Minimizing Expected Deviation in Upper Level Outcomes Due to Lower Level Decision Making in Hierarchical Multiobjective Problems
Kalyanmoy Deb, Zhichao Lu, Ian Kropp, J. Sebastian Hernandez‐Suarez, Rayan Hussein, Stephen R. Miller, Amir Pouyan Nejadhashemi · IEEE Transactions on Evolutionary Computation · 2022
Many societal and industrial problem-solving tasks involving search, optimization, design, and management are conveniently decomposed into hierarchical subproblems. While this process allows a systematic procedure to have a multistakeholder solution, the independent decision-making process for the lower level problem causes a deviation in the expected outcome of the upper level problem. In this article, we provide a new and computationally efficient evolutionary approach allowing upper level decision makers to analyze the vagaries of lower level decision making when choosing a preferred solution with the minimum deviation from their expectations. This concept is novel and pragmatic. We demonstrate the concept through a search for optimistic–pessimistic tradeoff solutions found by an evolutionary multiobjective optimization approach first on two difficult test problems, then on a watershed management problem and a telecommunication management problem. The approach is generic and can be applied to similar hierarchical management problems to achieve minimum deviation with a more predictive and reliable outcome. The proposed solution procedure is found to choose an optimistic solution that has approximately 31%–65% reduced deviation compared to another optimistic solution chosen at random in the test problems and approximately 85%–95% reduced deviation in the two practical problems, making the method of this study applicable to practical hierarchical problems.