A Survey on Preference-Based Multi-Objective Evolutionary Algorithms

Multi-objective optimization problems (MOPs) are ubiquitous in scientific research and engineering applications, where multiple conflicting objectives must be optimized simultaneously. Unlike traditional multi-objective evolutionary algorithms, which aim to approximate the entire Pareto front, preference-based multi-objective evolutionary algorithms (PBMOEAs) incorporate decision maker preferences to guide the search toward a region of interest (ROI). This paper presents a focused survey of PBMOEAs using a three-dimensional analytical framework consisting of preference articulation mode, preference modeling approach, and preference incorporation mechanism. Under this framework, the survey reviews major preference modeling approaches and preference incorporation mechanisms, and further discusses three practical issues critical to PBMOEA design: the effectiveness of preference incorporation, the practical limits of interactive methods, and the role of visualization in ROI-oriented decision support. Findings indicate that while PBMOEAs enhance decision-making efficiency, their performance hinges on preference quality, problem characteristics, cognitive load, and method suitability. Future directions include unified modeling, comprehensive evaluation, human-centered design, and integrating surrogate models, explainable learning, and LLM-assisted interfaces.