A Unified Hybrid Evolutionary Multi-Objective Optimization Algorithm for Decision Making
Date
2017-12-22
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Multi-objective optimization involves the simultaneous optimization of two or more objectives. The objectives to be optimized are often conflicting to each other, whereas there is no single optimal solution in the search space that are superior to other solutions when all objectives are considered. Therefore, a set of trade-of optimal solutions, also known as Pareto-optimal solutions, are required to give decision makers an informed decision-making process within the acceptable time frame. Evolutionary algorithms, also known as genetic algorithms, are well suited to address the multiplicity of objectives in solutions in its search procedure and therefore used for solving multi-objective optimization problems. There have been a number of research conducted on using genetic algorithms to solve multi-objective optimization problems in the past decades, and many variations of multi-objective genetic algorithms in literature. However, the application of multi-objective genetic algorithm to solve real world problems is not documented much or cited due to the challenges and complexities presented in real world situations. Furthermore. real-world situations often require the Pareto-set to be obtained in a timely and efficient manner for decision makers, and subject expertise is required to be incorporated in the initialization and search process interactively. Furthermore, the visualization of Pareto-optimal sets is an important aspect in the decision-making process for decision makers to use and understand the impacts of choosing the solutions from the Pareto-optimal sets.
This research presents an innovative unified hybrid framework with a novel multi-objective genetic algorithm with an integrated expert module that can be applied directly in helping decision makers to solve real world multi-objective optimization problems. The validity and effectiveness of the proposed algorithm are verified by conducting experiments with three well cited benchmark data sets and comparing with previous studies in the literature. The experiments conducted with benchmark datasets proved that it can find a much better spread of solutions and better convergence near the true Pareto-optimal front compared to other research on Pareto evolutionary algorithm.
The implementation of the proposed framework uses parallel and asynchronous design to achieve high performance computing and faster convergence of the algorithm. Sets of unique parallel, asynchronous parallel genetic and K-mean operators are used to reach a global optimality through population diversity.
An expert module is included in the framework to integrate domain subject matter experts’ knowledge, experiences and preferences. The integration makes more it realistic and practical to apply this framework to solve real-world problems. A graphical reporting submodule for data visualization on the generated Pareto-optimal set is included in the expert module to visualize the results and outcomes for decision makers. The proposed framework was applied to solve two real-world multi-objective optimization problems with expected Pareto-optimal solution sets.
Description
Keywords
Multi-Objective Optimization, Genetic Programming, Decision Making
Citation
Peng, P. X. (2017) A Unified Hybrid Evolutionary Multi-Objective Optimization Algorithm for Decision Making (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.