| Summary: | 碩士 === 國立臺灣大學 === 電機工程學研究所 === 98 === Evolutionary computations (ECs) have been applied to many real-world problems. One of the challenges of real-world problems is the constrained search space resulted from limited resources, requirement, et al., and another challenge is that some problems are involving human preference. Typically, solutions to conquer the first challenge in ECs rely on constraint-handling techniques such as decoder, and the second challenge is usually conquered with manually constructed objective functions by researchers. However, it is too strong such an assumption that the constructed objective function is close to the psychological preference in the human mind. This thesis presents a psychological preference-based optimization framework (PPOF) to solve the problems which are constrained and where the objective functions is involving human preference. PPOF combines interactive ECs and constraint-handling techniques in ECs. PPOF consists of three components: guidable fast search, surrogate fitness synthesizer, and EC method. This thesis discussed the characteristics of the three components and the assumptions behind them. Moreover, this thesis presents two implementations of PPOF on real-world problems: nurse scheduling problem (NSP) and space layout problem (SLP). The experiment shows that PPOF is able to arrange monthly schedules of realistic NSPs of the National Taiwan University Hospital, and PPOF also has the ability to find preferred layouts in SLP.
|
|---|
