Multiobjective heuristic search

Více o knize

A large number of problems require optimizing multiple criteria, which are often non-commensurate and sometimes conflicting, complicating the optimization task. Creating a combined optimization function can be challenging, and the sensitivity of the solution space can impact the decision-making process, with trade-offs frequently being non-linear. In practice, we typically address these issues by proposing several non-dominated solutions instead of just one. This approach is particularly beneficial in multistage optimization problems, where solutions from one stage inform the next. A classic example is circuit design, where high-level synthesis, logic synthesis, and layout synthesis represent critical stages of optimization. Transferring a set of non-dominated partial solutions between stages usually leads to improved global optimization. This work introduces a novel method for multi-criteria optimization utilizing heuristic search techniques. Traditional multicriteria optimization methods depend on single criteria optimization algorithms, requiring either the optimization of one criterion at a time (with constraints on others) or a single scalar combined optimization function. In contrast, the multiobjective search approach assigns each optimization criterion to a distinct dimension of a vector-valued cost structure.