CABINS system. The ill-structuredness of both the solution space and the desired objectives make many optimization problems difficult to formalize and costly to solve. In such domains, neither the system nor the human expert possess causal domain knowledge that can be used to guide solution optimization. Current optimization technology requires explicit formulation of a single global optimization criterion to control heuristic search for the optimal solution. Often, however, a user's optimization criteria are subjective, situation dependent, and cannot be expressed in terms of a single global optimization function. In CABINS, situation-dependent user's preferences that guide solution revision are captured in cases along with contextual information. During iterative revision of a solution, cases are exploited for multiple purposes, such as revision action selection, revision result evaluation and recovery from revision failures. Our approach was tested in the domain of job shop scheduling. Extensive experimentation on a benchmark suite of job shop scheduling problems has shown that CABINS
- is capable of acquiring user optimization preferences and tradeoffs,
- can improve its own competence through knowledge refinement,
- is a flexible schedule optimization methodology that produces high quality schedules in both predictive schedule generation and reactive schedule management in response to unexpected events during schedule execution.