ADNSGA-II: A Trend-Aware Evolutionary Algorithm for Dynamic Multi-Objective Optimization
提出一种趋势感知进化算法，通过跟踪帕累托非支配前沿的质心移动来引导搜索方向，解决动态多目标优化中环境变化适应慢和收敛质量下降的问题，在基准测试和微服务调度应用中表现优于现有算法。
A fundamental challenge in dynamic multi-objective optimization lies in simultaneously ensuring rapid adaptation to environmental changes and maintaining the convergence quality of the evolving solution set. Most existing dynamic multi-objective evolutionary algorithms rely on a stage-wise modeling paradigm, where dynamic problems are decomposed into a sequence of static subproblems solved independently. This structural assumption often leads to delayed adaptation and search drift when the objective functions evolve continuously. To overcome these limitations, we reformulate the static population evolution in the existing solutions as a trend-aware adaptive process and propose a novel trend-aware multi-objective optimization algorithm capable of dynamically adapting environment changes. The algorithm conducts trend-guided search based on the centroid movement of Pareto non-dominated front to ensure that the search process for each objective is consistent with the evolving direction of the optimization landscape. It estimates a global trend vector <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">T<sub>g</sub></i> based on the centroid shift of the non-dominated front across generations, and integrates a synergy factor to evaluate the directional consistency of each solution. This factor dynamically adjusts selection priorities, enabling the algorithm to adapt more effectively to continuous environmental changes. We conducted extensive experiments for both benchmark problems and a real-world application of microservice scheduling. The experimental results demonstrate the performance improvements of our proposed method over the popular baseline algorithms, and its cosistent achievement of globally or locally optimal values in key indicators for solving both benchmark and real-world application problems.